Introduction to linguistics

Lecture Notes: Linguistics
Edward Stabler, Winter 2012
http://upload.wikimedia.org/wikipedia/en/f/f7/Human Language Families Map.PNG

Lecture notes: (page numbers will change during the quarter)
1 The nature of human languages 1
2 Morphology 11
3 Syntactic constituents and categories 23
4 The anatomy of a phrase 33
5 Syntactic heads and rules 39
6 Other syntactic dependencies: movement 53
7 Clauses, tense, and questions 61
8 Review: The perspective so far 71
9 Semantics: What it all means 81
10 How the semantics ﬁts the syntax 89
11 Phonetics 99
12 Phonology introduced 111
13 Phonemes and rules of variation 119
14 Syllables and stress 129
15 Review 139
i

Stabler - Linguistics 20, Winter 2012
ii

Linguistics 20
Introduction to Linguistics
Lecture MW2-4 in Royce 190 Prof. Ed Stabler
Office Hours: M4-5, by appt, or stop by Office: Campbell 3103f
stabler@ucla.edu
Prerequisites: none
Contents: What are human languages, such that they can be acquired and used as they
are? This class surveys some of the most important and recent approaches to this question,
breaking the problem up along traditional lines. In spoken languages, what are the basic
speech sounds? How are these sounds articulated and combined? What are the basic units of
meaning? How are the basic units of meaning combined into complex phrases? How are these
complexes interpreted? These questions are surprisingly hard! This introductory survey can
only briefly touch on each one.
Text: Linguistics: An introduction to linguistic theory. V. Fromkin (ed.) Blackwell, 2000
Notes and homework will be posted at http://wintermute.linguistics.ucla.edu/Ling20/.
Requirements and grades: There will be 4 homework assignments, assigned on Wednesdays
and due the following Monday in lecture, at the beginning of class. No late assignments will be
accepted. The homework will be graded by the TAs and discussed in the discussion sections.
There will be 3 mid-term quizzes during the quarter, and an in-class final exam. The exams
will be analytic problems very similar to those given in the homework.
4 homeworks 36% (9% each)
3 quizzes + final 64% (16% each)
The final exam will have 4 parts, 1 part corresponding to each of the 3 earlier quizzes, and 1
part for the new material. So for each of the first 3 parts of the class, we will have 2 grades:
the original quiz and the grade on corresponding section of the final. Your grade will be the
higher of those 2. The part of the final on new material will be worth 16%. The idea is to
make the quizzes much less stressful (and a better indication of what you have learned) than
just 1 or 2 huge exams. You get a second shot at first 3 parts of the material.
Quiz and final exam dates (all held in class) are posted on the website,
http://wintermute.linguistics.ucla.edu/Ling20/,
where lecture notes, and reading assignments will also be posted each week.
iii

iv

Lecture 1 The nature of human languages
We are using a good text, but it has more than we can cover in a 10 week class! So these lecture
notes will contain the essential material, and the text can be used for more background and
more examples. I will be clear about which things you are expected to understand completely
– basically, it will be everything in these notes. This lecture is based on Chapter 1 of the text,
but if you compare you will see that we select just some parts and extend other parts.
Introduction . . . . . . . . . . . . 1
1.1 Productivity, Zipf’s law . 2
1.3 Creativity . . . . . . . . . . . . . . 4
1.4 Flexibility . . . . . . . . . . . . . . 5
1.5 Words and rules . . . . . . . . 5
1.6 Summary . . . . . . . . . . . . . . . 8
Human language is the most familiar of subjects, but most people do not devote much
time to thinking about it. The basic fact we start with is this: I can make some gestures that
you can perceive (the marks on this page, or the sounds at the front of the classroom), and
almost instantaneously you come to have an idea about what I meant. Not only that, your
idea about what I meant is usually similar to the idea of the student sitting next to you. Our
basic question is: How is that possible?? And: How can a child learn to do this?
The account of how sounds relate to meaning separate parts (which you may have seen
already in the syllabus), reasons that will not be perfectly clear until the end of the class:
1. phonetics - in spoken language, what are the basic speech sounds?
2. phonology - how are the speech sounds represented and combined?
3. morphology - what are words? are they the basic units of phrases and of meaning?
4. syntax - how are phrases built from those basic units?
5. semantics - how is the meaning linguistic units and complexes determined?
A grammar is a speaker’s knowledge of all of these 5 kinds of properties of language. The
grammar we are talking about here is not rules about how one should speak (that’s sometimes
called “prescriptive grammar”). Rather, the grammar we are interested in here is what the
speaker knows that makes it possible to speak at all, to speak so as to be understood, and to
understand what is said by others. The ideas are relatively new: that an appropriate grammar
will be a component of psychological and neurophysiological accounts of linguistic abilities, and
that it should be divided into something like the 5 components above. But our descriptive
goals are similar to ancient ones.
P¯an.ini (600-500 BCE) wrote a grammar
of 3,959 rules for Sanskrit, and is said
to have been killed by a lion. He is the
only linguist I know of who has been on
a postage stamp.
Considering the ﬁrst two components of grammar listed above, note the indication that
we will focus on spoken languages, and particularly on American English since that is the
language we all share in this class. But of course not all human languages are spoken. There
are many sign languages. American Sign Language (ASL) is the most common in the US, but
at the beginning of 2012, the online Ethnologue1
lists 130 sign languages. The phonetics and
phonology of sign languages involves manual and facial gestures rather than oral, vocalic ones.
The “speech” in these languages is not oral. As we will see, the patterns of morphology, syntax,
and semantics in human languages are remarkably independent of the physical properties of
1http://www.ethnologue.com/show family.asp?subid=23-16
1

speech, in all human languages, and so it is no surprise that the morphology, syntax, and
semantics of sign languages exhibit the same kinds of structure found in spoken languages.
In each of the 5 pieces of grammar mentioned above, there is an emphasis on the basic
units (the basic sounds, basic units of phrases, basic units of meaning), and how they are
assembled.
I like to begin thinking about the project of linguistics by reflecting on why the problems
should be tackled in this way, starting with “basic units.” There is an argument for that
strategy, which I’ll describe now.
1.1 Productivity, and Zipf’s law
Productivity: Every human language has an unlimited number of sentences.
Using our commonsense notion of sentence (which will be refined with various technical con-
cepts later), we can extend any sentence you choose to a new, longer one. In fact, the number
of sentences is unlimited even if we restrict our attention to “sensible” sentences, sentences that
any competent speaker of the language could understand (barring memory lapses, untimely
deaths, etc.).
This argument is right, but there is a stronger point that we can make. Even if we restrict
our attention to sentences of reasonable length, say to sentences with less than 50 words or
so, there are a huge number of sentences. The text says on page 8 that the average person
knows from 45,000 to 60,000 words. (I don’t think this figure is to be trusted! For one thing,
the text has not told us yet what a word is!) But suppose that you know 50,000 words. Then
the number of different sequences of those words is very large. Of course, many of those word
sequences are not sentences, but quite a few of them are, so most sentences are going to be
very rare. Empirical sampling confirms that this is true. What is more surprising is that even
most words are very rare.
To see this, let’s take a bunch of newspaper articles (from the ‘Penn Treebank’) – about
10 megabytes of text from the Wall Street Journal – about 1 million words. As we do in a
standard dictionary, let’s count am and is as the same word, and dog and dogs as the same
word, and let’s take out all the proper names and numbers. Then the number of differentIf word w is the n’th most frequent word,
then we say it’s ‘rank’ r(w) = n. Zipf
proposed that in natural texts the fre-
quency of w,
freq(w) =
k
r(w)α
.
for some constants k and α.
words (sometimes called ‘word types’, as opposed to ‘word occurrences’ or ‘tokens’) in these
articles turns out to be 31,586. Of these words, 44% occur only once. If you look at sequences
of words, then an even higher proportion occur only once. For example, in these newspaper
articles 89% of the 3-word sequences occur just once. Since most sentences in our average day
have more than 3 words, it is safe to conclude that most of the sentences you hear, you will
only ever hear once in your life.
The fact that most words are rare, but the most frequent words are very frequent, is often
called Zipf’s law. For example, with those newspaper articles again, plotting the actual
5000 10000 15000 20000 25000 30000
20
40
60
80
100
Fig. 1.1: Word frequency vs rank
frequencies of the most frequent word to the least frequent word gives us the graph shown
in Figure 1.1. The top of the curve gets chopped off so that I can fit it on the page! Here,
word 1 on the x-axis is the most frequent word, the, which occurs 64628 times – off the top
of the graph. Word 10 is say, which only occurs 11049 times – still off the top of the graph.
Word 2500 is probe, which occurs only 35 times and so it is on the displayed part of the curve.
Words 17,606 to 31,586 are all tied, occurring only once – these are words like zigzag, zealot,
yearn, wriggling, trifle, traumatize,. . . You have heard all these words, and more than once,
but that’s because you’ve heard many more than a million words. The surprising thing is that
2

as you increase the sample of texts, Zipf’s law stays the same: new unique words appear all
the time. Zipf’s law says that the frequencies in this plot drop off exponentially. This is the
reason that most words are rare. Given Zipf’s law about word frequencies, it is no surprise
that
most sentences you hear, you only hear once.
1.2 Compositionality
How can people understand so many sentences, when most of them are so rare that they will
only be heard once if they are heard at all? Our understanding of exactly how this could work
took a great leap early in this century when mathematicians noticed that our ability to do
this is analogous to the simpler mathematical task of putting small numbers or sets together
to get larger ones:
It is astonishing what language can do. With a few syllables it can express an
incalculable number of thoughts, so that even a thought grasped by a terrestrial
being for the very first time can be put into a form of words which will be
understood by someone to whom the thought is entirely new. This would be
impossible, were we not able to distinguish parts in the thought corresponding to
the parts of a sentence, so that the structure of the sentence serves as an image of
the structure of the thought. (Frege 1923)
The basic insight here is that the meanings of the limitless number of sentences of a productive
language can be finitely specified, if the meanings of longer sentences are composed in regular
ways from the meanings of their parts. We call this:
Semantic Compositionality: New sentences are understood by recognizing the meanings
of their basic parts and how they are combined.
This is where the emphasis on basic units comes from: we are assuming that the reason you
understand a sentence is not usually that you have heard it and figured it out before. Rather,
you understand the sentence because you know the meanings of some basic parts, and you
understand the significance of combining those parts in various ways.
We analyze a language as having some relatively small number of basic units, together with
some relatively few number of ways for putting these units together. This system of parts and
modes of combinations is called the grammar of the language. With a grammar, finite beings
like humans can handle a language that is essentially unlimited, producing any number of
new sentences that will be comprehensible to others who have a relevantly similar grammar.
We accordingly regard the grammar as a cognitive structure. It is the system you use to
“decode” the language.
In fact, human languages seem to require compositional analysis at a number of levels:
speech sounds are composed from basic articulatory features; morphemes from sounds; words
from morphemes; phrases from words. We will see all this later. The semantic compositionality
is perhaps the most intriguing, though. It is no surprise that it captured the imaginations of
philosophers of the early 1900’s (especially Gottlob Frege, Bertrand Russell, Ludwig Wittgen-
stein). In effect, a sentence is regarded as an abstract kind of picture of reality, with the
parts of the sentence meaning, or referring to, parts of the world. We communicate by passing
these pictures among ourselves. This perspective was briefly rejected by radically behaviorist
3

approaches to language in the 1950’s, but it is back again in a more sophisticated form – more
on this when we get to our study of meaning, of “semantics.”
From Barwise and Perry (1983). Here the words mirror the visual scene, but how often does that happen?
1.3 Another fundamental: “creativity”
Meaningful productivity is explained by compositionality, and compositionality brings
with it the emphasis on basic units and how they are combined. These notions should not be
confused with another idea that is often mentioned in linguistic texts, and in this quote from
the well-known linguist Noam Chomsky:
[The “creative aspect of language” is] the distinctively human ability to express
new thoughts and to understand entirely new expressions of thought, within the
framework of an “instituted” language, a language that is a cultural product
subject to laws and principles partially unique to it and partially reﬂections of
general properties of the mind. (Chomsky 1968)
Chomsky carefully explains that when he refers to the distinctive “creativity” of human
language use, he is not referring to productivity or compositionality. He says that although
linguists can proﬁtably study (productive, compositional) cognitive structures like those found
in language, our creative use of language is something that we know no more about than did
the Cartesian philosophers of the 1600’s:
When we ask how humans make use of ... cognitive structures, how and why they
make choices and behave as they do, although there is much that we can say as
human beings with intuition and insight, there is little, I believe, that we can say
as scientists. What I have called elsewhere “the creative aspect of language use”
remains as much a mystery to us as it was to the Cartesians who discussed it....
(Chomsky 1975, p.138)
Here the point is that we humans are “creative” in the way we decide what to say and do.
Chomsky suggests that we produce sentences that are in some sense appropriate to the context,
4

but not determined by context. Our behavior is not under “stimulus control” in this sense.
What we say depends in complex ways on our internal, psychological states which represent
the context in various ways, still poorly understood.
Regardless of whether we accept Chomsky’s scepticism about accounting for why we say
what we do when we do, he is right that this is not what most linguists are trying to account for.
This is an important point. What most linguists are trying to account for is the productivity
and compositionality of human languages. The main question is: What are the grammars of
human languages, such that they can be acquired and used as they are?
1.4 One more fundamental: flexibility and change
One thing that the first quote from Chomsky suggests is that language has a certain flexibility.
New names become popular, new terms get coined, new idioms become widely known – the
conventional aspects of each language are constantly changing. We are inventing the language
all the time, extending it in ways that are not predicted simply by the possibility of new
compositions from familiar elements (productivity and compositionality). Linguists have been
especially interested in what remains constant through these changes, the limitations on the
flexibility of human languages. It is easy to see that there are some significant limitations,
but saying exactly what they are, in the most general and accurate way, is a challenge. We
can adopt a new idiom naturally enough, at least among a group of friends, but it would not
be natural to adopt the convention that only sentences with a prime number of words would
get spoken. This is true enough, but not the most revealing claim about the range of possible
human languages. You can name your new dog almost anything you want, but could you give
it a name like -ry, where this must be part of another word, like the plural marker -s (as in
dogs), or the adverbial marker -ly (as in quickly)? Then instead of Fido eats tennis balls would
you say eatsry tennis balls or dory eat tennis balls or eats tennisry balls or what? None of these
are natural extensions of English. What kinds of extensions really get made and adopted by
others? This is partly a question of language learning, and partly a sociological question about
how groups come to adopt a new way of speaking.
1.5 Words and rules
If you have never studied language, the common ideas is that the atoms of a language are
its words. To learn a language, you learn the words and the rules for putting them together.
Let’s begin with this idea. The text says (p.26) that the following sentence from Shakespeare
has 13 words (12 different words, since 1=11):
The friends promised to inquire carefully about a schoolmaster for the fair Bianca
1 2 3 4 5 6 7 8 9 10 11 12 13
All of these words are meaningful, and some of them are complex in the sense that their
meanings are composed from meanings of their parts:
(1) a compound is a word that has other words as parts,
like school-master or looking glass
(2) a word can be composed of a root (or “base”) together with 0 or more affixes
like friend-s, promis-ed, care-ful-ly
(an affix is a prefix or suffix – and some other possibilities are mentioned later)
5

Why is -ed a suffix of promised, but pr- is not a prefix? The answer proposed in the text is:
(3) A morpheme is the smallest meaningful unit in a language, and
(4) Roots and affixes are morphemes.
So -ed is an affix because it means Past, but pr is not a morpheme of any kind because it is
not meaningful. So the morphemes of the first sentence are these:
The friend -s promis -ed to inquire care -ful -ly about a school -master for the fair Bianca
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
But we cannot understand definitions (1) or (2) without saying what a word is.
So, what is a word? The text says (p25) “Words are meaningful linguistic units that can be
combined to form phrases and sentences.” But all morphemes are by definition meaningful, and
they can be combined to form phrases – but we are told (p.26) that words are not the smallest
units of meaning. So what is a word? The text glossary says that words are meaningful units
that can be ‘freely reordered’ (p.320). Prefixes and suffixes are morphemes that cannot ‘occur
freely’ in this sense – they are sometimes said to be ‘bound’. Words can ‘stand on their own’.
Of course, phrases and sentences like the one dissected into 17 different morphemes above, are
meaningful units that can stand on their own too. We see that when roots and affixes are
combined, as in care-ful-ly or promis-ed, the result is a word that contains many morphemes.
Two words can combine as in school-master to form a new word that contains two words that
are each morphemes. This is a ‘compound’ word. But when Shakespeare combines the 16
words to form our example sentence about Bianca, although this is a meaningful unit that can
occur freely, it is not a word. So let’s tentatively adopt this definition:
(5) A word is a morpheme, or a complex of roots and affixes, or a compound, that can
occur freely.
To understand this definition exactly, we would need to explain more carefully what it means
“to occur freely.” This interesting idea is not easy to pin down. What is special about
the breaks between words, which we do not find between roots and affixes (and perhaps not
between the words in a compound)? There are various ideas about this, but let’s rely on our
intuitions for the moment.
1.5.1 Roots and affixes
We find another interesting connection between morphemes and pronunciation when we look
more carefully. For example, notice that our first example contained the past tense morpheme
that is usually spelled “-ed”, pronounced [t]:
@ fôEnd -z "pôamIs -t tu In"kwaIô keô -f@l -li. . .
The friend -s promis -ed to inquire care -ful -ly. . .
1 2 3 4 5 6 7 8 9 10. . .
Looking at a wider range of examples, we see that there are other past tense forms. Pro-
nouncing the following examples, you will hear 3 different pronunciations. (We use a phonetic
alphabet to indicate speech sounds, as dictionaries do. This notation will be introduced later.
– For now, it is enough to hear the different pronounciations of the past tense.)
spelling sound spelling sound spelling sound
promis-ed pôamIs-t penn-ed pEn-d load-ed lod-@d
dropp-ed dôap-t hugg-ed h2g-d nett-ed nER-@d
kick-ed kIk-t snar-ed sner-d fat-ed feR-@d
6

The way the past tense is pronounced is not random! We will consider this more carefully
later. These variant pronunciations of the morpheme are sometimes called allomorphs – that
just means: alternative form of the morpheme.
1.5.2 Are irregular forms complex?
Many English verbs have irregular past tense forms that are not root+suffix combinations,
some of them falling into groups like this:
i. blow, grow, know, throw, draw, withdraw, fly
ii. take, mistake, shake, forsake
iii. bind, find, grind, wind
iv. bend, send, spend, lend, build
v. swear, tear, wear, bear, get, forget
vi. keep, sleep, sweep, weep, creep, leap, feel, deal, mean, dream, leave, lose
vii. ring, sing, spring, drink, shrink, run, swim, begin The text says (p.67) that we need to
“memorize irregularly related pairs like
ran and run,” but these sub-regularities
(i-xii) suggest that maybe there is a rule
for a collection of verbs that includes run.
What would that rule say?
viii. sink, stink, spin, win
ix. string, swing, sting, cling, sling, wring, fling, stick, dig
x. bleed, read, breed, feed, meet, lead, mislead, speed
xi. teach, catch, buy, bring, seek, think
In many of these cases, forming past tense involves not adding a suffix, but changing the vowel.
Should the vowel change be regarded as a kind of morpheme? When we hear a verb like grew,
do we decompose it into grow+past where the past is pronounced not as a suffix, but as a
vowel change? And if we say yes, should we say the same thing about the past tenses of even
more irregular verbs like the verb be? Why do the verbs in (xii), verbs that do
not change in the past, all end in [t] or
[d]? Do any verbs ending in [t] or [d]
have regular past tenses?
And should we say the same about verbs whose irregularity is that nothing gets added to
form the past?
xii. hit, slit, split, quit, knit, bid, rid, shed, spread, wed, let, set, upset, beset, wet, cut, shut,
put, burst, cast, cost, thrust, hurt
There are various kinds of evidence bearing on these questions.
First, notice that the irregular past tense forms, like the regular ones, never allow an
additional affix:
(6) * He was promis-ed-ing everything
(7) * He was knew-ing everything
(8) * He was found-ing everything
(9) * He was taught-ing everything
This is true with irregular plurals too:
(10) a. * He looks mice-y
b. * He looks mouse-y
7

Second, notice that forming a yes-no question seems to involve splitting the tense away from
the verb, and this happens for both regular and irregular forms:
(11) a. You promis-ed Bianca
b. Di-d you promise Bianca?
(12) a. You knew Bianca
b. Di-d you know Bianca?
(13) a. You found Bianca
b. Di-d you find Bianca?
(14) a. You taught Bianca
b. Di-d you teach Bianca?
We see a similar thing when we use do emphatically, as in these cases:
(15) a. I never promised Bianca, but you di-d promise her
b. I never knew Bianca, but you di-d know her
c. I never found Bianca, but you di-d find her
d. I never taught Bianca, but you di-d teach her
In these cases, when the past tense is expressed on do, the verb takes its usual bare form,
whether it is regular or not. This suggests that both regular and irregular forms really are
complex, with a verb morpheme and a past morpheme.
The processes of word recognition are very rapid. In ordinary fluent speech, we often
are able to recognize words before hearing the end of them. In both listening to speech and
reading, word recognition is influenced by many factors, and it is hard to disentangle effects
of finding the stem from the effects orthographic/auditory similarity and frequency. But
there is some evidence of decomposition of stem and affix even in the earliest stages of word
recognition. Monitoring magnetic fields on the scalp (with MEG, magnetoencephalography),
there is evidence of a response to lexical recognition of a stem in forms like farmer and refill
as compared with simple forms like winter and resume (Zweig and Pylkkänen 2008). Another,
slightly slower response may indicate recognition of the stem, which happens with regulars
and irregulars alike (Stockall and Marantz 2006, Morris and Stockall 2011).
1.6 Summary
The basic questions we want to answer are these: how can human languages be (1) learned
and (2) used as they are? These are psychological questions, placing linguistics squarely in
the “cognitive sciences.” (And our interest is in describing the grammar you actually have,
not in prescribing what grammar you “should” have.)
The first, basic fact we observe about human languages shows that the answer to these
questions is not likely to be simple! Our first, basic fact about the nature of all human languages
is that they are productive – No human language has a longest sentence. It follows from this
that you will never hear most sentences – after all most of them are more than a billion words
long!
Zipf’s law gives us a stronger claim, more down to earth but along the same lines. Al-
though the most frequent words are very frequent, the frequencies of other words drop off
8

exponentially. Consequently, many words are only heard once, and it is a short step from
there to noticing that certainly most sentences that you hear, you hear only once.
To make sense of how we can use a language in which most sentences are so rare, we
assume that the language is compositional, which just means that language has basic parts
and certain ways those parts can be combined. This is what a language user must know, and
this is what we call the grammar of the language. This is what linguistics should provide an
account of.
It turns out that compositional analysis is used in various parts of linguistic theory:
1. phonetics - in spoken language, what are the basic speech sounds?
2. phonology - how are the speech sounds represented and combined?
3. morphology - what are the basic units of meaning, and of phrases?
4. syntax - how are phrases built from those basic units?
5. semantics - how can you figure out what each phrase means?
You will understand something about each of these by the end of the class.
We begin with the most familiar topic: morphology, the theory of words and their parts.
We introduced words and morphemes, roots and affixes (suffixes, prefixes, infixes). Words
with parts are called “complex.” English has some complex words, traditionally separated
into “regular” and “irregular” examples. Even the irregular forms seem to be complex in the
sense that the speaker recognizes the relation between related forms (e.g. the forms in different
tenses or different ‘conjugations’). Obviously, these complexes can only go together in certain
ways – more on this next time.
Questions:
Questions about lecture material will be addressed in discussion sections, but you can also
stop by my office M4-5 or anytime. Short questions can also be emailed to me.
To: stabler@ucla.edu
Subject: question
In today’s lecture on Zipf’s law, when you plotted the graph,
what did the x and y axis stand for?
On the x-axis, 1 represents the most frequent word, the, 2 represents the second most frequent
word, be, word 3 is a, word 4 is of, word 5 is to, word 6 is in, word 7 is and, word 8 is for, word
9 is have, word 10 is say, and so on. On the y-axis, I plotted how frequent each word was.
Instead of writing the words on the x-axis, I just put the numbers 1, 2 ,3,. . . , partly because
writing all those words there is hard work, and partly because what I wanted to show was just
the shape of the curve. The shape of the curve by itself shows that the most frequent words
are very frequent, and the other words are rather rare!
9

References and further reading
[1] Barwise, Jon and John Perry. (1983) Situations and Attitudes. Cambridge, Massachusetts: MIT
Press.
[2] Chomsky, Noam (1968) Language and Mind. NY: Harcourt Brace Javonovich.
[3] Chomsky, Noam (1975) Reflections on Language. NY: Pantheon.
[4] Frege, Gottlob (1923) Compound Thoughts. Translated and reprinted in Klemke, ed., 1968,
Essays on Frege. University of Illinois Press.
[5] Huang, Yi Ting and Steven Pinker (2010) Lexical semantics and irregular inflection. Language
and Cognitive Processes: 25(10): 1411-1461.
[6] Morris, Joanna and Linnaea Stockall (2011) Early, equivalent ERP masked priming effects for
regular and irregular morphology. Brain and Language, forthcoming.
[7] Stockall, Linnaea and Alec Marantz (2006) A single route, full decomposition model of morpho-
logical complexity. The Mental Lexicon 1(1): 85-123.
[8] Turing, Alan (1936) On computable numbers with an application to the ensheidungs problem.
Proceedings of the London Mathematical Society 42(2): 230-265, 544-546.
[9] Zweig, Eytan and Liina Pylkkänen (2008) A visual M170 effect of morphological complexity.
Language and Cognitive Processes: 24(3): 412-439.
[10] Zipf, George K. (1949) Human Behavior and the Principle of Least Effort: An Introduction to
Human Ecology. Houghton-Mifflin, Boston.
10

Lecture 2 Morphology
In the language of decimal numbers, each digit has a meaning on its own, and we calculate
the meaning of any sequence of digits accordingly. Human language apparently does not work
like that. Many sequences of morphemes cannot go together at all. Let’s consider how this
might work.
2.1 Assembling words, morphemes, roots, and affixes
Last time we considered the idea that the “atoms” of morphology are morphemes, the smallest
meaningful units. Since the study of meaning is called semantics, this is a picture according
to which morphemes are semantic atoms. But when we try to describe how they can be
assembled, it turns out that we need to refer not just to their meanings, but also to their roles
in building phrases.
(1) The regular past tense affix -ed attaches only to verbs.
(2) The regular plural affix -s attaches only to nouns.
(3) The adverb-forming affix -ly attaches only to adjectives and nouns.
Morphemes have particular “parts of speech:” they are nouns, verbs, adjectives, etc. These
notions refer to the function of the morpheme in assembling a phrase, and the study of phrases
is syntax, so these are syntactic categories. Let’s abbreviate the “parts of speech,” these
syntactic categories in this way:
part of speech: noun = N verb = V adjective = A adverb = Adv determiner = D preposition = P
example, spelling: friend say happy quickly the with
pronunciation: [fôEnd] [se] [hæpi] ["kwIkli] [ @] [wI ]
As indicated in the text (by the “frames” on pp31-32), each part of speech plays a certain kind
of role in building up a phrase, and can only occur in certain positions. We will get to syntax
soon, and have much more to say about parts of speech then. But just using what you may
know from using the dictionary, we can label many of the morphemes in our first sentence:
D N ? V ? P V N ? ? P D N N P D A N
It is not immediately obvious what parts of speech, if any, the bound morphemes have – we
will return to this later. But we might notice right away that we typically assign these familiar
parts of speech to the complex words in this sentence too. (We will return to see whether it is
right to do so in just a moment – (it is!))

N V A Adv N
friend-s promis-ed care-ful care-ful-ly school-master
This suggests that it is not just morphemes that are nouns, verbs, adjectives, and so on.
So what are these “parts of speech”? They are the basic elements of syntax. The parts of
speech formed by the morphology are syntactic atoms. These parts of speech are familiar
from dictionaries, and now we notice that having a part of speech seems, at least at first, to
go with being a semantic atom. So we can ask:
Q1 Are morphemes syntactic atoms?
(= Are the smallest units of meaning also the smallest units of phrases?)
Q2 How do morphemes relate to the units relevant to pronunciation (basic speech sounds,
syllables)?
These are good questions, but let’s postpone them! We should first see some morphology
before asking how it relates to everything else. The questions interanl to morphology are
these:
Q3 How do words get put together to make compounds?
Q4 How do roots and affixes get put together to form words?
It turns out there are some clear and interesting things that can be said about these problems.
We return to Q1 and Q2 after we see what morphology says about Q3 and Q4.
2.2 English morphology
We consider English first, and then look at some different phenomena in other languages.
2.2.1 Compounds
Compounds are words formed from other complete words. For example:
bartend, apple pie, jet black, part supplier,
boron epoxy rocket motor chamber instruction manual writer club address list
2.2.2 Roots + affixes
Some suffixes can combine quite freely
phon spelling effect examples
-ô
"
-er changes V to N kill-er
-@b@l -able changes V to A manage-able
-nEs -ness changes A to N happi-ness
Other affixes are much more fussy. We have, for example,
[N [N Reagan] -ism] [N [A modern] -ism] [N [N library] -ian] [N [A [N Darwin] -ian] -ism]
but not
*[A [N [N Darwin] -ism] -ian] *[A [N [N Reagan] -ism] -ian]
12

And we have the possessive or plural [N Reagan -s], but not
*[N [N Darwin -s] -ism] *[A [N Darwin -s] -ian] *[A [N Darwin -s] -ian] -ism
Notice how the brackets indicate how the elements combine together. The restrictions on these
combinations are not just memorized on a word-by-word basis, but apply to whole collections
of similar elements. If we invent a new verb glark, and say that Schwartzenegger is glarkable,
then we immediately know that this means that Schwartzenegger can be glarked, and we know
that we have said something about glarkability. We can describe some of these regularities as
follows:
Some suffixes can combine only with roots
Some examples, many from Fabb (1988):
-@n -i@n -an -ian changes N to N librari-an, Darwin-ian
changes N to A reptil-ian
-@dZ -age changes V to N steer-age
changes N to N orphan-age
-@l -al changes V to N betray-al
-@nt -ant changes V to N defend-ant
changes V to A defi-ant
-@ns -ance changes V to N annoy-ance
-et -ate changes N to V origin-ate
-d -ed changes N to A money-ed
-f@l -ful changes N to A peace-ful
changes V to A forget-ful
-hUd -hood changes N to N neighbor-hood
-IfaI -ify changes N to V class-ify
changes A to V intens-ify
-IS -ish changes N to A boy-ish
-Iz@m -ism changes N to N Reagan-ism
-Ist -ist changes N to N art-ist
-Iv -ive changes V to A restrict-ive
-aIz -ize changes N to V symbol-ize
-li -ly changes A to A dead-ly
-li -ly changes N to A ghost-ly
-mEnt -ment changes V to N establish-ment
-ori -ory changes V to A advis-ory
-2s -ous changes N to A spac-eous
-i -y changes A to N honest-y
-i -y changes V to N assembl-y
-i -y changes N to N robber-y
-i -y changes N to A snow-y, ic-y, wit-ty, slim-y
(You do not need to memorize the tables of affixes! It is enough to know how to read them, and
you can see roughly how many suffixes we can list with rough indications of the restrictions
on where they occur. Some languages have many more affixes than English.)
13

Some suffixes can combine with a root, or a root+affix
-eôi -ary changes N-ion to N revolut-ion-ary
-eôi -ary changes N-ion to A revolut-ion-ary, legend-ary
-ô
"
-er changes N-ion to N vacat-ion-er, prison-er
-Ik -ic changes N-ist to A modern-ist-ic, metall-ic
-(eS)2n -(at)ion changes N-ize (etc) to N symbol-iz-ation, impress-ion, realiz-ation
-(@t)oôi -(at)ory changes V-ify to A class-ifi-catory, advis-ory
Some suffixes combine with a specific range of suffixed items
-@l -al changes N to A natur-al
sometimes allows -ion, -ment, -or
-j@n -ion changes V to N rebell-ion
sometimes allows -ize, -ify, -ate
-Iti -ity changes A to N profan-ity
sometimes allows -ive, -ic, -al, -an, -ous, -able
-Iz@m -ism changes A to N modern-ism
sometimes allows -ive, -ic, -al, -an
-Ist -ist changes A to N formal-ist
-aIz -ize changes A to V special-ize
2.2.3 English morphological rules
Looking over the discussion so far, we can see some basic patterns. Let’s review some of them.
If we consider compounds first, we notice a striking pattern:
[V [N bar] [V tend]]
[N [N apple] [N pie]]
[A [N jet] [A black]]
[Npl [Nsg part] [Npl suppliers]
[Nsg [Npl parts] [Nsg supplier]
[N [N [N rocket] [N motor]] [N chamber]]
Instead of using brackets to show how these words are formed, it is often easier to use “tree
diagrams”, like this:
V
N
bar
V
tend
A
N
jet
A
black
N
N
N
rocket
N
motor
N
chamber
N
N
university
N
N
parking
N
lot
14

These trees are upside down: the point at the top is the root, along the bottom we have the
pronounced leaves. The root, the leaves, and the other labeled constituents in the tree are
called nodes of the tree. There is always just one root, and The branches never cross.
Notice that the different structures of the last two examples, what modifies what, is figured
out by considering what makes the most sense (more discussion of this in the text). Another
basic thing we see is that the roots combine in pairs. The pairs we see here can be described
with rules like the following (this rule format is not presented in the text):
V → N V
N → N N
A → N A
There is another regularity here. All of these rules have the form
X → Y X
This regularity in English compounds is described as follows:
(4) In English, the rightmost element of a compound is the head.
(5) A compound word has the category and features of its head.
This is called the English right hand head rule or the head-final principle (p68 in the
text).1
There is an analogous way to write affixation rules. The important thing to notice is that
the head-final rule in compounds predicts some of the patterns we see in affixation:
(6) an English suffix often changes category, but prefixes rarely do
(7) the conditions on affixation typically refer to the just the last suffix
The conditions for attaching a suffix never refer to the root, which may seem surprising to a
non-linguist, since, intuitively, it is usually the root that provides most of the meaning of the
word.
How can we exploit this insight that affixes and compounds both seem to have their prop-
erties determined by their righthand members? Well, we can just suppose that affixation
structures are head-final too. Then, considering the most productive affixes first, we can use
rules like the following to describe their requirements and their effects:
N → -er / [V ] (manager)
A → -able / [V ] (manageable)
N → -ness / [A ] (happiness)
The first rule says that the N -er is allowed when it can form a complex with a verb. And by
the head-final rule, we know that the resulting complex will be a noun N. We read the other
rules similarly. We can draw the resulting structures with trees.
1In syntax, we will see that phrases have heads too. The head of a phrase is generally different from the
head of a word, though. As we will see, there is no head-final rule for syntactic heads in English.
15

For affixation structures, the TEXT presents trees like the following:
N
V
manage
-er
A
V
manage
-able
N
A
happy
-ness
But if we use the rules given above, then INSTEAD, we can provide categories for the affixes,
conforming to the English head-final rule:
N
V
manage
N
-er
A
V
manage
A
-able
N
A
happy
N
-ness
Prefixes in English tend not to be category changing, but rather just modifiers, and so if
we had to assign categories to them, we could observe that
1. A modifies N, as in happy guy
2. Adv modifies V, as in he completely finished
3. Adv modifies A, as in completely happy
So we could assign trees like these to prefix structures:
A
Adv
un-
A
happy
V
Adv
un-
V
tie
V
Adv
re-
V
make
N
N
A
Adv
un-
A
happy
N
-ness
N
-es
N
A
anti-
N
N
race
N
-ist
These trees conform to the same generalization that we had for compounds: the right sister
determines category.
In fact, applying the head-final rule to each of the affixes in our first example sentence, we
obtain a category for all of the suffixes:
D N N V V P V N A Adv P D N N P D A N
16

2.3 How morphology relates to other things
OK. Now let’s reflect again on the general questions Q1 and Q2 from page 12 about how
morphological elements compare to the basic elements of speech sound, of phrases (syntax),
and of meaning (semantics).
2.3.1 Morphology and pronunciation
Sequences of sounds form syllables, but most syllables do not mean anything – they are often
just parts of morphemes. But it can also happen that a morpheme is a single speech sound,
smaller than a syllable.
(8) Morphemes can be smaller than syllables: the plural, possessive, or contracted
verb -s; and the affixes -y, -ic, -ed often are less than a full syllable
The text even considers the idea that a morpheme can be completely unpronounced
(pp75-78, and this idea will come up again on pp.279,322-3).
(9) Morphemes can be multi-syllabic: many nouns, verbs and proper names are mor-
phologically simple, but have more than one syllable: apple, ridicule, Merodachbaladan
(10) Morphemes may be pronounced in different ways in different contexts: al-
lomorphs. As discussed last time, past tense -ed is pronounced in various ways, plural
-s is pronounced in different ways. See slides 1b for notes on how to rec-
ognize allomorphs (also discussed in the
text).
(11) Some morphemes may be unpronounced. We mentioned last time that the
past tense sometimes has no pronunciation when it is on the verb, though it may be
pronounced again when it is separated from the verb to attach to do:
You wed Bianca yesterday Di-d you wed Bianca yesterday?
(12) Some morphemes are not sound sequences at all: the text presents several cases
of morphemes that attach to a word to reduplicate all or part of the sounds in that
word. For example, in Pima:
gogs gogogs "uvi "u"uvi jiosh jijosh toobi totobi
dog dogs woman women god gods rabbit rabbits
2.3.2 Syntactic atoms, semantic atoms, and morphemes
We asked earlier, in Q1 (on page 12), whether semantic atoms (the smallest units of meaning)
are always also syntactic atoms (the smallest units of phrases: verbs, nouns, adjectives, etc).
While these often go together, it is possible to find cases where syntactic atoms are not semantic
atoms, and vice versa. Interestingly, some of these are cases where the syntactic atoms are
nevertheless elements that our morphological rules can (and presumably should) apply to.
The traditional view about morphemes faces some problems. First: a semantic atom
can have many syntactic atoms in it. This is shown by the existence of multi-morphemic
idioms. An idiom is a complex expression whose meaning is not determined by the meanings
of its parts in the usual way. Phrasal idioms are often discussed, but there are also idiomatic
compounds and words – these expressions have special meanings that cannot be calculated
from the meanings of their parts:
17

(13) idiomatic phrases: He threw in the towel, He kicked the bucket, His goose is cooked.
(14) idiomatic compounds: pick-pocket, scare-crow, push-over, try-out, cut-throat, pain-
s-taking, pig-head-ed, carpet-bagg-er, water-melon, sun-flower, under-stand, with-stand,
down-load, under-take, under-tak-er, over-come, for-sake, red-wood, black-bird, hot-
dog, white cap, cold cream, looking glass.
(15) idiomatic root+affix: librar-ian, material-ist, wit(t)-y, ignor-ance, mis-take, sweat-
er.
It is natural to say that, in these idiomatic uses, the parts of these expressions are syntactic
atoms, but they do not have their usual meanings (i.e. they can be meaningful in other
contexts, with different meanings). Since we have to learn what these whole expressions mean,
rather than figuring the meanings out from the parts, so these are semantic atoms.
A related thing seems to happen in examples like these, where we seem to have morpheme-
like elements that cannot occur in any other context:
(16) cran-berry, rasp-berry, huckle-berry, un-couth, in-ept, dis-gust-ed, dis-gruntl-ed,
re-ceive, ami-able
This seems to be another kind of idiom, where one component is meaningful only in a unique
or narrow range of morphological complexes.
Second: a syntactic atom can have many semantic atoms in it, or so it seems.
Since we have not studied how phrases are formed, we are not really in a position to rigorously
assess this claim. But in the case of a compound noun like sunflower seed package, it seems
that the rules of the syntax never need to see the morphemes, only the whole word. The
compound itself is an atom in the rules for building phrases. Similarly for root-affix complexes
like the verb de-institut-ion-al-ize. The syntax does not need to see its parts (or so it seems).
With these puzzles, it is no surprise that the notions of word and morpheme are given
only a rough definition in most texts. Some say simply that “morphemes can be defined as
the smallest meaningful unit,” (Haspelmath 2002, p.16; Lieber 2010, p.3). Others say that
morphemes are the syntactic atoms, whether or not they have semantic or phonological content
(Halle and Marantz, 1993: 114). Our text implies a mixed view, saying a that the morpheme
is the “smallest meaningful linguistic unit” (p.703), but also assigning each morpheme a part
of speech – suggesting that morphemes are both semantic atoms and syntactic atoms. The
examples just above show that is not quite right. Another text says it is “the smallest unit of
language which carries information about meaning or function” (O’Grady et al 2010, p.117).
But strictly speaking, an initial [b] (or any other sound) on an English word carries some
information about both meaning and function, since it narrows down the possibilities (very
roughly, to those meanings and functions in the [b] section of the dictionary). Linguists
working more carefully conclude that the traditional notions of morpheme and word mix various
considerations that do not quite go together. The basic units of meaning (semantic atoms)“The conclusion drawn by many, if
not most, morphologists is that the
morpheme concept is inappropriate ex-
cept as a crude approximation. In-
stead,. . . morphemes are taken to be
markers serving to realize some feature
specification. . . not listed lexical entries
in their own right.” (Spencer 2006,
p.110)
and the basic units of grammatical functions, of phrase structure (syntactic atoms), are not
quite the same.2
. The notions of morpheme and word do not quite fit either characterization.
It is not quite clear how to resolve these problems, but these ideas are clear enough for us to
proceed.
2See for example Spencer (2006), or the introduction to Spencer and Zwicky (1998), or Di Sciullo and
Williams (1987), or Anderson (1992, esp. pp.51-6) for discussion of the ongoing controversy surrounding these
issues.
18

2.4 Other languages
Is the situation for words and morphemes different in other languages? Well, all languages
have semantic atoms and word-like units of course, but languages vary quite a lot in how much
gets put into a word. Languages that avoid combining many grammatical markers onto a stem
are sometimes called isolating or analytic, as in this example from Fijian (Bickel and Nichols
2005):
(17) Au
1sg
aa
pst
soli-a
give-tr
a=niu
art=coconut
vei
to
ira
3pl
‘I gave the coconut to them.’
Languages that tend to combine many grammatical morphemes with a stem can do so by
concatenating affixes, or by altering the morphemes, or with tones, as in this one word example
from the South American language Quechua (Stabler 1994):
(18) wañu-chi-chi-lla-sa-nku-ña-puni.
die-make-make-del-prog-3pl-dur-emp
‘they are still just having people killed as always’
Languages of this latter sort are sometimes called fusional (or agglutinating or polysynthetic).
When we look at a map, we see that these tendencies are localized in certain areas.
This map is from Bickel and Nichols 2005.
Many examples of various sorts of languages are discussed in the text.
19

2.5 Summary
The traditional view presented in Chapter 2 of our text says that morphemes are semantic
atoms, and morphology is about how words are built from morpheme roots, morpheme affixes,
and other words. (We stick with this picture for now.)
You should know what these morphological elements are: roots, affixes, suffixes, pre-
fixes, infixes, compounds, reduplicative morphemes. You do not need to memorize the
English suffixes! but know the English head-final rule, and you should understand our
rule notation for compounds and affix structures. You should be able to write the morpho-
logical rules which would describe some new data, and draw the trees for the word
structures – trees like the ones in these notes, where the affixes have categories.
For the general picture of what’s happening, you should know that the basic units of
morphology are quite different from syllables or basic speech sounds like vowels or consonants.
We pronounce words with sounds, but the units of sound don’t match up with the units of the
words in any simple way. And you should know that many words are idioms, in a sense; just
like many phrases are (and that this presents puzzles for the traditional view that morphemes
are semantic atoms).
In this lecture, we introduced rules like: X → Y X. That means that inside an X, another X
can recur – in this sense, these rules are recursive. In building root-affix complexes, these rules
tend to be tightly restricted, and sometimes we indicate the restrictions on their applicability
with conditions after a slash:
X → Y X/conditions.
In English noun compounds, the recursion is quite free and much easier to see. With free
recursion, the number of possible words becomes infinite. And this is just the beginning.
Recursion of roughly the sort we see in compounding appears in many parts of the grammar.
[1] Anderson, Stephen R. (1992) A-Morphous Morphology. NY: Cambridge University Press.
[2] Baker, Mark (1988) Incorporation: A Theory of Grammatical Function Changing. Chicago:
University of Chicago Press.
[3] Bickel, Balthasar and Johanna Nichols (2005) Fusion of selected inflectional formatives. In M.
Haspelmath, M.S. Dryer, D. Gil and B. Comrie, eds., The World Atlas of Language Structures. NY:
Oxford University Press.
[4] Di Sciullo, Anna-Maria and Edwin Williams (1987) On the Definition of Word. Cambridge,
Massachusetts: MIT Press.
[5] Fabb, Nigel (1988) English suffixation is constrained only by selection restrictions. Natural
Language and Linguistic Theory 6: 527–539.
[6] Halle, Morris and Alec Marantz (1993) Distributed morphology and the pieces of inflection. In K.
Hale and S. J. Keyser, eds., The View from Building 20, pp. 111-176. Cambridge, Massachusetts:
MIT Press.
[7] Haspelmath, Martin (2002) Understanding Morphology. London: Hodder Education.
[8] Lieber, Rochelle (2010) Introducing Morphology. NY: Cambridge University Press.
[9] O’Grady, William, John Archibald, Mark Aronoff, and Janie Rees-Miller (2010) Contemporary
Linguistics: In Introduction, Sixth Edition. NY: St. Martin’s.
20

[10] Selkirk, Elisabeth O. (1982) The Syntax of Words. Cambridge, Massachusetts: MIT Press.
[11] Spencer, Andrew (2006) Morphological universals. In R. Mairal and J. Gil, eds., Linguistic
Universals. Cambridge: Cambridge University Press.
[12] Spencer, Andrew and Arnold M. Zwicky, eds. (1998) The Handbook of Morphology. Oxford:
Blackwell.
[13] Stabler, Edward (1994) The ﬁnite connectivity of linguistic structure. In C. Clifton, L. Frazier and
K. Rayner, eds., Perspectives on Sentence Processing. Hillsdale, New Jersey: Lawrence Erlbaum.
21

22

Lecture 3 Syntactic constituents and categories
3.1 Productivity begins in morphology
3.1.1 First: morphemes, words and parts of speech are different!
Chapter 3 briefly makes a couple of claims about morphemes which are not quite right when
the term “morpheme” is used as we are using it – as a semantic atom.
• on page 89, the text says “In this chapter, we will see how words are combined with each
other to form grammatical sentences in a. . . rule-governed way.” Given our definition of
word from Chapter 2 (and p.6 of lecture notes 4a), though, we might conclude that in the
sentence “He’s bad!” [hiz"bæd], the contracted form is a word. But now we will treat it as
2 parts of speech, the pronoun he and a form of the verb be.
• on page 98, the text says “Although the number of grammatical English sentences is infinite,
the number of English morphemes (and words) is not. . . ” We saw that with recursive rules
in morphology, the number of English words is infinite. In particular, according to the
view presented in Chapter 2 and in class, the number of English nouns infinite. It is perhaps
possible though to regard those rules as outside of morphology, restricting morphology to
root+affix structures. But even with root+affix structures, the text acknowledges that
these can be quite productive in English and many other languages. So although English
has finitely many morphemes, from which the words are built, it has infinitely many words.
• on page 99, the text says “. . . within the mental lexicon of any single speaker, there is a
pronunciation associated with each morpheme.” But we saw in Chapter 2 and in class
that some morphemes are not associated with pronounced material. There can be null
affixes, but the more interesting case are the reduplicative morphemes: rather than being
associated with any particular pronounced material, these morphemes copy all or parts of
the morphemes they are attached to.
3.1.2 productive affixation
We noticed last week that following are among the most “productive” affixes – meaning that
these elements combine with the wides range of elements:
phon spelling effect meaning examples
-ô
"
-er changes V to N one who Vs kill-er
-ô
"
-er changes A to A more A light-er
-@b@l -able changes V to A capable of being V-ed manage-able
-nEs -ness changes A to N state of being A happi-ness
23

But when we look at what -er attaches to, we find some restrictive patterns. It attaches to
many single syllable adjectives (ES judgements):
bigg-er kind-er ripe-r *intelligent-er *adequate-r *haphazard-er
tall-er nice-r strong-er *active-r *placid-er *us-able-r
redd-er blu-er green-er *violet-er *crimson-er
pink-er whit-er black-er *ivory-er *turquoise-r *azure-r
browner blonder gray-er *indigo-er
Affix -er also attaches to some 2-syllable adjectives, most often to syllables with no final
consonant:
happy-er tidy-er friend-ly-er mellower heavier
purpl-er orange-r little-r
Puzzling cases:
quiet-er *just-er *apt-er ?dead-er
un-happy-er un-tidy-er un-health-y-er un-luck-y-er
unruly-er
Maybe a combination of features determines -er acceptability
number of syllables cold-er happi-er ?water-y-er *temporary-er
weight of final syllable happi-er *hapless-er
non-initial stress un-happi-er *slippery-er
historical group (‘latinate’) fast-er *just-er
degree vs binary stronger ?dead-er
The affixation rules are recursive, but with these largely unspecified restrictions on pro-
ductivity, there are limits to how many suffixes we get:
symbol-ize to put into symbols
idol-ize to make into an idol
? kill-er-ize to make into a killer
? kill-er-ize-r someone who makes people into killers
? kill-er-ize-r-ize-r someone who makes people make people into killers
3.1.3 productive compounding
Now let’s compare noun compounding in English. Earlier, we considered examples like these:
blackbird apple pie student protest biology laboratory safety precautions
We saw that these these nouns can be grouped into pairs in a way that makes semantic sense.1
We could write a rule that says that a noun can be made from two nouns:
N → N N.
1The text notes that stress works in an interesting way in compounds. There is an exercise later in the text
(we get to it in a few weeks) on how the default stress for these compounds is influenced by the grouping, the
bracketing.
24

It seems that the number of syllables in each noun and such phonological properties of the
nouns do not matter. These are productive enough that the recursion is apparent: there is no
fixed limit on the length of noun compounds in English.
This is interesting for many reasons. First, this is the kind of rule that makes human
languages infinite. Second, these compounds have such a simple structure, and they are
present in some form or other in most languages, that it has been proposed that they are the
fossils of the earliest structures of human language (Jackendoff 2009, Bickerton 1990), possibly
a relic of the earliest ‘protolanguages’. Third, productive noun compounds with more than 2
words are highly ambiguous – you have to know what is meant to understand them.2
As
we will see, most English sentences have some similar structural ambiguities. Let’s briefly
consider compounds a little more closely, to correct some possible misconceptions from our
earlier discussions and to set the stage for other parts of syntax.
We observed that many familiar noun compounds are idiomatic, but we did not really say
what the non-idiomatic noun compounds mean. One linguist has claimed that there are 9
basic relations:
NN a. onion tears cause (Levi 1978)
b. vegetable soup have as part
c. music box make
d. steam iron use
e. pine tree be
f. night flight in
g. pet spray for
h. peanut butter from
i. abortion controversy about
And when we consider nouns related to verbs, it seems various grammatical roles are possible
N[N V] a. heart massage obj
b. sound synthesizer obj
c. child behavior subj
d. car lover obj
e. soccer competition (at|in? modifier)
f. government promotion subj|obj
g. satellite observation subj|obj
If you don’t know the meanings of the compound (or the words in it), you will not be able to
tell how it should be interpreted.3
These English examples look similar to
2The number of different binary branching trees for n leaves is (2(n−1))!
(n)!(n−1)!
(Catalan, 1838). For n =
2, 4, 8, 16, . . . leaves, the number of possible trees is 1, 5, 429, 9694845, . . .. So for a noun compound with 8
words, like boron epoxy rocket motor chamber sales market potential, there are catalan(8) = 429 different
binary trees.
3Jackendoff mentions a New York Times story (3 June 2007) about child camel jockey slavery which many
people do not know how to interpret, at first. It is a story about children serving as camel jockeys, and so it
is bracketed [[child [camel jockey]] slavery]. There is a wikipedia page about child camel jockeys.
25

Quechua rumi ñan stone road
warmi wagra female cow
German Geshirr-spüler dish-washer
Eisen-bahn iron track
Spanish auto-escuela car school
cine-club cinema club
tele-novela television novel
video-arte video art
It is sometimes claimed that Mandarin Chinese is left and right headed
Chinese [[du]N [fàn]N ] drug-criminal (Ceccagno&Basciano 2009; Packard 2000)
[[hán]N [shou]V ]V letter-sell ‘order by mail’
[[j´ın]V [du]N ]V prohibit-poison ‘ban the sale and abuse of drugs’
The following VN examples seem to have a different kind of structure, and these have
mainly idiomatic meanings:
French essui-glace wipe-window windshield wiper
couche-tard lie-late one who stays up late
Italian lava-piatti washes dishes dishwasher
porta-sapone liquido container soap liquid liquid soap container
Spanish abre-latas opens cans can opener
saca-corchos removing corks corkscrew
English [jump-rope]V [jump-rope]N
[pick-pocket]V [pick-pocket]N
[break-fast]V [break-fast]N
[cut-throat]N [scatter-brain]N [twinkle-toes]N
3.2 Parts of speech, syntactic atoms
Syntax is the theory of phrases. A sentence is a certain kind of phrase, and when we look at
these phrases, we see that many of them have parts. The idea that an expression can be a
complex of parts is already familiar syllable structures in phonology and word structures in
morphology. A tree structure is a diagram of what the parts are and how they are assembled.
(Compare a diagram of a car, which might show that the engine and the body are different
parts, each of which is in turn assembled from more basic parts.) Linguists often call the parts
of a sentence its “constituents,” but you can just as well call them parts or pieces or units.
The justification for the particular units of structure that we use in syntax comes from
their use in our account of how structures are built. We used the same logic in phonology.
The motivation for a phonological feature like voice and sonorant, for example, came from
its utility in describing speech sounds – we found that vowels were lengthened before +voice,
−sonorant sounds. Similarly in morphology. Some suffixes can only attach to roots, so we
conclude that roots are important units of word structure. Now we introduce some basic units
of phrase structure. The discussion of syntax begins in Chapter 3 of the text.4
In syntax, building phrases, the action begins at the parts of speech – roughly the word
level — though at some points we will dig in to look at affixes. As we will see, many of the
categories of expressions at this level and above will have infinitely many elements in them.
4Although the whole chapter is good reading, we will focus initially on the introductory pages 89-101 and
then, next week, on the methods for figuring out what the structure is on pages 138-185.
26

We have already introduced word-level categories: N (noun), V (verb), A (adjective), P
(preposition), Adv (adverb), D (determiner). A category is just a class of expressions that
share some important properties. We have already seen some of the distinctive morphological
properties of these categories of expressions. For example, they differ with regard to which
affixes they allow. Many adjectives can be suffixed with -ly to form an adverb; but no de-
terminers can. We have also seen that these different sorts of expressions differ with regard
to the possibilities for compounding. English allows noun compounding quite freely, but does
not freely allow determiner compounding. So when we label an expression N, we are saying,
among other things, that this is an expression of the kind that can be used to build compounds
in a certain way.
These same categories of expressions, (N, V, A, P, Adv, D) have distinct roles in syntax.
We will begin with the hypothesis that these are the “syntactic atoms”, the basic units that
will be referred to in our account of how phrases are built in human languages. We get a
first, basic appreciation of the importance of these categories by noticing that these categories
predict certain “distributional” properties, properties having to do with where a word can
occur in a sentence.
In our discussion of morphology, parts of speech were introduced by where they occur, by
their distribution in the language. For example, where one verb can occur, other similar verbs
can usually occur:
(1) They can run/hide/fly/go/break
Where one determiner can occur, others can too:
(2) The/a/some/one/every/each book was read by every student. The/some/two/all/most
books were read by every student.
Contexts where adverbs occur:
(3) He serves us politely/badly/happily
(4) He quickly/slowly served the dinner
And adjectives:
(5) the exciting/big/hot/huge/depressing/gleeful/rugged/red book
Some word classes are less familiar than the ones we have already mentioned (N, V, A, P, Adv,
D), and so we will introduce some additional categories. For example, at the beginning of the
following sentence, the words that can occur are called modals:
(6) Can/could/shall/should/must/would I be frank?
There are some contexts where the only word that can occur is some form of the verb to be:
(7) Is/was he going out
This suggests that although be may be a kind of verb, it’s really in a category by itself!
Notice that the words that/this can appear in the position of determiners (at least to a
good first approximation):
(8) this/that/the/every book is on sale
27

(9) I bought this/that/the/every book
So here, the words that/this are presumably determiners. We might call them “demonstrative”
determiners. There is another use for that which is quite different though:
(10) I know that it is on on sale
Here, what follows that is a sentence, not a noun. In this use, it is a complementizer (C).
There are other complementizers:
(11) I wonder whether it is on on sale
(12) I wonder if it is on on sale
We will say more about complementizers later.
The distributional properties we have been looking at correspond with certain basic se-
mantic properties. These may be more familiar, but they are actually less useful.
(13) a. verbs V typically denote actions (run, eat, buy, break)
b. nouns N typically denote entities (dog, cat, number, word)
c. adjectives A typically denote states (hot, sick, old)
d. adverbs Adv typically denote manners (slowly, cynically, painfully)
e. prepositions P typically denote locations or relationships (near, far, with, about)
f. determiners D serve to specify or quantify a noun (the, this, which, five)
g. complementizers C introduce “clauses” or “sentences”
These rough semantic criteria are very unreliable! Running specifies an activity, but is a
perfectly good noun. Love specifies a relationship, but is also perfectly good noun. fast may
seem to specify a manner, but it is an adjective. What we are interested in is the role these
words play in building phrases, so we get much more reliable evidence from “distributional”
arguments of various kinds than from intuitions about what these expressions mean or denote.
3.3 Categories and “finest” categories
It is easy to see that dog and dogs cannot occur in all the same positions: they are not exactly
the same category. If we carve up the categories in the finest way possible, singlular nouns
and plural nouns (as we are now calling them) will have to be in different categories. In the
same “finest category” with dog we certainly have cat, horse, rabbit,. . . , but not plural nouns.
Similarly, for the (traditionally masculine) name Bill; it is certainly in the same finest category
as John, Fred, Sam,. . . . But there are other words that seem to be in a special category by
themselves, like is, or has – make sure you can present evidence that, for example, is and was
are not in exactly the same category, nor are is and has, nor is and reads, or anything else.
We will give a lot of attention later to these words that are in special finest categories by
themselves.
28

3.4 Substitutions and Phrases
Consider the distribution of the pronoun “she.” We can see that this does not have the distri-
bution of any of the categories already mentioned (N(sg,pl),V(aux,main),A,P,Adv,D(sg,pl)).
We find it in contexts like this:
(14) she reads
(15) she is a doctor
(16) the report says she discovered the answer
We notice that in these cases, the pronoun appears to be the subject of a sentence. What
other things can be substituted into those contexts? That is, what other things can be a
subject in these contexts?
a. other (sg, 3rd person) pronouns: he, it
b. demonstrative determiners: this, that
c. certain phrases (containing determiners): a scientist, the manager, every manager
of the department, a friend of mine, the former White House chief-of-staff
Checking other pronouns, we find that they often occur where certain determiners and
phrases with determiners can occur. To a first approximation, we can divide the contexts up
this way
nominative case pronouns, in subject position: I/you/he/she/it/we/they studied the
report
accusative case pronouns, in object position: the project studied me/you/him/her/it/us/them
genitive (possessive case) pronouns, in determiner position: Al read my/your/his/her/its/our/their
book
Notice that nouns can have adjective modifiers and determiners associated with them;
verbs can have adverb and prepositional modifiers associated with them, adjectives can have
degree phrases and adverb modifiers associated with them, adverbs can have degree phrases
and adverb modifiers associated with them:
(17) [the happy students] are reading linguistics
(18) the balloon [quickly rose up]
(19) He is [so completely happy]
(20) He is [so totally completely] finished
But pronouns, demonstrative determiners, and the phrases with determiners cannot have any
other modifiers or determiners added to them:
(21) * [the happy he] reads linguistics
(22) * [one expensive this/that] is being sold
(23) * [the happy a scientist] reads linguistics
29

In a sense, pronouns are complete: no more modifiers (*tall he left) or determiners (*the he
left) are allowed.. For these reasons, we give pronouns the category: determiner phrase (DP).
We give them the phrasal category because they are complete in the same way that the phrase
the happy student is complete.
So we have identified our first phrases: these are the things that have the same distribution
as prounouns do. And we can use a pronoun substitution test to identify these things.
(The reasons for calling these things ‘phrases’ will become clear later.) So we have two kinds
of tests for identifying the parts, the “constituents” of a phrase so far:
First constituency tests:
(24) Substitution tests for N(sg,pl),V(aux,main,forms),A,P,Adv,D(sg,pl): ‘frame-
based’ reasoning – finding items that occur in similar contexts
(25) Pronoun substitution test for determiner phrases (DPs): The unit we will call
a determiner phrase (DP) can often be replaced by a pronoun
a. The store with the top hats has closed
b. It has closed
This perspective has introduced a couple of basic ideas. First, obviously, words go together
to form larger units of structure. That is, in our account of how phrases are assembled, it is
natural to refer to units that have more than one word in them. We did this earlier when we
talked about “sentences”, “subjects,” “objects” and so on, but now we will try to be more
systematic about it. The justification for assuming that some sequence of words forms a unit
comes from the fact that we refer to this unit in the syntax. We use “constituency tests” to
provide some preliminary evidence. They are preliminary checks on the units that our syntax
is likely to refer to. These are our starting point, and so they are very important!
3.5 Manipulations
We not look at some more significant structural manipulations. Instead of just trying to put
one piece of an expression in for another, we try “twisting” the sentence into various related
forms, seeing what pieces tend to hang together in these twists, like in a puzzle where part ot
the problem is to figure out what the pieces really are.
One thing we can do to a sentence to get a very closely related form is to move a piece of
it to the front. For example, consider the student from Venice in the following sentences,
where it can appear in different positions, but not left out:
(26) a. I like the student
b. the student , I like
c. * I like
If we try moving arbitrary other parts of the sentence to the front, we get really strange results:
(27) a. * student, I like the
b. * like the student, I
The element that moves easily, the student, is a determiner phrase. Other kinds of things seem
to be able to move in the same way, but they are all phrases:
30

Two more constituency tests:
(28) Preposing (topicalization) test for phrases: Only phrases can be preposed.5
a. i. I saw the picture of the statue
ii. the picture of the statue, I saw
iii. * of the statue, I saw the picture
iv. * picture of the statue, I saw the
b. i. I ate with a spoon
ii. with a spoon, I ate
iii. * a spoon, I ate with
c. i. he may be completely reckless
ii. completely reckless, he may be
iii. * reckless, he may be completely
iv. * completely, he may be reckless
d. i. the witches stirred the cauldron very slowly
ii. very slowly, the witches stirred the cauldron
iii. * slowly, the witches stirred the cauldron very
iv. * very, the witches stirred the cauldron slowly
(29) Postposing test for phrases: Only phrases can be postposed (much less ﬂexible).
a. The student [that I told you about yesterday] arrived.
b. the student arrived that I told you about yesterday
[1] Bickerton, Derek (1990) Language and Species. Chicago, University of Chicago Press.
[2] Booij, Geert (2009) Compound construction: Schemas or analogy? In S. Scalise and I. Vogel,
eds., Compounding, Philadelphia: Benjamins.
[3] Ceccagno, Antonella, and Basciano, Bianca (2009) Sino-Tibetan: Mandarin Chinese. In R. Lieber
and P. Stekauer, eds., The Oxford Handbook of Compounding, 478-490. Oxford: Oxford University
Press.
[4] Downing, Pamela (1977) On the creation and use of English compound nouns. Language 53(4):
810-842.
[5] Jackendoﬀ, Ray S. (2009) Compounding in the parallel architecture and conceptual semantics. In
R. Lieber and P. ˇStekauer, eds., The Oxford Handbook of Compounding, Oxford: Oxford University
Press.
[6] Lapata, Maria (2002) The disambiguation of nominalizations. Computational Linguistics 28(3):
357–388.
[7] Levi, Judith (1978) The Syntax and Semantics of Complex Nominals. NY: Academic Press.
[8] Packard, Jerome (2000) The Morphology of Chinese: A Linguistic and Cognitive Approach.
Cambridge: Cambridge University Press.
5The text, on page 152, says that nouns, adjectives, and adverbs can be topicalized. But the examples below
show that it is noun phrases, adjective phrases, and adverb phrases can be topicalized. In fact, this is what
the text on page 153 is driving at. You cannot prepose a noun out of the phrase that it is the head of. The
text also says that verbs can be preposed. Verbs and verb phrases are a special case, with special properties,
so I leave it out until we come back to look at verb phrases in detail later.
31

32

Lecture 4 The anatomy of a phrase
4.1 More consituency tests
So far we have seen the most important
(1) Substitution tests for N(sg,pl),V(aux,main,forms),A,P,Adv,D(sg,pl),
noting that some of these categories are not “ﬁnest categories,” but something a little
broader than those.
(2) Pronoun substitution test for determiner phrases (DPs)
We can also identify elements by “manipulations:”
(3) Topicalization (preposing) test for phrases: Only phrases can be preposed.
(4) Postposing test for phrases: Only phrases can be postposed.
a. The student [that I told you about yesterday] arrived.
Another thing that can be noticed about phrases, is that they often constitute acceptable
answers to questions, even when the answers are not complete sentences:
Constituency test 5
(5) Sentence fragment test for phrases: Only phrases can be answers to questions
a. Who came? Maria
b. Who came? the president of the student union
c. Who came? * the
d. Who came? * the president of
e. What did she do? ran the student council for 2 years.
f. When did she do it? last year
g. How did she do it? with lots of hard work
h. How did she do it? quickly
i. How did she do it? * with
Constituency test 6
(6) Coordination test for constituents of the same type: For the most part, con-
stituents are coordinated (using and, or, but,. . . ) with other constituents of the same
category.
33

a. [DP The old man] and [DP Maria] went over the books.
b. They [V wrote] and [V rewrote] the address.
c. * They [V wrote] and [NP Maria] the address.
d. They looked [P up] and [P down]
e. They [VP wrote it] and [VP sent it out]
4.2 Determiner phrases: first thoughts
Our constituency tests show us, no surprise, that in the following sentences, the sequences in
the brackets form units:
(7) [cats] sleep
(8) those [cats] sleep
(9) I saw [cats]
(10) I saw those [cats]
(11) [Orange cats] sleep
(12) [Orange orange cats] sleep
(13) [Orange orange orange cats] sleep
(14) . . .
(15) I saw [orange cats]
(16) [Terrifying orange cats] sleep
(17) I saw [terrifying orange cats]
(18) [big terrifying orange cats] sleep
(19) I saw [big terrifying orange cats]
(20) . . .
In these examples we see that the plural noun cats can be modified by any number of adjectives
(A). What kind of unit is [orange cats]? Is it a noun? To say it is a noun would suggest that
[orange cat] is a singular noun that can enter into noun compounds the way cat does, but this
is not right:
(21) tiger cat
(22) tiger cat salesperson
(23) * tiger orange cat
(24) * tiger orange cat salesperson
This sugests that [orange cat] is not a noun. What are the properties of expressions like [orange
cats]? One is: they do not enter into compounding the way nouns do. Another property we
see in (7)-(20) is that expressions like orange cat can be modified by more adjectives. So NP
is also a category that can be modified by adjectives. One way of putting this (a way that
will get further confirmation from other sorts of considerations later) is to say that when we
add a modifying adjective to a NP, we get another NP, another category that can be modified
in the same way. The units [orange cats], [big orange cats], and so on are not nouns. They
are phrases. Pronoun replacement suggests that it is a DP in examples like (7), but we also
34

see examples like (8) and (10) where is seems to be part of a DP. So we could assume that
[cats] and [orange cats] are some kind of phrases in both (7) and (8), but in (7) it is a DP
in addition. So let’s call [cats] and [orange cats] a noun phrase (NP) in both (7) and (8) –
which seems reasonable – and this is compatible with saying that in (7) that NP is also a DP.
While plural expressions like [orange cats] can be the subjects of sentences, where they can
be replaced by pronouns:
(25) [orange cats] sleep.
(26) [they] sleep.
We notice that this is not possible with the singular [orange cat]
(27) *[orange cat] sleeps
We also notice that DPs like [the orange cats] or [they] cannot be modified the way [cats] or
[orange cats] can be:
(28) * big the orange cats sleep
(29) * big they sleep
So what we want to say here is that a NP can form a DP either by itself or with modifiers.
This is getting complicated, so let’s start drawing out the situation with trees:
S
DP
D
the
NP
AP
A
orange
NP
N
cats
VP
V
sleep
S
DP
NP
AP
A
orange
NP
N
cats
VP
V
sleep
S
DP
NP
N
cats
VP
V
sleep
Notice that, in all of these trees,
(30) the is a D
(31) orange is an A
(32) cats is a N
But only in the third tree is cats a DP.
So we call the kind of things that we see as subjects in these sentences determiner phrases
(DPs). Then we see that [cats] is a N; it is also a NP, and it is also a DP. The expression
[orange cats] is a NP but not a N. And the expressions [the orange cats], with the determiner
the in it, is a DP but not a NP and not a N.
Putting all of this together, we can think of the N cats as having the potential of collecting
certain associated units to make a phrase: it can combine with 0 or more modifiers to make a
NP. Then the NP can combine with 0 or 1 determiner to make a DP. We will see that other
sorts of phrases have collect similar kinds of units.
35

As the text observes on page 165, to allow for any number of modifiers, we need a recursive
rule. The text proposes a rule that combines an adjective with a NP to make a NP, but since
the adjectives can themselves be modified, we will propose this rule:
NP → AP NP
This says that a NP can be formed from an adjective phrase and a NP. With this idea, we get
structures like this:
S
DP
D
the
NP
AP
A
big
NP
AP
A
orange
NP
N
cats
VP
V
sleep
S
DP
D
the
NP
AP
A
big
NP
AP
A
frightening
NP
AP
A
orange
NP
N
cats
VP
V
sleep
Coordination tests confirm the idea that the adjectives and noun form a constituent to the
exclusion of the determiner:
(33) [the [red book] and [green book]] cost 15 dollars.
And we show this in our trees:
S
DP
D
the
NP
AP
A
exciting
NP
N
book
VP
V
surprised
DP
me
S
DP
D
the
NP
NP
AP
A
red
NP
N
book
Coord
and
NP
AP
A
green
NP
N
book
VP
V
cost
DP
D
15
NP
dollars
4.3 Arguments and modifiers
Sentences are formed when words – heads – combine in appropriate ways with other units.
In effect, each head can have other constituents as “satellites”. We will distinguish two basic
kinds of satellites: arguments and modifiers. We will consider arguments more carefully next
time.
36

4.4 Summary
We introduced important tests for constituency – (we will get much more practice, but you
need to know them!) We showed how phrase structure could be indicated in trees or with
brackets. The relations in the trees can be represented with rules.
37

Summary: syntactic constituency tests
(1) Substitution tests for N(sg,pl),V(aux,main,forms),A,P,Adv,D(sg,pl): we started
with these
(2) Pronoun substitution test for determiner phrases (DPs):
a. The store with the cool stuﬀ has closed
b. It has closed
(3) Topicalization (preposing) test for phrases: Only phrases can be preposed.1
a. i. I saw the picture of the statue
ii. the picture of the statue, I saw
iii. * of the statue, I saw the picture
iv. * picture of the statue, I saw the
b. i. I ate with a spoon
ii. with a spoon, I ate
iii. * a spoon, I ate with
c. i. he may be completely reckless
ii. completely reckless, he may be
iii. * reckless, he may be completely
d. i. the witches stirred the cauldron very slowly
ii. very slowly, the witches stirred the cauldron
iii. * slowly, the witches stirred the cauldron very
iv. * very, the witches stirred the cauldron slowly
e. i. Robin said she would sing a song. . . and
ii. sing a song, she did
iii. * sing, she did a song
(4) Postposing test for phrases: Only phrases can be postposed.
a. The student that I told you about yesterday arrived
(5) Sentence fragment test for phrases:
a. Who came? Maria
b. Who came? the president of the student union
c. Who came? * the
(6) Coordination test for constituents of the same type: For the most part, con-
stituents are coordinated (using and, or, but,. . . ) with other constituents of the same
category.
a. [NP The old man] and [NP Maria] went over the books.
b. They [V wrote] and [V rewrote] the address.
c. * They [V wrote] and [NP Maria] the address.
d. They looked [P up] and [P down]
e. They [VP wrote it] and [VP sent it out]
1The text, on page 152, says that nouns, adjectives, and adverbs can be topicalized. But the examples
below show that it is noun phrases, adjective phrases, and adverb phrases can be topicalized. This is what the
text on page 153 is driving at. For example, we see that you typically cannot prepose a noun out of the phrase
that it is the head of!

Lecture 5 Syntactic heads and rules
[NB: as mentioned in class, we are still covering material from last week’s reading. The second
syntax chapter in the book (Chapter 4) covers several things that we will not go into in detail,
so the new required reading for this week is quite short. The most important things will be in
the lecture notes.]
Looking over the syntactic structures we drew last time and reviewing especially that last
two pages of the previous lecture notes, we can see some basic things going on:
• D, N, V, A, Adv, P, C are heads (=syntactic atoms, ‘basic parts of speech’) which
combine with 0 or more phrases (e.g. direct or indirect objects) to form DP, NP, VP,
AP, AdvP, PP, CP.
The heads are given to us by morphology, and there are infinitely many of them. Notice that
whenever we have a noun N, we will have an NP. And in the class, we will assume that the
same goes for D, V, A, P, Adv. (This is a slight departure from the text, where it is assumed
that, for example, you can have an A without an AP.)
• Modifiers are optional, and they can combine (recursively) with NP, VP, AP, AdvP,
PP
• When we think about where modifiers appear, we notice that a modifier of a head X
usually cannot appear between X and its complements:
the student finished the homework
slowly * * *
big * * * *
• So in cats sleep. In that sentence, we say that the noun cats is a DP, because it plays
the same role in the syntax as a phrase like the cats with a determiner does.
As discussed in class, we can understand these options now, simply with the idea that APs
attach to the left of NPs, and AdvP attaches on either side of VP (unless they are preposed
– in that special case they must attach to something else).
5.0.1 Arguments of VP introduced
The direct object of a verb is called its argument or complement, such as Elizabeth in
(1) Mary [VP praised Elizabeth]
Mary is the subject, external to the VP. The subject is usually also regarded as an argument.
Semantically, the subject and the direct object refer to things that are essentially involved
in the action named by the verb. This is typical of arguments, and distinguishes them from
modifiers. It is not natural to think of Elizabeth as modifying praised in (1).
39

S
DP
Mary
VP
V
praised
DP
Elizabeth
Notice that the DP Elizabeth is a sister of the head V. In this configuration, the pattern VP
→ V DP is not recursive.
An indirect object is a argument of the verb as well, such as Elizabeth in
(2) I [VP gave an exciting book to Elizabeth]
(3) I [VP gave Elizabeth an exciting book]
Here we have two DPs inside the VP, two arguments: a direct object (an exciting book) and
an indirect object (Elizabeth).
S
DP
Mary
VP
V
gave
DP
D
an
NP
AP
A
exciting
NP
N
book
PP
P
to
DP
Elizabeth
S
DP
Mary
VP
V
gave
DP
Elizabeth
DP
D
the
NP
AP
A
exciting
NP
N
book
Notice that, in the tree on the right, the DP Elizabeth is again a sister of the head V. In this
configuration, the pattern VP → V DP DP is still not recursive, which fits with the fact that
we cannot just stick more DPs into the VP:
(4) * I [VP gave Elizabeth the book the door Friday Mary Hamlet]
So just a limited number of DP arguments are possible in VPs, in a certain sense which we
will try to get clear about. (Notice for example, that we can have all these VPs The maximum
number of complements seems to be 3, in all human languages.1
Some English verbs require 2 complements, like put, but I do not know of any that require
3. But some verbs appear to allow 3, as in I bet [Mark] [5 dollars] [that they would win].
S
DP
I
VP
V
bet
DP
Mark
DP
D
5
NP
N
dollars
CP
C
that
S
DP
he
VP
V
would
VP
win
1For discussion of this idea, see for example Pesetsky 1995.
40

It is an interesting fact that it is hard to think of verbs other than put which require 2
complements. I know only a few: hand, maybe also set, lodge.
Sometimes complements are mandatory, sometimes optional. But modifiers are always
optional.
(5) I praised Mary.
(6) ? I praised.
(7) I put the car in the garage.
(8) * I put in the garage.
This confirms the intuitive idea that arguments play a fundamentally different role in syntax
from modifiers. When you look up a verb in a good dictionary, it will usually tell you something
about what objects it can take, since verbs are fussy about this, but the dictionary will not
tell you what modifiers the verbs allow, since most verbs allow most modifiers!
5.0.2 Modifiers in VP
As we have seen, Like the modifiers in NP, we can have any number of modifers in a VP:
(9) I [VP gave Elizabeth the book]
(10) I [VP happily gave Elizabeth the book]
(11) I [VP happily gave Elizabeth the book on Friday]
(12) I [VP happily gave Elizabeth the book on Friday in class]
(13) I [VP happily gave Elizabeth the book on Friday in class with the rest of my notes]
(14) . . .
In order to add any number of modifiers like this, we need recursive rules, like the one we
proposed for noun phrases.
VP → AdvP VP
VP → VP AdvP
VP → VP PP
VP → PP VP
Drawing these ideas out in trees, we get structures like these:
S
DP
I
VP
V
gave
DP
Elizabeth
DP
D
the
NP
N
book
S
DP
I
VP
AdvP
Adv
happily
VP
V
gave
DP
Elizabeth
DP
D
the
NP
N
book
S
DP
I
VP
VP
AdvP
Adv
happily
VP
V
gave
DP
Elizabeth
DP
D
the
NP
N
book
PP
P
on
DP
Friday
Now lets go back to consider what counts as an argument versus a modifier in a VP. Direct
and indirect objects are not modifiers, but arguments, as in:
41

(15) I sent money
(16) I sent Mary money
(17) I sent money to Mary
Notice that the indirect object can appear in a PP. We cannot have multiple indirect objects:
(18) * I sent [Bill] money [to Mary] [to Sam]
However, the number of PPs that can be included in a VP seems essentially unlimited:
(19) I worked on Sunday in the city on that project without a break.
Are all of these PPs arguments? Or are some of them modifier modifiers? In fact, it seems
that there is a principled difference between arguments and modifier PPs, revealed by tests
like the following.
Tests for modifiers and arguments in VP beginning with the easiest. . .
(20) Optionality: Arguments are sometimes required, sometimes optional. Modifiers are
always optional.
a. I praised Mary
b. ? I praised
c. I sang [on Saturday]
d. I sang
(21) Iteration: The number of arguments is strictly limited. There can be any number of
modifiers. That is, modifiers can be “iterated” or “repeated.”
a. I sang with gusto on Saturday with Mary about love at the auditorium.
(22) modifiers can modify coordinated XPs:
a. Robin [VP wrote a book] and [VP sang three songs] [with Sandy.]
Linguists sometimes try trickier kinds of “manipulation” tests – more tests where we con-
sider the possible ways of rephrasing a sentence by rearranging its parts. (You do not need
to remember the following ones, and they wiill not be on the quizzes, but they should be
understandable, and we might mention them in class.)
(23) modifiers can be left behind in VP-preposing: Is [with Sandy] a complement of
V in [Robin said she would sing a song with Sandy]? No!
a. Robin said she would sing a song, and [sing a song] she did, [with Sandy].
b. * Robin said she would give Mary a book, and [give Mary] she did, [a book].
(24) modifiers can modify “do so” VPs: Is [with Mary] a complement of V in [I sang
a song with Mary]? No!
a. I sang a song with Mary while you did so [with Bill]. (modifier)
b. * I saw Bill while you did so [Mary]. (argument)
(25) unlike arguments, modifiers are OK in “do what” pseudoclefts: Is [with Bill]
a complement of V in [Mary sang a song with Bill]? No!
a. What Mary did [with Bill] was sing a song. (modifier PP)
42

b. * What Mary did [Bill] was give a book. (argument NP)
Usually, these tests provide convergent evidence about the status of any given PP. When these
tests yield different results, it is less clear what to say about the structure.
In sum, when a verb V and its direct object DP form a verb phrase VP, we say that the DP
is a complement. A complements is a sister of the head. Complements, together with the
subject, are also called arguments. The notion “complement” usually means the structural
relation, sister of the head, while the word “argument” usually means the semantic relation:
usually a specification of what the verb applies to, one of the essential components of the
event.
5.1 S(emantic)-selection and argument roles, ‘θ-roles’
For understanding the difference between arguments and modifiers, reasoning about what the
sentences mean can be helpful. I think of these as “shortcuts” – These will not be on the
quizzes, but it is often useful to think about what the sensible elements of phrases are. For
example, subjects often name the agent of the action, and the object often names the patient
or theme of the action. It is hard to pin down these notions precisely, but the text (pp.129,
227) points out that sentences like the following are very odd (the text uses ! to mark these
odd sentences):
(26) ! the rock murdered the tree
(27) ! the forest described Macbeth
This oddity comes from facts about what the verbs mean, facts about the semantic role that
the subject and object are expected to play. The verbs murder and describe normally have
subjects that are ‘agents’, animate agents. Since this is a matter of what the DPs mean, what
they refer to, they are called s(emantic)-selection requirements.
Consider a bunch of simple English sentences like these:
S
DP
Othello
VP
V
murders
DP
Desdemona
S
DP
Othello
VP
V
describes
DP
Desdemona
S
DP
Othello
VP
V
calls
DP
Desdemona
S
DP
Othello
VP
V
attacks
DP
Desdemona
In all of these sentences, the subject is, in some sense, the agent of the action denoted by the
verb, while the object is more passive – the text says that the objects in these sentences are
patients of the actions denoted by the verbs. In contrast, if we compare
S
DP
Othello
VP
V
likes
DP
Desdemona
S
DP
Othello
VP
V
pleases
DP
Desdemona
In the first of these latter sentences, the subject is the experiencer and the object is the
theme – the thing that the psychological attitude denoted by the verb is about. But in
the second sentence, the subject is the theme and the object is the experiencer. We expect
that our sentences will not only be assembled appropriately, but that the elements will have
appropriate sorts of meanings.
43

5.2 Syntactic rules
The way each constituent – each “node” in a tree – can be composed of parts – its “children”
of the node – can be represented with a rule. For example, we see in each of the trees just
displayed that each S “node” has two children – a DP and a VP. We represent this fact with a
“phrase structure rule”: S → DP VP. Remember that the arrow in this rule can be read “can
be formed from.” Collecting all the rules from all the trees we have displayed, we find these:
basic rules for rules for
selected elements modifiers
(137) S → DP VP
(124a) DP →



(D) NP
Name
Pronoun



(123) NP → N (
PP
CP
)
(130) VP → V (DP) (



PP
CP
VP
AP



)
(120) PP → P (DP)
(122) AP → A (PP)
(138) CP → C S
AdvP → Adv
(123’) NP → (A) N (PP)
(73c.iii) NP → AP NP
NP → NP PP
VP → AdvP VP
VP → VP AdvP
VP → VP PP
AP → AdvP AP
α → α Coord α (for α=S,D,V,N,A,P,C,Adv,VP,NP,AP,PP,AdvP,CP)
Remember that parentheses go around optional elements. And set brackets go around elements
when you have to choose exactly one. These rules are not complete! but they are a good
starting point, and they are, I think, intuitive.2
2These rules are are slightly simpler and more general that the “complete list of phrase structure rules” on
page 175 of the text. Our rules differ from the text primarily in these respects:
a. the “complete list” on p175 of the text includes rule (122) for APs, but it does not provide any rule that
let’s an AP occur in a sentence, so we add a few modifier rules.
b. As discussed in class, the text does provide rule (123’) on page 165, but once we have (73c.iii) also from
page 165, we can eliminate (123’).
c. To allow sentences like cats sleep, we need to allow the determiner in rule (124a) to be either unpronounced
(“empty”) or optional. The rules in the text do not indicate the fact that there can be no (pronounced)
determiner before the noun.
d. We add a rule for coordinate structures. This rule is special, because it works on so many categories.
44

5.3 Arguments in PP, NP and AP
Just as the traditional ‘object’ of a verb is an argument, so the object of a preposition is its
argument. It is not always required, and there are other possible arguments besides DP:
(28) He looked [PP up]
(29) He looked [PP up [DP the chimney]]
(30) He walked [PP up [PP to the chimney]]
(31) He is [PP beside [DP me]]
(32) He walks [PP with [S you leading the way]]
These phrasal arguments are attached as sister to P.
The rules given above also include the possibility of arguments for NP and AP. Intuitively,
Sarajevo is a argument of V in the sentence:
(33) They destroyed Sarajevo in 1993
In a similar way, the phrase [PPof Sarajevo] seems to be an of N in the sentence:
(34) The destruction of Sarajevo in 1993 was terrible
We will accordingly attach [PPof Sarajevo] as a sister to N, and we attach [PPin 1993] as an
modifier. This fits with the fact that neither of the following are any good:
(35) * They destroyed in 1993 Sarajevo
(36) ?? The destruction in 1993 of Sarajevo was terrible
We use this idea for nouns that are clearly related to verbs that take complements, like claim,
study, idea, proposal, thought, fear, desire, argument,. . . . For example, consider
(37) The argument [of the noun] [in this sentence] is familiar
(38) ?? The argument [in this sentence] [of the noun] is familiar
In this case too, we say [of the noun] is a argument, a sister of the N argument, but [in this
sentence] is a modifier, a sister of NP.
We can also get arguments in APs, arguments which are rather similar to DP arguments.
For example,
(39) That was [AP clever]
(40) That was [AP clever [PP of Mary]]
(41) It is [AP important]
(42) It is [AP important [PP to Bill]]
5.4 Review so far.
We have observed that phrases are built according to some very general principles:
45

• Each head of category X combines with its complements (possibly none) to form
XP.
So complements are always sisters of the head.
• Modifiers are optional phrases, and they can combine (recursively) with the phrase
they modify
In English specifically, we find phrases combining in certain orders, which we represent with a
set of rules like the ones given on page 44 of these lecture notes.
We noticed that the rules given in class and on page 44 are too permissive – for example,
while some verbs select DP objects, others do not – so we have to factor in the specific require-
ments of each lexical item. We say that each lexical item can “select” certain other phrases
that it needs to combine with. There are 2 kinds of selection: ‘categorial’ and ‘semantic’.
5.4.1 C(ategorial)-selection
Words get put into the syntactic structures according to their category. For example, a verb
can go where the rules allow a V to occur. However, an intransitive verb like laugh can only
go into a tree that does not have a direct object. A transitive verb like surprise, on the other
hand, can go into a tree with a direct object, but not one with both a direct and indirect
object.
We can assume that each word is associated with a specification of the arguments that it
can occur with. The arguments of a verb are its subject and its complements. Let’s adopt
the policy of underlining the subject argument, so that we will not distinguish it from the
complements.
word category arguments
laugh V DP
surprise V DP DP
send V DP DP
send V DP DP DP
These are important distinctions among elements of the category V – the category divides
into subcategories according to the categories of the arguments each verb requires.
Some verbs allow sentences, or CP clauses of the form [CP that S] as arguments. For
example,
(43) John knows [CP that [S Mary laughs]]
(44) John told [DP Bill] [CP that [S Mary laughs]]
Infinitival clauses (using the ’infinitive’ form of the verb, with to) can also occur as arguments:
(45) John knows [S her to be a dedicated linguist]
(46) I prefer [S her to explain it]
These infinitival clauses have accusative case subjects, and the verb in them is never present
or past, but infinitival.
We can represent the way that different verbs assign different roles to their arguments by
augmenting the lexical entries. For example, we could use a notation like this – similar to the
proposal on page 231 except that we will indicate the subject by underlining, so that it can
46

be distinguished from the complements. (Here I also indicate the “roles” of the subject and
object, but this is just for your intuitions and we will not emphasize these roles in this class.)
laughs V[tns]
agent
—
DP
murders V[tns]
agent patient
— —
DP DP
likes V[tns]
experiencer theme
— —
DP DP
pleases V[tns]
theme experiencer
— —
DP DP
So in sum, we have
c-selection: a lexical item can impose category restrictions on its complements.
s-selection: a lexical item can impose semantic restrictions on its arguments.
The lexical entries include these restrictions, so for example, we have already seen that we can
think of the word describes as coming with the following information, which indicates that its
subject argument (indicated by underlining) plays the semantic role of being the agent, while
the complement argument (not underlined) plays the role of the theme:
describes V[+tns]
agent patient
— —
DP DP
Other verbs can have diﬀerent selection requirements:
laughs V[+tns]
agent
—
DP
dives V[+tns]
agent
—
DP
murders V[+tns]
agent patient
— —
DP DP
likes V[+tns]
experiencer theme
— —
DP DP
pleases V[+tns]
theme experiencer
— —
DP DP
47

5.4.2 c-selection of clauses
A sentence has a subject DP and a predicate. We will call sentence-like phrases ‘clauses’ and
apply this term to a large range of constructions that have a subject and predicate – I indicate
some of them here in brackets:
(47) [S Kate defies Petruccio] simple clause
(48) She knows that [S
Kate defies Petruccio] clausal complement of V
She wonders whether [S Kate defies Petruccio]
She prefers for [S
Kate to defy Petruccio] (infinitival clause, pp131,281)
She prefers [S Kate’s defying Petruccio] (‘possessive -ing clause)
(49) The claim that [S Kate defies Petruccio] is true. clausal complement of N
Let’s consider that-clauses first:
Kate defies Petruccio]
What is the complement of the verb knows here? This particular kind of that is called a
complementizer and so a standard idea is that here we have the verb selecting a comple-
mentizer phrase (CP):
S
DP
Pronoun
She
VP
V
knows
CP
C
that
S
DP
Name
Kate
VP
V
defies
DP
Name
Petruccio
(Here I have put in the categories ‘Pronoun’ and ‘Name’, but they are special instances of DP,
so sometimes I will leave these out – They are special instances of DP because they seem not
contain determiners in them.) The important point about this tree is the phrase CP formed
from the C that: this kind of complementizer phrase is selected by knows. We already have
the phrase structure rule that allows this:
CP → C S
Notice that this is perfectly good even without the complementizer that. These two sen-
tences mean exactly the same thing, so it is natural to assume that the structures with and
without the complementizer are exactly the same:3
3With this idea, that a complementizer position can have no pronounced material in it, it becomes possible
to assume that
(1) simple clauses are CPs with silent C.
In fact, this idea has some appealing features that we will notice when we consider questions. As we will see,
this idea provides some positions, positions in the CP, that get filled in questions. This idea is proposed in the
text on page 278. We’ll return to this.
48

S
DP
Pronoun
She
VP
V
knows
CP
C S
DP
Name
Kate
VP
V
defies
DP
Name
Petruccio
Although we have rules for building CPs, and rules that allow Vs to select them,
CP → C S
VP → V CP
it is clear that not all verbs c-select CPs. For example the verb dive allows the modifier from
the cliff, but not a DP or CP complement:
(51) a. He dives from the cliff
b. * He dives the ocean
c. * He dives that Kate defies Petruccio
But verbs like know can select a CP complement, and the C can be that, it can be silent, or it
can be whether:
(52) a. He knows whether Kate defies Petruccio
b. He knows that Kate defies Petruccio
c. He knows Kate defies Petruccio
(For reasons that will be discussed soon, we will not count why or what or which as comple-
mentizers.)
knows V[+tns]
experiencer theme
— —
DP CP
49

5.5 Two additional rules, mentioned before
5.5.1 Case
We have seen various examples of DPs:
(53) Mary, Elizabeth, Bill (proper names)
(54) I, you, he, she, it, we, they (nominative case pronouns)
(55) me, you, him, her, it, us, them (accusative or objective case pronouns)
(56) my, your, his, her, its, our, their (genitive case or possessive pronouns)
(57) an exciting book, the door, the car, the garage (descriptions)
We have already noticed that these DPs have diﬀerent distributions:
(58) *him laughs
(59) *Mary surprises he
Now it is easy to be more precise about how the form of the pronominal DP depends on its
position in the structure:
Case assignment to DPs
a. the subject of a tensed clause is nominative case
b. the complements of a verb are accusative (objective) case
c. the complement of a preposition is accusative (objective) case
5.5.2 Subject-verb agreement
It is also a familiar fact that the highest verb in a sentence is tensed, and present tense forms
agree with their subjects. For example,
(60) She is[+tns] smart!
(61) * She are[+tns] smart!
(62) She is[+tns] describing[prespart] the solution
(63) * She are[+tns] describing[prespart] the solution
(64) She prefers[+tns] for us to solve[-tns] it
(65) * She prefer[+tns] for us to solve[-tns] it
Looking at just the highest verb for the moment, we see that it agrees with the subject. This
is a completely familiar idea:
Subject-verb agreement: A present tense verb must agree with its subject.
50

5.6 Summary
Considering VPs, we went over the distinction between complements and adjuncts (arguments
and modifiers). Usually this distinction is easy – roughly, the the complements name ‘partic-
ipants’ in the event while the modifiers say something about why, where, when, or how. But
the distinction between argument and modifier can be slightly tricky. The complements of a
V are sometimes obligatory, and always limited in number; the complements of an N are never
obligatory, but are still always limited in number. Some tests for V modifiers were presented,
and they are summarized again below.
Finally, in this lecture we saw how we can write down the rules for building the phrases
that we have drawn in these trees. You should be familiar with all the rules in the table on
page 44.
What is emerging now is a picture of the syntax in which the basic elements are lexical items,
which have categories, sometimes requiring complements in addition.
Then the phrases get assembled according to a few rather simple rules, at least so far. Special
attention needs to be given to one issue:
Constituents of VP. This is probably the trickiest part of the syntax so far. The things
traditionally called the subject and the (direct or indirect) objects of the verb are arguments,
typically naming participants in some sense. The object and indirect object are comple-
ments, sisters of the V An argument can be a DP, PP, CP, or other category:4
The king doesn’t give his money to charity very often
The king knows that rich people should pay income taxes, and he doesn’t like them!
He prefers for the wealthy to prosper
He wonders whether his plans for the kingdom will succeed
The modifiers of verbs, on the other hand, can be adverbs or prepositional phrases, and they
are sisters of VP. That is, there is recursion on VP in this pattern.
The queen listens politely
She completely supports her husband
She spoke on Sunday at the rally in the rain
How can we tell arguments and modifiers apart?? We repeat our tests here:
(66) Optionality: arguments are sometimes required, sometimes optional. Modifiers are
always optional.
(67) Iteration: The number of arguments is strictly limited. There can be any number of
modifiers. That is, modifiers can be “iterated” or “repeated.”
a. I sang about love with gusto on Saturday at the auditorium with my band
(68) modifiers can modify coordinated VPs:
a. Robin [VP[VPwrote a book] and [VP sang three songs]] with Sandy.
4The text suggests that the adjective phrase in She is beautiful is not the argument of the verb be, but a
predicate. There is something right about this proposal. Notice for example that beautiful does not have a
θ-role (that is, beautiful is not an agent, patient, theme, experiencer,. . . ). We may come back to these issues
later, but for now we will think of beautiful as a complement, sister of V, even if it is not an argument.
51

(69) modifiers can be left behind in VP-preposing:
a. Robin said she would sing a song, and [sing a song] she did, [with Sandy].
b. * Robin said she would give Mary a book, and [give Mary] she did, [a book].
So a speaker knows some basic things about each word that are relevant to assembling
phrase structure. The phrase structure is assembled to respect all these different things:
possible combinations
(specified by phrase structure rules)
⇓
case requirements ⇒ acceptable phrase ⇐ agreement requirements
⇑
specific requirements of words
(c-selection, s-selection, from lexicon)
[1] Pesetsky, David (1995) Zero Syntax: Experiencers and Cascades. Cambridge, Massachusetts:
MIT Press.
52

Lecture 6 Other syntactic dependencies: movement
This lecture begins material that is introduced in Ch 5 of the text. We will introduce ideas
from that chapter bit-by-bit in the next few lectures.
6.1 One more instance of c-selection: Auxiliary verbs
One particularly interesting case of selection happens with auxiliary verbs. When we consider
sentences with auxiliary verbs, a simple pattern is easy to see:
(1) He might have been eating
(2) He has been eating
(3) He is eat-ing
(4) He eats
(5) He has been eating
(6) He has eat-en
(7) He might be eating
(8) He might have eaten
(9) He might eat
If we put the modal verbs in any other orders, the results are no good:
(10) * He have might been eating
(11) * He might been eating
(12) * He is have eaten
(13) * He has will eat
The regularities can be stated informally as follows:
(14) English auxiliary verbs occur in the order MODAL HAVE BE. So there can be as many
as 3, or as few as 0.
(15) A MODAL (when used as an auxiliary) is followed by a tenseless verb, [-tns]
(16) HAVE (when used as an auxiliary) is followed by a past participle, [pastpart]
(17) Be (when used as an auxiliary) is followed by a present participle, [prespart]
(18) The ﬁrst verb after the subject is always the one showing agreement with the subject
and a tense marking (if any), [+tns]
53

We have already seen a mechanism for telling us what can follow a verb: its complement
list! So these auxiliary verbs can be treated as having special complements: VPs. We can list
the requirements as follows:1
word category complements
will,would,shall,should,may,might,must,can,could V[+tns] VP[-tns]
has,have,had V[+tns] VP[pastpart]
is,are,was,were V[+tns] VP[prespart]
(Notice that the lexical entries on page 47 of these lecture notes listed tensed forms of some
verbs, but now we list also inﬁnitive forms, and present and past participles.)
While it is natural to think of the verb in a simple sentence like he dived – the agent of the
diving – it is not so natural to think of the auxiliary in he is diving as having a subject. For
this reason, it is usually assumed that auxiliary verbs do not have subjects (at least, not in
the way that main verbs do). With that assumption, we can list the complements selected by
each verb as follows (showing underlined subjects only for main verbs, not auxiliary forms).
We also list here the participle forms that do not need to be in the lexicon since they can be
derived by regular morphological processes:
have
having
had
V[-tns]
V[prespart]
V[pastpart]
VP[pastpart]
be
being
been
V[-tns]
V[prespart]
V[pastpart]
VP[prespart]
laugh
laughing
laughed
V[-tns]
V[prespart]
V[pastpart]
—
DP
murder
murdering
murdered
V[-tns]
V[prespart]
V[pastpart]
— —
DP DP
like
liking
liked
V[-tns]
V[prespart]
V[pastpart]
— —
DP DP
1Many of these auxiliary verbs have other uses too, which will require other entries in the lexicon. Consider,
for example:
(1) He willed me his fortune. His mother contested the will. (WILL as main verb, or noun)
(2) They can this beer in Canada. The can ends up in California. (CAN as main verb, or noun)
(3) The might of a grizzly bear is nothing to sneeze at. (MIGHT as noun)
(4) I have hiking boots. (HAVE as main verb)
(5) I am tall. (BE as main verb)
54

With lexical entries like these, we can account for all of the patterns of auxiliary verbs that
we saw in (1-13), but examples like (20) and (27) remain mysterious:
(19) I know [he is describing which problem]? (bold to indicate stress)
(20) I know [which problem John is describing]?
6.2 Wh-questions
We now introduce another kind of dependency: movement. Movement gets discussed in Chap-
ter 4 of the text. (The whole chapter is good reading, but we will focus on pages 224-245.)
We have briefly surveyed a range of clause-types found in English (a similar range is found
in other human languages):
(21) [S Kate defies Petruccio] simple clause
Kate defies Petruccio] V complement clause
(23) The claim that [S
Kate defies Petruccio] is true. N complement clause
(24) She prefers for [S Kate to defy Petruccio] V infinitival complement clause (pp131,281)
(25) She sees [S
Kate defy Petruccio] small clause (p133)
But now let’s consider questions. In the text and in class, we saw that in both of these examples
(repeated from above)
(19) I know [he is describing which problem]? (bold to indicate stress)
(20) I know [which problem John is describing]?
it is natural to regard which problem as the object, the theme of describe. This idea fits with
the argument structure given above, in which this verb takes both a subject and an object.
This idea also explains why it is no good to put an object in that position:
(26) * Which problem is John describing the book?
The only problem is that the expression which problem in (20) is not in the usual object
position – it is in a more “remote,” preposed position.
Similar observations can be made about which students in (27)
(27) Which students are you thinking want to take the exam?
In sentences like these, it is natural to assume that which students is the subject of want. This
explains why we have subject-verb agreement:
(28) * Which student are you thinking want to take the exam?
(29) Which student are you thinking wants to take the exam?
(30) * Which students are you thinking wants to take the exam?
Again, the only puzzle is how does the subject get into the “remote” “preposed” position, at
the front of the sentence.
To explain these facts, the text proposes another kind of dependency:
55

wh-movement (first version): A phrase containing a wh-word can be preposed.
Let’s postpone just for a moment showing any hypotheses about what the tree structure
of these wh-questions are. The text shows a structure on p244, but does not explain it. We
will postpone the question for just a minute while we set the stage for understanding what’s
going on.
There is another puzzle about sentences (20) and (27) which we have not tackled yet!
Namely, what is going on with that auxiliary verb is and are. It is easy to see that the
auxiliary verb in wh-questions like these has to agree with the DP that follows it!
(31) Which book are you buying?
(32) * Which book is you buying?
(33) Which book is she buying?
This looks like subject-verb agreement, but the subject and verb seem to be reversed from their
usual order! To understand this, let’s first be a little bit more explicit about the categories of
auxiliary verbs.
There is a range of wh-constructions which contain sentences:
(34) [Who [S defies Petruccio]]? wh-question (questioning the subject)
(35) She knows [who [S
defies Petruccio]]. V wh-complement clause
(36) The woman [who [S
defies Petruccio]] is Kate. N wh-modifier (“relative clause”)
(37) [Who does [S Kate defy]]? wh-question (questioning the object)
(38) She knows [who [S
Kate defies]] V complement clause
(39) The guy [who [S
Kate defies]] is bad news. N wh-modifier (“relative clause”)
We saw that both CPs with declarative sentences and CPs with questions can be embedded.
(And so the “relative clauses” are CP questions, attached as modifiers in NP – see new modifier
rule on page 58.) There are a number of puzzling things going on here that we should look at
more carefully, but it will take several steps. . .
6.3 Yes/no-questions
First, let’s go to material Chapter 5 “The distribution of verbal forms” (esp.pp257-300, and
some things in this chapter – aux verbs, morphology – which have already been discussed).
When we look over our earlier examples of (1-13), we see that each one allows the first
auxiliary verb to be preposed:
(40) a. He is diving
b. Is he diving?
(41) a. He has dived
b. Has he dived?
(42) a. He has been diving
b. Has he been diving?
(43) a. He might have been diving
56

b. Might he have been diving?
(44) a. He might be diving
b. Might he be diving?
(45) a. He might have dived
b. Might he have dived?
After looking at wh-movement, it is natural to propose some kind of “preposing” in this
case too, except that in this case, the preposing moves just a verb, not a whole verb phrase.
What position does the verb move to? Well, we have seen that sentences can be preceded by
complementizers (C), so one idea is that in these sentences, the auxiliary verb moves to the C
position:
Subject-auxiliary inversion (ﬁrst version): The highest auxiliary in a tensed sentence
can be preposed to C to make a question.
CP
C S
DP
Macbeth
VP
V[+tns]
is
VP[prespart]
V[prespart]
describing
DP
D
the
NP
N
forest ⇒
CP
C
is
S
DP
Macbeth
VP
V[+tns] VP[prespart]
V[prespart]
describing
DP
D
the
NP
N
forest
(Draw an arrow
from V[+tns] to
to make clear
what happened here!)
Notice that we can make these into “echo” questions with stress and intonation:
(46) Macbeth is describing WHAT FOREST?
(47) Is Macbeth describing WHAT FOREST?
Or, with normal question intonation,
(48) What forest is Macbeth describing?
Now we can make sense of the tree (90) on page 244 of the text:
wh-movement (second version): A phrase containing a wh-word can be preposed to at-
tach to CP. (When we do this, we assume that C+S form an intermediate phrase
C’.)
57

CP
C
is
S
DP
Macbeth
VP
V[prespart]
describing
DP
D
what
NP
N
forest ⇒
CP
DP
D
what
NP
N
forest
C’
C
is
S
DP
Macbeth
VP
V[prespart]
describing
DP
6.4 Summary
After class in response to a question, I mentioned that the rules could be simpler if we assume
sentences are CPs (but this brings a few more small changes next week. . . ).
(137) S → DP VP
(124a) DP →



(D) NP
Name
Pronoun



(123) NP → N (PP)
(130) VP → V (DP) (



PP
CP
VP



)
(120) PP → P (DP)
(122) AP → A (PP)
(138) CP → C S
(new!) CP → C’
(new!) C’ → C S
AdvP → Adv
NP → NP PP
(new!) NP → NP CP
VP → AdvP VP
VP → VP PP
AP → AdvP AP
α → α Coord α (for α=D,V,N,A,P,C,Adv,VP,NP,AP,PP,AdvP,TP,CP)
movement rules:
wh-movement: A phrase containing a wh-word can be preposed to attach to CP.
Subj-Aux inversion: A tensed auxiliary Verb can move to attach to C
These lists are incomplete but this is a good start.
Noam Chomsky (1956) originally argued that we need more than phrase structure rules to
describe the patterns in human languages more simply, more “succinctly.” But later math-
58

ematical studies of grammars with movement show that they can describe patterns that no
set of phrase structure rules can describe, no matter how complex the phrase structure rules
are (Peters & Ritchie 1973, Savitch et al. 1987, Kobele 2009).. Deﬁning simple languages of
arithmetic or standard programming languages does not require movement rules, but human
languages do need these (or something with equivalent descriptive power).
[1] Chomsky, Noam (1956) Three models for the description of language. IRE Transactions on
Information Theory IT-2: 113-124.
[2] Kobele, Gregory M. (2009) Without remnant movement, MGs are context-free. Proceedings of
the 11th Mathematics of Language (MOL).
[3] Peters, P. Stanley and Robert W. Ritchie (1973) On the generative power of transformational
grammar. Information Sciences 6: 49-83.
[4] Savitch, Walter J., Emmon Bach, William Marsh, and Gila Safran-Naveh. (1987) The Formal
Complexity of Natural Language. Boston: Reidel.
59

60

Lecture 7 Clauses, tense, and questions
We have observed that simple clauses have tense, but other clauses don’t. We could add the
requirement [+tns] to our sentences, but then we will need a separate account for “infinitival
clauses”. Is there a better way to say what’s going on with +tns? We have also seen that
when a clause is tensed, we usually see the tense only on the first verb:
(1) Marie will[+tns] have been eating cake
(2) * Marie will[+tns] has[+tns] been eating cake
(3) * Marie will[+tns] have is[+tns] eating cake
(4) Marie has[+tns] been eating cake
(5) * Marie has[+tns] is[+tns] eating cake
But this is not always the case:
(6) The desire to succeed[-tns] in business drove[+tns] her on
(7) For Marie to succeed[-tns] is[+tns] all that matters
What’s going on here?
In Chapter 3 “Syntax I”, on pages 160, 181-182, there was a suggestion about this which
may seem rather surprising: the suggestion is that simple sentences are Tense Phrases, TPs:
(8) sentences are TPs
So now we have a couple of different sentence-like units. Sentence-like phrases are often called
clauses. Putting the TP proposal together with our earlier idea about CPs, the idea is that
T precedes the VP and that we should elaborate our tree structures like this:
S
DP
Kate
VP
V
defies
DP
Petruccio becomes
CP
C’
C TP
DP
Kate
T’
T
-s
VP
V
defy
DP
Petruccio
The idea then is that somehow, -s defy gets pronounced as the appropriate, tensed form defies.
How is this possible when they are in the wrong order?
61

How does the tense affix get onto the verb? The simplest idea is that an affix can just hop
onto an adjacent verb, so let’s propose this new movement rule, restricting it to this particular
case:
Affix hopping: A tense suffix can move down to the (non-auxiliary) verb of an adjacent VP.
So affix hopping applies in our example:1
CP
C’
C TP
DP
Kate
T’
T
-s
VP
V
defy
DP
Petruccio affix-hop
−−−−−→
CP
C’
C TP
DP
Kate
T’
T VP
V
defy -s
DP
Petruccio
And this structure is exactly the same when it appears as the complement of a verb, as in
example (22):
CP
C’
C TP
DP
He
T’
T
-s
VP
V
know
CP
C
that
TP
DP
Kate
T’
T
-s
VP
V
defy
DP
Petruccio affix-hop
−−−−−→
CP
C’
C TP
DP
He
T’
T VP
V
know -s
CP
C
that
TP
DP
Kate
T’
T VP
V
defy -s
DP
Petruccio
1In the trees on this page I label “defy -s” and “know -s” simply V, but in class I wrote V[+tns] and included
the original categories V and T too. These morphological details will be untangled if you take more syntax
classes. For now, this simple labelling (or the one in class) is good enough.
62

And of course the structure is exactly the same even when the complementizer position C does
not have that in it:
CP
C’
C TP
DP
He
T’
T
-s
VP
V
know
CP
C TP
DP
Kate
T’
T
-s
VP
V
defy
DP
Petruccio affix-hop
−−−−−→
CP
C’
C TP
DP
He
T’
T VP
V
know -s
CP
C TP
DP
Kate
T’
T VP
V
defy -s
DP
Petruccio
(It is a good exercise to draw arrows in these trees from the T where -s starts, to the position
where it ends up after affix hopping!)
Now let’s turn to a basic question that we neglected earlier:
Question 1: What is the structure of infinitival clauses like the one in (24) on page 55?
With the new structures we are drawing for sentences, we can assume that the structure of
(24) is almost exactly the same:
CP
C’
C TP
DP
She
T’
T
-s
VP
V
prefer
CP
C
for
TP
DP
Kate
T’
T
to
VP
V
defy
DP
Petruccio affix-hop
−−−−−→
CP
C’
C TP
DP
She
T’
T VP
V
prefer -s
CP
C
for
TP
DP
Kate
T’
T
to
VP
V
defy
DP
Petruccio
NOTICE! that affix hopping cannot apply in the embedded sentence because there is no affix!
Instead, we have the word to. Now let’s look at the verb forms that occur in these clauses.
63

7.1 Auxiliaries, Negation and the verb DO
Before looking at verb forms more carefully, it is interesting to make one quick observation
about a respect in which do is unlike the auxiliary verbs (modals, BE, HAVE). We have seen
that auxiliary verbs can invert with the subject to form a yes/no question:
(9) Othello is the one
(10) Is Othello the one?
(11) Othello would be it.
(12) Would Othello be it?
Other verbs cannot form questions in this way:
(13) Othello killed it
(14) * Killed Othello it?
So what about DO? It does not act like the other auxiliaries:
(15) Othello did it
(16) *Did Othello it?
It is also easy to see that when an object is questioned, as in (37), there must be an auxiliary
verb (Modal, HAVE, BE) or else DO, and subject-auxiliary inversion must take place. Chapter
5 observes that a similar thing happens when negation is added to simple clauses:
(17) * [Kate not defies Petruccio]
(18) * [Kate defies not Petruccio]
(19) [Kate does not defy Petruccio]
Why is this? When we look at the structure here, it suggests the simple idea that the not gets
in the way of affix hopping:
CP
C’
C TP
DP
Kate
T’
T
-s
NegP
Neg
not
VP
V
defy
DP
Petruccio
What can save this structure? Putting the “dummy verb” do into the T position so that the
affix can attach to it. We propose this rule (p286):
Do-support: Insert DO to save a stranded suffix in T.
64

So we get:
CP
C’
C TP
DP
Kate
T’
T
-s
NegP
Neg
not
VP
V
defy
DP
Petruccio do-supp
−−−−−→
CP
C’
C TP
DP
Kate
T’
T
do -s
NegP
Neg
not
VP
V
defy
DP
Petruccio
This theory, with affix hopping and do-support, gives us an analysis of Kate does not defy
Petruccio, but it leaves a puzzle about the auxiliary verbs. How are negated sentences with
auxiliaries possible?
(20) She will not defy Petruccio
(21) She has not defied Petruccio
(22) She is not defying Petruccio
(23) * She not has defied Petruccio
(24) She will not have been defying Petruccio
(25) ?? She will have not been defying Petruccio
(26) ?? She will have been not defying Petruccio
(27) ?? She will have been defying not Petruccio
These examples suggest that the way an auxiliary verb gets associated with tense suffixes is
different from the way that main verbs do. How could these auxiliaries get associated with T
properly, even when there is a not. Consider the structure:
CP
C’
C TP
DP
Kate
T’
T
-s
NegP
Neg
not
VP
V
have
VP[pastpart]
V[pastpart]
defied
DP
Petruccio
It is possible to propose a different mechanism to associate HAVE with the suffix in structures
like this:
65

V-to-T head movement: an auxiliary verb can raise attach to T
If we assume that this rule is not blocked by negation, then we get the required contrasts that
we want, with these derivations:
CP
C’
C TP
DP
Kate
T’
T
-s
VP
V
have
VP[pastpart]
V[pastpart]
defied
DP
Petruccio V-to-T
−−−−→
CP
C’
C TP
DP
Kate
T’
T
have -s
VP
V VP[pastpart]
V[pastpart]
defied
DP
Petruccio
CP
C’
C TP
DP
Kate
T’
T
-s
NegP
Neg
not
VP
V
have
VP[pastpart]
V[pastpart]
defied
DP
Petruccio V-to-T
−−−−→
CP
C’
C TP
DP
Kate
T’
T
have -s
NegP
Neg
not
VP
V VP[pastpart]
V[pastpart]
defied
DP
Petruccio
(It’s a good exercise to draw an arrow again to indicate the movement from V to T.)
7.1.1 Subject-Auxiliary inversion again
Now that we have modified our structures (using CP and TP instead of S), and we have put
the tense suffixes under T, it is worth considering Subject-Auxiliary inversion again. It can
now be expressed like this:
T-to-C head movement T can raise to an empty C.
So the steps in deriving Has Kate defied Petruccio are these:
CP
C’
C TP
DP
Kate
T’
T
-s
VP
V
have
VP[pastpart]
V[pastpart]
defied
DP
Petruccio
V-to-T
−−−−−−→
CP
C’
C TP
DP
Kate
T’
T
have -s
VP
V VP[pastpart]
V[pastpart]
defied
DP
Petruccio
T-to-C
−−−−−−→
CP
C’
C
have -s
TP
DP
Kate
T’
T VP
V VP[pastpart]
V[pastpart]
defied
DP
Petruccio
66

Some auxiliaries contain both a verb or modal and the past tense morpheme, and so in these
cases we can either say that a tense suﬃx attaches but then the V+tns is pronounced with
one word. For example:
CP
C’
C TP
DP
Kate
T’
T
tns
VP
V
would
VP[-tns]
V[-tns]
defy
DP
Petruccio
V-to-T
−−−−−−→
CP
C’
C TP
DP
Kate
T’
T
would tns
VP
V VP[-tns]
V[-tns]
defy
DP
Petruccio
T-to-C
−−−−−−→
CP
C’
C
would tns
TP
DP
Kate
T’
T VP
V VP[-tns]
V[-tns]
defy
DP
Petruccio
And when there is no auxiliary verb, then there is no V-to-I raising, but do-support applies,
so we derive the sentence Does Kate defy Petruccio in the following steps:
CP
C’
C TP
DP
Kate
T’
T
-s
VP
V
defy
DP
Petruccio T-to-C
−−−−−−→
CP
C’
C
-s
TP
DP
Kate
T’
T VP
V
defy
DP
Petruccio do-supp
−−−−−−−→
CP
C’
C
do -s
TP
DP
Kate
T’
T VP
V
defy
DP
Petruccio
(For practice, make sure again that you can draw the arrows in the derived trees indicating
all the movements that have been done.)
A tricky detail. Usually main verbs cannot move, as we expect if V-to-T movement does
not apply to them:
(28) * Runs she the marathon?
(29) * Has she a knee injury? (acceptable in some dialects, not standard American)
But BE seems special – it acts like an auxiliary verb even when it is the only verb in the
sentence, as in (31):
(30) Is she running the marathon? (BE as aux verb)
(31) Is she a runner? (BE as main verb)
So we need to assume that, unlike HAVE in (29), when BE is the only verb in the sentence,
as in (31), it is an auxiliary, so that V-to-T, and possibly also T-to-C, can apply.
67

7.2 Ambiguity!
We observed that noun compounds can be interpreted different ways. For example, the phrase
health insurance premium can have two different structures with slightly different interpreta-
tions:
N
N
N
health
N
insurance
N
premium
N
N
health
N
N
insurance
N
premium
[[health insurance] premium [health [insurance premuim]]
a premium for health insurance an insurance premium for your health
(this is the likely one) (this one is peculiar!)
This noun compound with 3 nouns has 2 different trees. If you check, you will see that a
compound with 4 nouns, like air -port terminal security has 5 possible trees. A noun compound
with 5 nouns like air -port terminal security officer has 14 possible trees.
Phrasal modifiers can often be attached in different positions too. For example, in the
sentence I saw the man with a telescope, the PP with a telescope can modify either the man or
the seeing. I think this kind of syntactically ambiguity was mentioned a few times in class
but maybe not in the notes yet.
CP
C’
C TP
DP
I
T’
T VP
VP
V
saw
DP
D
the
NP
N
man
PP
P
with
DP
D
a
NP
N
telescope
CP
C’
C TP
DP
I
T’
T VP
V
saw
DP
D
the
NP
NP
N
man
PP
P
with
DP
D
a
NP
N
telescope
(Really I should draw saw as see+Past so that we can see the affix hopping in these trees
(with arrows!). You can do that for me.) A sentence with more prepositional phrases and
other modifiers can have many possible trees – even hundreds of them. This sentence from a
plaque at the UBC Museum of Anthropology has many possible structures:
The government [of the province] [of British Columbia] has contributed [to the building] [of this
museum] to honour the centenary [of the province’s entry] [into confederation] [in the conviction]
[that it will serve and bring pleasure] [to the people] [of the nation].
This sentence is comprehensible. But as we mentioned in class some ambiguous sentences (like
Buffalo buffalo buffalo Buffalo buffalo) can be hard to understand even when they have simple
structures.
68

7.3 Summary
We have now introduced a bunch of “little” changes to handle tense properly! The main ideas
are (1) Sentences are phrases with tense, TPs. To get T and V together, we have one rule for
main verbs (Affix hop) and one for auxiliary verbs (T-to-C).
(137) S → DP VP
(new!) TP → DP T’
(new!) T’ → T NegP
(new!) T’ → T VP
(new!) NegP → Neg VP
(124a) DP →



(D) NP
Name
Pronoun



(123) NP → N (PP)
(130) VP → V (DP) (



PP
CP
VP



)
(120) PP → P (DP)
(122) AP → A (PP)
(138) CP → C S
(new!) CP → C’
(new!) C’ → C TP
AdvP → Adv
(123’) NP → (A) N (PP)
NP → NP PP
(new!) NP → NP CP
VP → AdvP VP
VP → VP PP
AP → AdvP AP
movement and insertion rules:
Wh-movement: A phrase containing a wh-word can be preposed to attach to CP.
Affix hopping: T affix can move to main V head of its complement VP (not over Neg).
V-to-T head movement: auxiliary verbs can raise to T (possibly over Neg).
T-to-C head movement (subj-aux inversion): T can raise to an empty C.
This list is incomplete, but a very good start. These rules now cover many English structures
(infinitely many, of course, since they are recursive)! Maybe I did not talk about the rule
“NP → NP CP” yet? Modifying an NP with a CP? We can mention that next time.
69

70

Lecture 8 Review: The perspective so far
Last class we introduced a standard analysis of English tense and auxiliaries. Today we review
a range of structures, connecting various ideas from the past few weeks together without adding
any new rules or principles.
8.1 Noun complements
We spent some time talking about verb complements (sisters of V, not recursive) versus verb
modiﬁers (sisters of VP, recursive). For example, in the following sentence, the DP every
human is a complement while in every game is a modiﬁer:
CP
C’
C TP
DP
Watson
T’
T VP
VP
V
V
claim
T
PRES
CP
C’
C
that
TP
DP
he
T’
T VP
VP
V
V
beat
T
PRES
DP
D
every
NP
NP
N
human
PP
P
in
DP
D
every
NP
N
game
Make sure you can correctly draw the ar-
rows from the original position of T to
its position next to claim, and from the
original position of PRES to its position
next to beat
It is natural to assume that the relation between the noun claim and the embedded CP in the
following sentence is also a head-complement relation, even though the embedded CP becomes
71

optional. Still, the embedded CP tells us what was claimed:
Again, make sure you can correctly draw
the arrows that relate the verbs and
tenses! In the following tree, mark 1 aﬃx
hop and 1 V-to-T
CP
C’
C TP
DP
DP
Watson’s
NP
N
claim
CP
C’
C
that
TP
DP
he
T’
T VP
VP
V
V
beat
T
PRES
DP
D
every
NP
NP
N
human
PP
P
in
DP
D
every
NP
N
game
T’
T
V
be
T
PRES
VP
V AP
A
preposterous
The question of whether this is the right idea – whether nouns really have complements the
way verbs do – deserves more attention than we can give it here. But the basic idea is roughly
that when the relation between a N and a phrase is similar to the relation between a related
V and its complement, we may want to say that the N is taking a complement too.
8.2 Wh-questions
Last week we mentioned the idea that wh-phrases can be preposed. Now, if we treat these
phrases as the “subjects” of the CP, just like regular subjects are related to TP, we get the
following simple 3-step analysis of a question like What forest is Macbeth describing?.
72

STEP 1: move the auxiliary verb up to Tense using “V-to-T”,
CP
C’
C TP
DP
Macbeth
T’
T[+tns]
pres
VP
V
be
VP[prespart]
V[prespart]
describing
DP
D
what
NP
N
forest
V-to-T
−−−−→
CP
C’
C TP
DP
Macbeth
T’
T[+tns]
V
be
pres
VP
V VP[prespart]
V[prespart]
describing
DP
D
what
NP
N
forest
Draw arrows for all the movements!
STEP 2: move aux verb + tense to C using “T-to-C”:
CP
C’
C TP
DP
Macbeth
T’
T[+tns]
V
be
pres
VP
V VP[prespart]
V[prespart]
describing
DP
D
what
NP
N
forest T-to-C
−−−−→
CP
C’
C
T[+tns]
V
be
pres
TP
DP
Macbeth
T’
T VP
V VP[prespart]
V[prespart]
describing
DP
D
what
NP
N
forest
STEP 3: move the wh-phrase to attach to CP, using the “wh-question formation” rule:
CP
C’
C
T[+tns]
V
be
pres
TP
DP
Macbeth
T’
T VP
V VP[prespart]
V[prespart]
describing
DP
D
what
NP
N
forest wh-move
−−−−−→
CP
DP
D
what
NP
N
forest
C’
C
T[+tns]
V
be
pres
TP
DP
Macbeth
T’
T VP
V VP[prespart]
V[prespart]
describing
DP
73

8.3 Wh-questions as complements
A wh-question can be a complement of V, as in The salesman wonders what the customer
wants. Here, what the customer wants is a question CP, where what is the direct object of the
verb want. We can capture these facts with this structure:
CP
C’
C TP
DP
D
the
NP
N
salesman
T’
T VP
V
V
wonder
T
PRES
CP
DP
what
C’
C TP
DP
D
the
NP
N
customer
T’
T VP
V
V
want
T
PRES
DP
Draw arrows for 2 aﬃx hops and 1 wh-
movement. CP
C’
C TP
DP
I
T’
T VP
V
V
know
T
PRES
CP
DP
D
which
NP
N
book
C’
C TP
DP
you
T’
T VP
V
V
buy
T
PAST
DP
movement.
74

8.4 Wh-questions as modifiers: relative clauses
In the person who you insulted left, the question who you insulted modifies person. This
question is called a relative clause. Intuitively, the question modifies person to tell us which
person left.
CP
C’
C TP
DP
D
the
NP
NP
N
person
CP
DP
who
C’
C TP
DP
you
T’
T VP
V
V
insult
T
PAST
DP
T’
T VP
V
V
leave
T
PAST
movement.CP
C’
C TP
DP
D
the
NP
NP
N
assignment
CP
DP
which
C’
C TP
DP
you
T’
T VP
VP
V
V
finished
T
PAST
DP
PP
P
on
DP
NP
N
time
T’
T
V
will
T
PRES
VP
V VP
V
V
help
DP
D
your
NP
N
grade
Draw arrows for 1 affix hop, 1 V-to-T, 1
wh-movement.
75

CP
C’
C TP
DP
D
the
NP
NP
N
salesman
CP
DP
who
C’
C TP
DP T’
T VP
V
V
wonders
T
PRES
CP
DP
what
C’
C TP
DP
D
the
NP
N
customer
T’
T VP
V
V
wants
T
PRES
DP
T’
T VP
V
V
leave
T
PAST
Draw arrows for 3 aﬃx hops, 2 wh-
movements.
76

CP
C’
C TP
DP
D
the
NP
NP
N
student
CP
DP
who
C’
C TP
DP T’
T
V
DO T
PAST
NegP
Neg
not
VP
V
like
DP
D
the
NP
N
subject
T’
T VP
V
V
drop
T
PAST
DP
D
the
NP
N
class
Draw arrows for 1 affix hop, 1 wh-
movement, and notice the DO-support.
8.5 Infinitival clauses, very briefly!
With the new rules for tense and clauses, infinitival clauses have the same kind of tree as
tensed clauses. We showed this tree already to illustrate this:
CP
C’
C TP
DP
She
T’
T
-s
VP
V
prefer
CP
C
for
TP
DP
Kate
T’
T
to
VP
V
defy
DP
Petruccio affix-hop
−−−−−→
CP
C’
C TP
DP
She
T’
T VP
V
prefer -s
CP
C
for
TP
DP
Kate
T’
T
to
VP
V
defy
DP
Petruccio
NOTICE! that affix hopping does not apply in the embedded sentence because there is no
affix! Instead, we have the word to. It is not an affix, so it does not move.
(There are other types of infinitival clauses too, but we leave them to your next syntax class.. . )

Syntax review
We list the syntax rules again, with the small revisions we made for T this week.
1. constituency tests: substitution for word level categories, substitution by a pro-
noun, substitution by do or do so, phrasal preposing, coordination, sentence fragments
(phrasal answers to questions)
tests for VP modifiers: optionality, iteration, modification of coordinations, VP
preposing (leaving modifier behind)
2. c-selection: a lexical item can impose category restrictions on its complements.
3. s-selection: a lexical item can impose semantic restrictions on its arguments.
4. Case assignment to DPs
a. the subject of a tensed clause is nominative case
b. the complements of V and P are accusative case
5. Subject-verb agreement: A present tense verb must agree with its subject
6. lexical entries: (with subject arguments underlined)
laugh
laughing
laughed
V[-tns]
V[prespart]
V[pastpart]
agent
DP
execute
executing
executed
V[-tns]
V[prespart]
V[pastpart]
agent patient
DP DP
executed V[passpart]
patient
DP
like
liking
liked
V[-tns]
V[prespart]
V[pastpart]
experiencer theme
DP DP
have
having
had
V[-tns]
V[prespart]
V[pastpart]
VP[pastpart]
be
being
been
V[-tns]
V[prespart]
V[pastpart]
VP[prespart]
or
V[passpart]
78

7. phrase structure rules: sentences are TPs (inside CPs) formed by these
rules
(137) S → DP VP
TP → DP T’
T’ → T NegP
T’ → T VP
NegP → Neg VP
(124a) DP →



(D) NP
Name
Pronoun



(123) NP → N (PP)
(130) VP → V (DP) (



PP
CP
VP



)
(120) PP → P (DP)
(122) AP → A (PP)
(138) CP → C S
CP → C’
C’ → C TP
AdvP → Adv
(123’) NP → (A) N (PP)
NP → NP PP
NP → NP CP
VP → AdvP VP
VP → VP PP
AP → AdvP AP
8. After phrase structure and lexical reqs are met, then movement and inser-
tion rules:
(Tricky detail: English be acts like an auxiliary, even when it’s the main verb.)
Now that we are beginning to get a rough understanding of how English sentences are put
together, at least in some simple cases, a question that might begin to haunt you is: Why do
humans ﬁnd it natural to do things this way? Wouldn’t a much simpler system work just as
well? This is similar to the questions about the “profusion of form” in phonology, mentioned
in Chapter 12 (p.519) of the text.

80

Lecture 9 Semantics: What it all means
What is it to understand a sentence?1
So far we have this picture: we recognize sound
sequences, syllables, morphemes, words, parts of speech, and phrases. But how to these mean
anything?
Uttering a sentence is sometimes compared to making a move in a game. A move in a
game cannot be understood in isolation; it has a certain role with respect to the rest of the
game; and it is partly governed by conventions or rules but some room for creativity is often
allowed. Uttering a sentence has these same properties. Clearly the things we say play an
important role in our interactions; the language is partly governed by conventions but leaves
room for creativity; and understanding a sentence involves recognizing a large collection of
relations between the sentence and other things. Many utterances seem to be related to our
perceptions: if I hold up a yellow pencil for you to see (assuming that you have normal vision)
and I say “That’s a yellow pencil” you will know that I have said something true partly because
you can hear the sounds, syllables, morphemes, words, parts of speech, and phrases. But also
you can perceive the pencil and its color, you know that the word “yellow” names that color,
etc. Giving an account of these connections between the linguistic things and me and my
pencil goes well beyond the bounds of linguistic theory! We don’t expect linguistic theory to
say anything all about me, or about pencils, or about colors.
So if an expression’s having a certain meaning amounts to its having a certain role in a
whole network of activities, the prospects for semantics may seem slim.
But we are saved by the fact that many of the most important relations that a sentence
enters into are purely linguistic. For example, while it is beyond the scope of linguistic theory
to say whether a sentence like “I have a yellow pencil” is true, it is not beyond the scope
of the theory to account for the fact that if “I have a yellow pencil” is true, then so is “I
have a pencil.” This relation is independent of whether either sentence is actually true. And
obviously, someone who does not know this basic relation between these 2 sentences cannot
be said to understand them. This kind of relation is especially important because it gives us
a purely linguistic approach to the important semantic property of being true or not. Let’s
define the relation “entailment” as follows:
(1) Sentence S1 entails sentence S2 just in case in any possible situation where S1 is true,
S2 is also true.
(And in making these judgements, we hold the context fixed, so that a name refers the same
thing in S1 as it does in S2, etc. . . The “speaker” of S1 is assumed to be the same in S2. And
if S1 uses “pencil” to refer to the usual writing implement, then I do not understand it in S2
to refer to a particular kind of brush that artists use for fine detail in paintings.)
It is important that this entailment relation just involves the possible situations in which the
1Here we cover material from Chapter 7 of the text.
81

sentences are true, and does not involve any more complicated considerations about whether
the two sentences are relevant to each other in any other ways. So, for example, all of the
following claims about entailment relations are all correct:2
(2) The sentence Stabler has a yellow pencil entails the sentence Stabler has a pencil.
(3) The sentence Stabler has a pencil does not entail the sentence Stabler has a yellow
pencil.
(4) The sentence Stabler has a pencil entails the sentence Either 5 is a prime number or 5
is not a prime number
(5) The sentence Stabler has a pencil entails the sentence Stabler has a pencil and 5 is an
odd number.
A competent speaker of English may not know whether the sentence Stabler has a yellow pencil
is actually true, but a competent speaker does have a grasp of simple entailment relations. So
really when we say something true, we are telling the audience that many many things are
true: both the sentence actually uttered and also other things that are entailed by it.
It is also no surprise to notice that when someone tells us what a phrase means, or we
look something up in a dictionary, we learn things about entailment relations (at least, to a
first approximation). For example, if you ask what the noun platypus means, and I say it is a
peculiar egg-laying mammal native to Australia, a simple idea is that I have told you:3
(6) The sentence Sophia saw a platypus entails Sophia saw a peculiar egg-laying mammal
native to Australia.
In developing our semantic theory, we will give a lot of attention to the semantic property of
being true or not, and particular attention will be given to the semantic relation of “entail-
ment,” which has the special technical sense that we just defined. The reason for this is not
just that being true is sometimes important, but more that it gives us a way of approaching se-
mantics without saying anything at all about pencils or the many other things our expressions
can refer to.
9.1 Compositional semantics
Gottlob Frege’s idea (discussed in Lecture 1) was that the meaning of a phrase is determined
by the meaning of its parts and by the way those parts are put together. We assume that the
relevant parts, the parts we will interpret, are the phrases and words of syntax and morphology.
A complete semantic theory should provide a list of all the morphemes and their meanings, and
show how the meaning of phrases is composed from the meanings of the parts. In particular,
we want to be able to determine the entailment relations of a sentence using this kind of
analytical strategy. Let’s begin with a simple sentence:
2These examples show that one sentence can entail another even when the sentences are about different
things. It is possible to study a different relation which holds only between sentences that are relevant to
each other, but it turns out that this relation is much more difficult to understand. The study of entailment
has been fruitful in linguistics, and it is the main subject of standard logic. The study of so-called “relevance
logics” is much more difficult and even the fundamentals remain rather obscure. In this class, we will stick to
entailment in this special sense. It is a simpler idea than relevance.
3The “rules” that specify entailments like this are sometimes called “meaning postulates” (Fodor et al,
1975), following a philosophical tradition developed by (Carnap 1956) and many others. There is a long-
standing philosophical controversy about whether a distinction can be drawn between those postulates which
hold in virtue of meaning and those which hold just because of facts about the world (Quine 1951), but we
will ignore this issue for the moment.
82

(7) Sophia laughs
Here it is natural to assume that (in a “normal” context of utterance), the proper name
Sophia refers to something, and the sentence (interpreted literally) asserts that this thing has
the property named by laughs. Let’s use double brackets to represent what expressions refer
to, so for example, [[Socrates]] refers to a person, and we will regard [[laughs]] as referring to
the set of things with the property of being something that laughs. Then we can say:
(8) The sentence [Sophia laughs] is true just in case the person [[Sophia]] is in the set
[[laughs]].
Expressing the matter in this slightly awkward form allows us to notice some general claims
about sentences of this form.
(9) When DP is a proper name, the sentence [DP T’] is true just in case [[DP]] is in [[T’]].
We can also notice that the set of things in the set [[sings beautifully]] is a subset of (is
completely included in) the set [[sings]]. Whenever we have this kind of subset relation, we
have a corresponding entailment relation. Since [[sings beautifully]] is a subset of [[sings]], the
sentence Maria sings beautifully entails Maria sings. This simple relation holds in general:
(10) When DP is a proper name, and whenever we have two verb phrases T’1 and T’2 where
[[T’1]] is a always subset of [[T’2]], then the sentence [DP T’1] entails [DP T’2].
Since the set [[laughs]] is completely included in the set [[either laughs or doesn’t laugh]], the
sentence Sam laughs entails Sam either laughs or doesn’t laugh. This kind of fact is very basic
to our understanding of a language, and we will have more to say about it later.
83

9.2 Determiners and nouns
We might guess that in all sentences, the subject DP names some thing, and the T’ names
a property which that thing has. But this guess would be wrong! It does not work when we
move to even just slightly more complicated sentences. Consider the sentences
(11) No cat laughs
(12) Every student laughs
(13) Most people laugh
(14) Less than 5 teachers laugh
In these sentences, the subjects do not name single objects. Each different determiner makes an
important contribution to the sentence. Gottlob Frege had an idea about this too! He suggests
that these sentences are not about particular objects, rather, the determiners specify a relation
that holds between the concept represented by the subject and the concept represented by the
rest of the sentence, the T’.
Frege found this puzzling at first. He says,
It is true that at first sight the proposition
“All whales are mammals”
seems to be not about concepts but about animals; but if we ask which animal then we are
speaking of, we are unable to point to any one in particular. . . If it be replied that what we are
speaking of is not, indeed, an individual definite object, but nevertheless an indefinite object,
I suspect that “indefinite object” is only another term for concept. . . (Frege 1884, §47)
So what we say is that All whales are mammals tells us something about how the collection of
whales is related to the collection of mammals, namely, the first collection is a subset of the
second.
For the different determiners D in sentence of the form [[D NP] T’], we have rules like the
following:
(15) [No NP T’] is true just in case nothing in [[NP]] is also in [[T’]].
(16) [Every NP T’] is true just in case [[NP]] is a subset of [[T’]].
(17) [The NP T’] is true just in case there is a particular thing (determined according to
context) in [[NP]] that is also in [[T’]].
(18) [Most NP T’] is true just in case the set of things in both [[NP]] and [[T’]] is larger than
the set of things that are in [[NP]] but not in [[T’]].
(19) [Less than 5 NP T’] is true just in case the set of things in both [[NP]] and [[T’]] has
less than 5 things in it.
These rules tell us what each of these determiners mean: each one represents a kind of relation
between [[NP]] and [[T’]].
84

9.3 Adjectives
We have not really provided the basis for determining the semantic properties of infinitely
many sentences yet: we have only considered simple Ds, NPs and T’s. In our syntax, one of
the first things we added which made the language infinite was adjectives. These make the
language infinite, because they are introduced by recursive rules, signifying that they can be
repeated any number of times. So let’s consider how semantic properties of sentences with
adjectives could be figured out, no matter how many adjectives there are.
Let’s start with simple sentences again. Consider the sentences
(20) Every student laughs
(21) Every Transylvanian student laughs
The first sentence is true when the set [[student]] is a subset of the set [[laughs]]. The second
sentence is true when the set of things that are in both [[student]] and [[Transylvanian]] is a
subset of [[T’]]. Adjectives like this are called “intersective”:
(22) An adjective A (and the AP it forms) is intersective if [every [AP NP] T’] means that
the set of things that are in both [[AP]] and [[NP]] is a subset of [[T’]].
The reason for the name “intersective” is obvious: the set of things that are in both [[AP]] and
[[NP]] is the intersection of [[AP]] and [[NP]]. The intersection of [[AP]] and [[NP]] is sometimes
written [[AP]] ∩ [[NP]].
Intersective adjectives can be iterated in a noun phrase, and we know what the result
will mean. A Transylvanian student is something that is in both the sets [[Transylvanian]]
and [[student]]. A female Transylvanian student is something that is in the sets [[female]],
[[Transylvanian]] and [[student]]. A Republican female Transylvanian student is something that
is in the sets [[Republican]], [[female]], [[Transylvanian]] and [[student]]. And so on.
Not all adjectives are intersective. Consider the adjective big. It does not really make sense
to look for a set of things with the property of being big. Consider the sun for example. We
usually think of it as big, and it is certainly a big member of the solar system – the biggest in
fact. But it is a tiny star, not big at all by stellar standards. So is it big or not? The question
does not even make sense. It does not make sense to have a set of big things, because we need
to know, big relative to what? Relative to the planets, relative to the stars, or relative to an
electron? Adjectives like big are sometimes called scalar because they refer to size: big, little,
short, wide, narrow, . . . . These adjectives are non-intersective.
The negative adjectives like fake, bogus, phony, false are also non-intersective. A fake
diamond may be made out of glass. But it is not fake glass. It is real glass but fake diamond.
Similarly non-intersective are the conjectural adjectives like ostensible, alleged, apparent,
possible, likely. An alleged thief may be undeniably a person, and not a thief at all. So the
adjective alleged is not intersective because whether a thing is alleged or not is relative to the
property being alleged. It does not make sense to have a set of objects with the property of
being alleged.
85

9.4 The simple semantics more concisely
The simple ideas about meaning that we have just sketched can be expressed easily with the
tools of set theory, using the following symbols:
x ∈ A x is an element of set A
A ∩ B the intersection of A and B, that is,
the set of things that are in both A and B
A ⊆ B A is a subset (or equal to) B
A − B the result of removing all elements of B from A
|A| the number of things in set A
∅ the empty set
With these symbols we can express (9), (15-19) and (22) more simply, like this:
(9’) [[Name T’]]=true just in case [[Name]]∈[[T’]]
(15’) [[No NP T’]]=true just in case [[NP]]∩[[T’]]=∅
(16’) [[Every NP T’]]=true just in case [[NP]]⊆[[T’]]
(17’) [[The NP T’]]=true just in case something (determined by context)
is in [[NP]]∩[[T’]]
(18’) [[Most NP T’]]=true just in case |[[NP]]∩[[T’]]| > |[[NP]]-[[T’]]|
(19’) [[Less than 5 NP T’]]=true just in case |[[NP]]∩[[T’]]| <5
(22’) [[AP NP]]=[[AP]]∩[[NP]] just in case AP is intersective
In this table, we can see that the truth of each sentence is determined compositionally, by the
meanings of the sentence parts.
Sentences like [two out of three NP T’] say something like: 2/3 of the things that are in
[[NP]] are also in [[T’]]. That is, we can divide [[NP]] into the things that are also in [[T’]],
[[NP]] ∩ [[T ′
]],
and the things that are in [[NP]] but not in in [[T’]],
[[NP]] − [[T ′
]],
and then the claim is that the ﬁrst set has twice as many elements as the second:
(new) [[2 out of 3 NP T’]]=true just in case |[[NP]] ∩ [[T ]]′
| = |[[NP]] − [[T ]]′
| × 2.
(The text gets this one wrong on page 381!)
The basic facts listed above also allow us to understand some entailment relations. For
example, the fact that every student laughs entails every Transylvanian student laughs can be
seen from this simple fact of set theory:
if [[student]]⊆[[laughs]] then ([[Transylvanian]]∩[[student]])⊆[[laughs]].
Here we are not paying much attention to the rather elaborate syntax that we provided for
sentences like this:
86

CP
C TP
DP
D
every
NP
N
student
T’
T VP
V
laugh -s
CP
C TP
DP
D
every
NP
AP
A
Transylvanian
NP
N
student
T’
T VP
V
laugh -s
But of course this structure does matter for the meaning – as we will see shortly!
[1] Carnap, R. (1956) Meaning postulates. In Meaning and Necessity: A study in semantics
and modal logic. Chicago: University of Chicago Press.
[2] Fodor, J.D., Fodor, J.A., and Garrett, M.F. (1975) The psychological unreality of semantic
representations. Linguistic Inquiry 6: 515-531.
[3] Frege, Gottlob (1884) Compound Thoughts. Translated and reprinted in Klemke, ed.,
1968, Essays on Frege. University of Illinois Press.
[4] Quine, Willard van Orman (1951) Two dogmas of empiricism. Reprinted in From a logical
point of view. NY: Harper.
87

88

Lecture 10 How the semantics fits the syntax
We have seen that sentences of the form [[D N] T’] can be regarded as saying that the sets
[[N]] and [[T’]] have the relationship denoted by [[D]]. Here, we notice first that determiners do
not denote just any relation: they are all “conservative.” We define “decreasing” DPs using
similar ideas and notice that “negative polarity items” are allowed when the have decreasing
DP determiners. And finally, we use these ideas again to notice some important restrictions
on the antecedents of pronouns.
10.1 What relations can determiners represent?
All the determiners we have looked at share an important property: they are all “conservative”
in the following sense.1
(1) A determiner D is conservative if:
the following conditions hold:
whenever [D N] is a singular DP in a sentence [D N T’], the sentence [D N T’] entails
and is entailed by [D N is a N that T’].
whenever [D N] is a plural DP in a sentence [D N T’], the sentence
[D N T’] entails and is entailed by [D N are Ns that T’].
So for example, the singular determiner every is conservative because the sentence every student
laughs entails and is entailed by every student is a student that laughs (and similarly for every
other sentence with every). And the plural most is conservative because the sentence most
platypuses sing entails and is entailed by most platypuses are platypuses that sing.
It is easy to make up a determiner that is not conservative in this sense. Let’s make up
the plural determiner aln, so that we can say aln platypuses are ordinary things. Let’s say
that this sentence means that everything that is not a platypus is an ordinary thing. That is
a sensible claim. In general, we can define aln as follows:
(2) [Aln N T’] is true just in case everything that isn’t in [[N]] is in [[T’]].
So it is true to say: aln the people who are in this room are missing this great lecture! That
just means that everything that is not a person in this room is missing this great lecture. The
sentence Aln squares are striped just means that everything that is not a square is striped.
The important point is that the determiner aln is not conservative. We see this by observing
that aln squares are striped does not entail aln squares are squares that are striped. The latter
1The definition is slightly complicated by the fact that some determiners like every can only occur in singular
noun phrases like [every student], while other determiners like most usually require plural noun phrases, like
[most students]. Notice that the definition just uses the verb is for singular noun phrases and the verb are for
the plural noun phrases.
89

sentence, aln squares are squares that are striped, means that everything that is not a square
is a square that is striped – that’s absurd! So the special entailment relation that identiﬁes
conservative determiners does not hold in the case of this made up determiner aln. That is,
aln is not conservative.
The surprising thing is that, at least to a very good approximation, no human language
has determiners that are non-conservative like aln is (Keenan & Stavi 1986):
(3) In all human languages, every determiner is conservative.
Notice that being conservative or not, in this sense, is a semantic property. You cannot tell
whether aln is conservative or not from its phonological or syntactic properties – whether it is
conservative or not depends on what it means.
10.2 Decreasing determiners and NPIs
Certain adverbials like at all, much, one bit and ever seem to require a context that is negative
in some sense. They are called negative polarity items (NPIs). (These were discussed on
pp245-252, and on pp198-224, in the text.)
(4) The fairies do not like witchcraft at all/much/one bit
(5) Nobody likes witchcraft at all/much/one bit
(6) The fairies will never like witchcraft at all/much
(7) No fairies like witchcraft at all/one bit/much
(8) * The fairies like witchcraft at all/much/one bit
(9) * The fairies will like witchcraft at all/much/one bit
These sentences suggest that the NPIs can only occur when there is something negative in
the sentence – a negated VP, a negative determiner, or a negative AdvP. But we see that the
situation is slightly more complicated when we look at a wider range of sentences:
(10) Nobody told the elves [that [the fairies would care at all]]
(11) The news [that [the fairies do not like witchcraft]] surprises the elves.
(12) The news [that [the fairies do not like witchcraft at all]] surprises the elves.
(13) * The news [that [the fairies do not like witchcraft]] surprises the elves at all.
These last sentences all contain negative elements, but the last one is no good! Apparently,
the negative elements have to appear in certain positions in order for NPIs to be allowed.
We now know how to draw the syntactic structures for all of these sentences, so let’s do
that and see if we can ﬁgure out what’s happening.
90

CP
C’
C TP
DP
nobody
T’
T
-s
VP
VP
V
like
DP
NP
N
witchcraft
PP
P
at
DP
D
all
aﬃx-hop
−−−−−→
CP
C’
C TP
DP
nobody
T’
T VP
VP
V
like -s
DP
NP
N
witchcraft
PP
P
at
DP
D
all
The NPI at all is allowed in this example, and in the following: (In the tree on the right, mark the
movement with an arrow!)CP
C’
C TP
DP
D
the
NP
N
fairies
T’
T
tns
NegP
Neg
not
VP
VP
V
like
DP
NP
witchcraft
PP
P
at
DP
D
all do-supp
−−−−−→
CP
C’
C TP
DP
D
the
NP
N
fairies
T’
T
do tns
NegP
Neg
not
VP
VP
V
like
DP
NP
witchcraft
PP
P
at
DP
D
all
CP
C’
C TP
DP
D
the
NP
N
news
CP
C’
C
that
TP
DP
D
the
NP
N
fairies
T’
T
tns
NegP
Neg
not
VP
VP
V
like
DP
NP
N
witchcraft
PP
P
at
DP
D
all
T’
T
-s
VP
V
suprise
DP
D
the
NP
N
elves
do-supp,aﬃx-hop
−−−−−−−−−−−→
CP
C’
C TP
DP
D
the
NP
N
news
CP
C’
C
that
TP
DP
D
the
NP
N
fairies
T’
T
do tns
NegP
Neg
not
VP
VP
V
like
DP
NP
N
witchcraft
PP
P
at
DP
D
all
T’
T VP
V
suprise -s
DP
D
the
NP
N
elves
(In the trees on the right, mark move-
ments (if any) with arrows!)
91

The problem is in examples like this:
* CP
C’
C TP
DP
D
the
NP
N
news
CP
C’
C
that
TP
DP
D
the
NP
N
fairies
T’
T
tns
NegP
Neg
not
VP
VP
V
like
DP
NP
N
witchcraft
T’
T
-s
VP
VP
V
suprise
DP
D
the
NP
N
elves
PP
P
at
DP
D
all
do-supp,affix-hop
−−−−−−−−−−−→
*CP
C’
C TP
DP
D
the
NP
N
news
CP
C’
C
that
TP
DP
D
the
NP
N
fairies
T’
T
do tns
NegP
Neg
not
VP
VP
V
like
DP
NP
N
witchcraft
T’
T VP
VP
V
suprise -s
DP
D
the
NP
N
elves
PP
P
at
DP
D
all
(In the tree on the right, mark the
movement with an arrow!) So what is wrong with this last example? One idea is that the negative element (not in this
example) and the NPI have to be in the same smallest CP, but the earlier example (10) shows
that’s not right. The text considers various ideas, but finally comes to this conclusion:
(14) the parent category of the negated phrase must include the NPI.
The relationship required between the negated phrase and the NPI turns out to be an important
one, so the proposal (14) gets formulated by first naming the relationship:2
(15) A node X c-commands Y if either Y is the sister of X, or Y is a descendant of a sister
of X.
(16) NPI licensing (first version): An NPI must be c-commanded by a negative element.
The negative element X can be a Neg head or a negative DP like “nobody.” (We will have
more to say later about what else counts as a “negative DP”.) The negative polarity items Y
that we have considered so far can be PPs like at all or other adverbials much, one bit.
This story seems pretty good, but there are cases where ever can occur in a sentence that
does not have any explicitly negative item:
(17) a. Less than 3 students ever laugh.
b. * Some student ever laughed.
(18) a. At most 30 students ever laugh.
b. * 30 students ever laugh.
(19) a. No student likes anyone
b. * A student likes anyone
2The text (p. 223) uses this slightly simpler definition: “A node X c-commands Y if the parent of X
includes Y.” This definition agrees with ours on all the cases we will consider here.
92

(20) a. No student saw anything
b. * A student saw anything
(21) a. No student budged an inch
b. * A student budged an inch (with the idiomatic reading)
What determiners can occur in this kind of sentence with the negative polarity items like ever?
We say in (16) that a “negative polarity” YP must be c-commanded by a negative phrase, but
which phrases are “negative phrases” in the relevant sense?? Now we can provide a partial
answer.
We did not get to this on Monday, but the usual answer is this: The determiners D that
can occur in a sentence of the form [D N ever laugh] are the “decreasing” determiners, where
this is again defined in terms of a certain pattern of entailment relationships:
(22) A determiner D is decreasing if whenever we have two verb phrases T’1 and T’2
where [[T’1]] is a always subset of [[T’2]], then [D N T’2] entails [D N T’1].
In these cases we will say that [D N] forms a decreasing DP.
With this definition, we can see that the following determiners form decreasing DPs: no, less
than 3, at most 30, fewer than 6, no more than 2,. . . . These are also the determiners that can
occur with negative polarity items.
The complete story about NPIs goes beyond what we can cover here, and has some suprises,
but this is a good start:
(23) NPI licensing (revised):
An NPI must be c-commanded by a negative phrase XP, and
when XP=DP, then XP=DP is a “negative phrase” only if it’s decreasing.
(To pursue this some more, you will want to take our semantics class, Linguistics 125!)
10.3 Names, pronouns and binding
Many names apply to many different people.3
Consider the sentence:
(24) Every student knows that Ed laughed
Obviously, on various occasions of use, the proper name will refer to different people. The
things that decide which person is referred to are in large part non-linguistic.
Pronouns are a little different. Consider the sentence:
(25) Every student knows that he laughed
Here, we can distinguish two different ways to interpret the pronoun. One possibility is that
the pronoun just refers to someone mentioned earlier or someone pointed to. This is sometimes
called the referential use of the pronoun. When used referentially, a pronoun is similar to
a proper name: its referent is determined by non-linguistic considerations. But the pronoun
has another option that the proper name did not have: it can refer to each person, each thing
3Here we cover material in Chapter 8 of the text, esp. pp.399-406. These sections elaborate on the story
about pronouns and NPIs that we mentioned very briefly a couple of weeks ago in syntax.
93

that every student picks out. The sentence seems to be ambiguous between these two readings.
Let’s use the following notation to refer to the latter reading:
(26) Every studenti knows that hei laughed
This means that it is part of the meaning of the sentence (on one reading) that he refers to
each of the individuals picked out by every student. In this kind of situation, we say that the
pronoun is bound by the antecedent every student.
These ways in which a pronoun picks up its referent is what makes it distinctive. It has
certain options that proper names do not have. The proposition expressed by the “bound”
reading of (26) cannot be expressed without pronouns!
Pronouns have more interpretive options that other DPs, but these options are also re-
stricted in certain ways. Consider the following for example:
(27) * Every studenti knows the joke. Hei laughed.
(28) * Hei knows that every studenti laughed.
(29) * Every studenti and every teacherj knows that hei laughed.
This raises the question, in what cases can a pronoun have its meaning given by the linguistic
context, being “bound” by an antecedent? (27) suggests that he cannot be bound to a DP
every student when that DP in in another sentence. (28) suggests that he cannot be bound to
a DP every student when the pronoun is “higher” in some sense. (29) suggests that he cannot
be bound to a DP every student when that DP not “high enough” in some sense – and this
example might remind you of some of the NPI examples! Let’s try an idea that is similar to
the one we proposed for NPIs.
The examples clearly suggest that whether a pronoun can have a certain antecedent depends
on details of the structural position of the pronoun relative to its antecedent. The required
relation is c-command (mentioned earlier), defined in the following way:
(30) (no change from before:) DP1 c-commands DP2 if either DP1 and DP2 are sisters,
or a sister of DP1 includes DP2.
Binding requirement:
An antecedent must c-command a pronoun in order to bind it.
Checking our earlier examples, notice that this Binding Requirment allows the binding in 26,
and explains why binding is not possible in 27, 28, and 29.(Make sure you can draw the trees for
27, 28, and 29 to show why the bind-
ing is not possible!)
This all seems fine. One other complication is illustrated by examples like the following:
(31) * [Every student]i helped himi.
(32) [Every student]i helped himselfi
(33) * [Every student]i knows that [Jimj helped himselfi]
(34) [Every student]i knows that [Jimj helped himi]
It seems that reflexive pronouns are distinguished from other pronouns in the way they can
get their meaning. Very roughly,
Binding principle A.
A reflexive pronoun must be bound (= must have a c-commanding antecedent) in the
smallest TP that contains it.
94

Binding principle B.
A non-reflexive pronoun cannot be bound in the smallest TP that contains it.
Binding principle C.
Names and phrases like every student cannot be bound.
So we see that pronouns are very sensitive to whether and where they find their antecedents.
Reflecting on this fact, it is natural to worry about whether these requirements apply before or
after movements, or both. We saw in our work on syntax that subcategorization requirements
apply before movements. For example, we check to see which kind of complement a verb
takes before moving the verb or the complement. What about the binding principles? Do
they apply before or after movements? To decide this, we need to consider examples where
the movement could be relevant – movements that change the position of a pronoun or its
antecedent. Consider these examples:
(35) John really likes that story about himself
(36) Which story about himself does John really like?
(37) I know which story about himself John really likes
we have not formulated the rule for the following kind of preposing yet, but it looks
similar:
(38) With himself in mind, he wrote a good story.
(39) * With him in mind, John wrote a good story.
So it appears that, at least in some cases, it is the original position that matters for binding
purposes.
(The whole story about how pronouns find their antecedents is one of the most complicated
parts of syntax, but these first versions of principles A,B,C are a step in the right direction.)
10.4 Summary
We assume that meanings of sentences are determined compositionally. That is, the “meaning”
of a sentence is calculated from the meanings of the phrases and words from which it is built.
What is the meaning of a sentence? We assume that the meaning of a sentence is given by a
network of relationships that the sentence has to other things. Some of the most important
relationships depend on the possible circumstances in which the sentence is true. So we want
to see how the circumstances in which a sentence is true depends on the meanings of its parts.
We began with simple 2 word sentences like Maria sings. In this sentence it seems that the
subject DP serves to name an object, the T’ names a property which a set of things have, and
the sentence says that Maria is in that set. So one aspect of the meaning of a subject DP like
Maria is that (in a given context) it refers to something, and one aspect of the meaning of a T’
is that it names a property which a set of things have. But it is not always true that the subject
of a sentence names something which the T’ tells us about! This is very important! Proper
names typically serve to name something, but DPs of the form [D N] are more complicated.
They do not simply name objects. Instead, they describe a relation the set of things with the
property named by N, [[N]], and the set of things with the property named by the T’, [[T’]].
When an (intersective) adjective A modifies a noun N, we are talking about the things which
are in both of the sets [[A]] and [[N]], that is,
95

[[A]]∩[[N]].
This kind of combination of sets can be repeated any number of times: When two intersective
adjectives A1 and A2 modify a noun N (and this is allowed by our syntax), then we are talking
about the things which are in all of the sets [[A1]], [[A2]] and [[N]], that is,
[[A1]]∩[[A2]]∩[[N]].
In summary, we have so far talked about semantic properties of sentences, proper names,
determiners, adjectives, nouns and T’s.
You should understand these claims (explained more carefully above):
(1) Ds are conservative, in English (and perhaps every human language)
(2) NPIs are OK if they are c-commanded by a decreasing DP
(A) Reflexive pronouns are OK if they are c-commanded by an antecedent DP in same TP
(B) Non-reflexive pronouns cannot be c-commanded by an antecedent DP in same TP
(C) Other DPs (like John or every student) do not have antecedents
where the underlined phrases are defined this way:
• D is conservative = [D N T’] entails [D N is/are N that T’]
• NPI = negative polarity item like ever, budge an inch, some uses of any
• DP is decreasing = [DP VP1] entails [DP VP2] whenever [[V P1]] ⊆ [[V P2]]
• A c-commands B = A is a sister or aunt of B (in the syntactic tree)
• A is the antecedent of B = B gets its reference from A
There are some challenges to (1), but it is at least a very good first approximation to the truth.
For (A-C), there are quite a few tricky issues, but these basic claims are fairly simple and you
should be able to apply them to the syntactic trees that we learned about earlier in the class.
10.5 Exercises not assigned, just for practice
(answers on next page, but don’t look! First, try these problems yourself!)
(1) Which nodes labeled with capital letters c-command node N in the following tree?
A
B
a
D
E F
G
H
b
I
c
K
L
d
M
N
e
O
P
f
Q
R
g
S
h
(2) If the determiner keek is defined as follows, is it conservative? (defend your answer)
[keek NP T’] is true if and only if |[[NP]]|=|[[T’]]|
96

[1] Barwise, Jon and Robin Cooper (1981) Generalized quantifiers and natural language.
Linguistics and Philosophy 4: 159–219.
[2] Keenan, Edward L. and Leonard M. Faltz (1985) Boolean Semantics for Natural Language.
Boston: Reidel.
[3] Keenan, Edward L. and Jonathan Stavi (1986) A semantic characterization of natural
language quantifiers. Linguistics and Philosophy 9: 253–326.
Answers to the exercises on the previous page
(1) capitalized nodes that c-command N (including the direct ancestors):
O, L, G, E, B
(2) With the given definition of keek, the sentence [keek students run] means that the
number of students is the same as the number of runners. But then [keek students are
students who run] means that the number of students is the same as the number of
students who run – something obviously different!
For example, suppose there are two students and neither of them run, and there are
two runners. Then it is true that keek students run since
|[[NP]]| = |[[VP]]| = |[[students]]| = |[[run]]| = 2.
But in this case it is not true that keek students are students who run, since
|[[NP]]| = |[[students]]| = 2
|[[are students who run]]| = 0
97

98

Lecture 11 Phonetics
As discussed in lecture 1, human languages are productive and compositional, like many other
much simpler representational systems. We saw that meanings (semantic values) for complex
expressions can be calculated from the syntactic structures (trees) for those expressions. We
Introduction . . . . . . . . . 99
11.1 Speech sounds . . . . . . . 100
11.2 Transcription . . . . . . . . 102
11.3 Explanations. . . . . . . . . 108
11.4 Articulatory processes 109
11.5 Summary . . . . . . . . . . . . 110
encountered some puzzles about the basic elements of these theories that we did not completely
resolve, but very roughly, it seems that morphemes (or possibly smaller units, like the parts of
idioms) combine to form syntactic expressions. We noticed that sometimes those morphemes
are less than a whole syllable long, but we did not give much attention to how morphemes get
pronounced. That is the focus of the last part of the class.
Obviously, morphemes typically have parts. They are pronounced with sequences of oral or
gestured movements, so the “most basic” elements must be smaller than typical morphemes.
For example, the pronounced forms of the words class and glass seem to have some sounds in
common, sounds that are also shared by many other words. So there might be a shorter list
of basic sounds which can cover all the sounds in all the morphemes of the lexicon.
We do not necessarily want the simplest list, though. What we want is the list of elements
that people, the users of the language, actually take to be basic. So the question is not just
whether there is a list of more basic elements, but whether people actually pay attention to
what those parts are. It is easy to see that we do. This will be completely clear by the end of
this chapter and the next, but just to start with, we can see that speakers of English actually
pay attention to the individual sounds by noticing that the plural of newt is formed by adding
an [s] sound, while the plural of nude is formed by adding a [z] sound. We can see that this is
not accidental in two ways. First, we can see that other “regular” plurals fall into a pattern
with these cases:
pluralize with [z] pluralize with [s]
load loot
mood moat
code coat
mode mote
road root
food foot
Second, if we make up new words that speakers have never used before, we can predict that
these will also fall into the same pattern. For example, if I say that a bad idea should be called
a “crod”, and then I ask you what 2 bad ideas would be called, I can predict that you will say
“2 crods”, pronouncing that plural with a [z] sound. But if I did the same thing with “crot”, I
would predict that you would pluralize with an [s]. This shows that English speakers are not
taking the words as indivisible units, but are noticing the individual sounds in them. We are
not consciously aware of this classiﬁcation of sounds, but it is implicit in the way we use the
language. Our implicit pluralization strategy shows that the list of basic elements of English
(and other spoken languages) are individual sounds like [s] and [z] and [t] and [d].
99

Part (ii) of the first basic idea about the language faces a problem too. It is not true
that we make expressions of the language just by putting words in a sequence. The sequence
“the dog barks” is a good expression of English, something you might say, but the sequence
“barks dog the” is not. The latter sequence is not an intelligible expression of the same sort as
the former one, and so if we are going to describe how the intelligible expressions are formed
from words, the story is going to be more complicated than it is for decimal numerals. Before
working on this problem, let’s go back to the first one and consider what the basic speech
sounds are.
11.1 Speech sounds
If you ask a physicist, sounds are vibrations in the air (i.e. variations in air pressure) produced
in various ways by our vocal apparatus, perceived by the vibration of the ear drum that results.
Like any other sounds, speech can be plotted in a familiar visual form, with the air pressure
on the vertical axis and with time on the horizontal axis. An example is shown in Figure 11.1.
samples[0]
−5000
0
5000
10000
0.00 0.20 0.40 0.60 0.80 1.00 1.20 1.40 1.60
Time: 0.85862sec
−8.050000e+02
D: 0.52600 L: 0.33263 R: 0.85862 (F: 1.90)
“t h e r e ’ s u s u a l l y a v a l v e”
Fig. 11.1: “There’s usually a valve” – deviation from average air pressure vs. time
It is very difficult to recognize the speech sounds relevant to humans in this sort of repre-
sentation, since there are waves of different frequencies and amplitudes caused by the different
aspects of articulation. We get a slightly more readable representation of the same data in a
spectrograph, as in Figure 11.2. Here we plot frequency on the vertical axis, with time on the
horizontal axis, with the magnitude of the departure from average air pressure (amplitude)
indicated by shading, increasing from light gray to dark grey to black to white. The white
bands of high amplitude are called formants. In both graphs, I have put two lines around
the sound of the word usually.
Even in spectrograms, it is difficult to see the linguistically significant distinctions, but one
thing is obvious: word boundaries do not stand out! There is no silence between words, or
any other obvious mark. This is no surprise to anyone who has listened to the speech of a
language they do not know: you cannot tell where one word ends and the next begins. In
fact, this is highly context dependent even when you are fluent in the language, as we see in
(nearly) homophonous pairs of English expressions:
(1) a. The good can decay many ways
b. The good candy came anyways
(2) a. The stuffy nose can lead to problems
b. The stuff he knows can lead to problems
(3) a. Gladly the cross I’d bear
b. Gladly the cross-eyed bear
100

0
1000
2000
3000
4000
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1.1 1.2 1.3 1.4 1.5 1.6
Time: 0.8595 Freq: 4000.00 Value: 72 D: 0.52600 L: 0.33263 R: 0.85862 (F: 1.90)
Fig. 11.2: frequency vs. time, amplitude indicated by shading
(4) a. I scream
b. Ice cream
(5) a. Was he the bear?
b. Wuzzy the bear?
So although we hear individual words, they are difficult to detect in our graphs. We also
hear various things as the same sound, even when they are quite different acoustically. For
one thing, absolute pitch is represented in our graph, and we can hear it, but it makes no
difference to the speech sounds. Also, changing the rate of speech will of course change the
acoustic representation and be perceived, even when the speech sounds are the same.
More interesting mismatches between the acoustic representation and our perception are
found when you look into them more carefully. A typical [i] sound has formants at 280 cps
(cycles per second), 2250 cps and 2890 cps (Ladefoged 1993).1
We can see this sound in the
spectrogram shown above, sliding by quickly as the final vowel of usually, between 0.80 and
0.86 on the horizontal (time) scale. (Check for yourself!) The acoustic properties of vowels
vary from one speaker to another, though. Ladefoged (1957), and many other studies, have
shown that our perception of vowels is actually adjusted to the voice we are hearing, so that
the very sounds we hear as bet in the context of one voice may be perceived as bit in the context
of another voice. The acoustic properties of consonants, on the other hand, vary much more
dramatically even for a given speaker, depending on the context in which they are spoken.
If you cut the first consonant sound out of [pi] (pea) and splice it onto [a] (ah), the resulting
sound is not [pa] but [ka] (Schatz 1945, Liberman 1967). In consonant sounds, we are very
sensitive to the brief changes in formants. Some sounds that you might think would be simple,
are not.
In any case, it is difficult to begin our linguistic theory with the representations of sounds
suggested by work in physics. What we want to do is to classify speech sounds in the way
1These pitches are all fairly high, as is no surprise considering the small size of the parts of the vocal tract
whose resonance gives rise to these formants. For reference: middle C is 221.63 cps; the highest C on a piano
keyboard is 4186 cps. So the main formants of [i] are at frequencies higher than the pitch of the first partial
of any normal speech. The fact that many different frequencies are present at once also explains how singing,
and the intonation we use in questions, etc. is possible: we can vary the fundamental frequency of our acoustic
signals (produced by the vibration of the vocal chords) preserving the basic formant structures of the speech
sounds (produced by the filtering, resonance effects of the shaping of the vocal tract).
101

nasal cavity
alveolar ridge
palate
velar region
lips
(labial region)
teeth (dental region)
tongue body
glottis
tongue root
Fig. 11.3: Places of articulation for consonants
that speakers of the language automatically do in their fluent use of the language.2
As a
first approximation, we begin with a classification of sounds based on how the sounds are
articulated and how they sound to our remarkably sensitive and complex auditory sense. At
some level, this classification should correspond to one based on standard physics, but not in
any simple way!
11.2 Articulation and transcription
The basic structure of the human vocal tract is shown in Figure 11.3. We list the basic sounds
of ‘standard’ American English, classifying them roughly according to the manner of their
production. X-rays of the mouth in action show that our intuitions about tongue positions are
really not very good, and the traditional classification scheme presented here is based largely
on perceived sound quality, i.e. on more or less subtle acoustic properties of the sounds.
Many sounds can be made using these parts of the mouth and throat. vowels can be
formed by vibrating the vocal chords with the tongue in various positions, and consonants can
be produced by stopping or affricating the sound. Writing systems are sometimes classified into
phonetic, syllabic, or morphemic, with English classified as phonemic, Japanese katakana as
syllabic, and Chinese as morphemic, but anyone who knows these writing systems will realize
that the names of these classifications do not match the real complexities of these systems.
It would be wonderful to have an universal alphabet that was truly phonetic, with one
symbol for each sound that is used in any human language. This would make it possible to
pronounce a sentence in any language just by reading it. This is not quite possible, but the
International Phonetic Alphabet comes close. We display it here, and then go through the
2The text says on p 483 “By basic sounds we mean the minimum number of sounds needed to represent each
word in a language differently from all other words, in a way that corresponds to what native speakers think
are the same sounds in different words.” This is not quite right, because two different words can sound exactly
the same: “are” is both a form of the verb be and also a unit of area; “bank” is both a financial institution
and the edge of a river; “nose” is something on your face, but “knows” is a verb. These different words can
be pronounced exactly the same, so we really do not want to represent each word “differently from all other
words.” What we want is to identify the classification of sounds that speakers of the language implicitly use.
102

parts of the alphabet that get used in “standard” American English. (An interactive version
is linked to the web page.)
103

We mark some additional distinctions with these little accents, “diacritics”:
Diacritics
And some further informations about pauses, syllables, etc., “suprasegmentals”, can also be
marked:
Suprasegmentals
Rather than going through everything in these charts, let’s just explore the parts that we need
for rough transcriptions of standard English.
104

Stop, fricative and affricate consonants:
manner voice place
1. [p] spit plosive stop −voice labial
1a. [ph
] pit plosive stop −voice labial
2. [b] bit plosive stop +voice labial
6. [t] stuck plosive stop −voice alveolar
6a. [th
] tick plosive stop −voice alveolar
20. [k] skip plosive stop −voice velar
20a. [kh
] keep plosive stop −voice velar
7. [d] dip plosive stop +voice alveolar
21. [g] get plosive stop +voice velar
[P] but ’n (button) glottal stop −voice glottal
3. [m] moat nasal stop +voice labial
8. [n] note nasal stop +voice alveolar
22. [ ] sing nasal stop +voice velar
4. [f] fit fricative −voice labiodental
5. [v] vat fricative +voice labiodental
10. [T] thick fricative −voice interdental
11. [ ] though fricative +voice interdental
12. [s] sip fricative −voice alveolar
13. [z] zap fricative +voice alveolar
14. [S] ship fricative −voice alveopalatal
15. [Z] azure fricative +voice alveopalatal
24. [h] hat fricative −voice glottal
The stops (plosive and nasal) momentarily block the airflow through the mouth. They are
sometimes called -continuant
The vowels, fricatives, glides, and liquids are continuants, +continuant, because they do not
block airflow through the mouth.
The nasals [n m ] are produced by lowering the velum to force the air through the nose.
The fricatives [s S f z v T h Z] do not quite block airflow, but constrict air passage enough
to generate an audible turbulence.
The affricates [Ù Ã] are represented as sound combinations: very brief stops followed by
fricatives.
105

Liquid and glide consonants:
manner voice place
16. [l] leaf lateral approximant +voice alveolar
16a. [l
"
] or [@l] bottle
syllabic
lateral approximant +voice alveolar
9. [ô] reef (central) approximant +voice retroflex
37. [ô
"
] or [@ô] or [Ä] bird
syllabic
(central) approximant +voice retroflex
[R] butter flap +voice alveolar
19. [j] yet (central) approximant +voice palatal
23. [w] weird (central) approximant +voice labiovelar
The approximants are less restrictive, more vowel-like than the fricatives.
The liquids [ô l] have less constriction than the fricatives.3
Liquids can appear in a syllabic form, sometimes written [@r @l], or alternatively with a
diacritic mark: [ô
"
l
"
].
The glides [j w] involve a rapid transition.
All of the consonants made by raising the blade of the tongue toward the teeth or alveolar
ridge are called coronals. They are the dental, alveolar and alveopalatal stops, fricatives,
affricates, liquids and alveolar nasals: [t d T s z n l r R S Z Ù Ã]. (Not labials, palatals, velars
or glottals.)
Sounds that do not restrict air flow enough to inhibit vibration of the vocal chords are called
sonorants: they are the vowels, glides, liquids and nasals. They are “singable.” Non-sonorants
(plosive stops, fricatives, affricates) are called obstruents.
(6) Every spoken language contrasts vowels with consonants, and sonorant consonants with
obstruents.4
Why would such a thing be so?
3As indicated, we use [ô] for the American “r” sound, following the standard IPA notation, though the text
uses [r]. In IPA, [r] represents a trill “r”. When I am talking and writing about American English, I sometimes
put the r rightside up too.
4In ASL, there is a very similar contrast between the positions assumed in a gesture and the movements
that occur between positions. It is natural to regard the movements as analogous to vowels and the positions
as analogous to consonants. In spoken languages, there are some syllabic consonants, like [r
"
l
"
] in English, but
they never occur adjacent to vowels. In ASL, there are syllabic positions, but never adjacent to movements.
This kind of description of ASL is developed by Perlmutter (1992), for example.
106

high (close)
mid
low (open)
i
I
E
o
u
U
2
æ a
e
@
front back
Fig. 11.4: Tongue position for vowel classiﬁcation
Simple vowels:
tongue body
height
tongue body
backness
lip
rounding
tongue root
tense (+ATR)
or lax (−ATR)
25. [i] beat high front unrounded +ATR
26. [I] fit high front unrounded −ATR
34. [u] boot high back rounded +ATR
33. [U] book high back rounded −ATR
28. [E] let mid front unrounded −ATR
32. [o] road mid back rounded +ATR
31. [O] caught mid back unrounded −ATR
36. [2] shut low back unrounded −ATR
27. [e] ate mid front unrounded +ATR
29. [æ] bat low front unrounded −ATR
30. [a] pot low back unrounded +ATR
35. [@] roses mid back unrounded −ATR
Diphthongs: vowels which change in quality in a single syllable
38. [aI] lies +ATR
39. [aU] crowd +ATR
40. [oI] boy +ATR
The list of relevant speech sounds varies from one dialect of English to another. For me the
vowel [O] in caught is diﬀerent from the vowel [a] in cot, but this distinction is not present for
many English speakers.
107

Tenseness:
The long or tense, +ATR vowels are [i u a o e] and all of the diphthongs [oI aI aU].5
(In elementary school, I was taught that the vowels were [e i aI o u], pronounced in their long
forms here. To this list of long vowels, we have added [a aU].)
The tense/lax distinction is harder to sense by tongue position, though you can feel the
tenseness in the tongue root in the tense/lax pairs like beat/bit, mate/met, shoot/should,
coat/caught.
Probably the best way to remember this feature of vowels is to use the following general-
ization about English:
(7) Monosyllabic words can end in tense vowels, but not in lax vowels.6
OK: bah,
[ba],
see,
[si],
sue,
[su],
say,
[se],
so,
[so],
sigh,
[saI],
now
[naU]
NOT: [sI], [sE], [sæ], [sU]
(8) Syllables with lax vowels other than [U] can end in [ ]; syllables with [U] or tense vowels
do not end in [ ]:
OK: sing,
[sI ],
length,
[lE T],
sang,
[sæ ],
sung,
[s2 ],
song
[so ]
NOT: [sU ], [sa ], [si ], [su ], [se ], [so ], [saU ]
11.3 Explaining the sounds of human languages
Why classify speech sounds into phones in just the way indicated here? One idea is this:
If two speech sounds distinguish two words in any language, they should be represented
as different phones;
Distinctions that are never relevant to distinguishing two words should not be repre-
sented (e.g. absolute volume, absolute pitch).
Notice, for example, that the sounds [t] and [th
] do not distinguish any two words in English.
But [t] and [th
] do distinguish words in Hindi, and so we mark the distinction in our classifi-
cation system. The ideal is that the classification system should be a notation for the sounds
of any spoken human language.
But we have not really stuck to this ideal of marking every distinction of every language in
the phones listed above. For example, [ma] is often used as a word for “mother” in English. But
in Mandarin Chinese, there are the variants [ma] with a high tone vowel meaning “mother,”
[ma] with rising pitch, meaning “hemp,” [ma] with falling pitch, meaning “scold,” and [ma]
5The vowel [o] of standard American English is sometimes classified as lax. In fact, the tenseness of this
vowel varies from one American English dialect to another, as Halle (1977) and others have observed. Eastern
New England dialects have a laxer [o] than most other parts of the country. For any particular speaker of
American English, though, the tenseness of [o] is fairly uniform across lexical items. In contrast, in standard
Southern British English (RP) some words seem to have a rather lax [o] while other words have tenser form.
Ladefoged (1993) suggests that tenseness is a phonological property and not phonetic at all – contrary to what
its name and association with the ATR feature would suggest.
6One of the most common words of English, the, pronounced [ @], is one of the few counterexamples to this
claim. This word the has quite a few special properties.
108

with a lowering and then rising tone meaning “horse.”7
The following notation is sometimes
used to mark these distinctions:
H L H M L H H L
[ma]
[ma] [ma] [ma]
So really, by the same logic that motivates including both [t] and [th
] in our inventory of sounds,
we should include all four of these tonal variations of [a]. Could there be other variations?
Another example is the [k] sound of English. For most English speakers, the [k] in keel
is high and more forward, more central (“scarcely a velar articulation at all”). On the other
hand, the [k] in cool is high and back. The sounds are slightly different, too. Ladefoged (1986,
pp.17ff) report that in some Australian and other languages, such slight variants of [k] are
used to distinguish words. So really they all should have different entries in our list of phones.
Other examples will come up later.8
It is in the context of such observations as these that we should assess the claim one
sometimes hears, that there could be a “completed” IPA chart of all the possible sounds. The
claim is:
(9) the class of phones, the class of possible speech sounds for all human languages, is
finite.
Is this believable? The diversity of languages needs to be weighed against universals such as
(6). And remember: finite sets can be enormous!9
A couple of other interesting points come up when we consider [t] and [th
] in English.
First, the use of one or another of these allophones in English is not random. The first
consonant in top is always [th
]. In almost every context, one or the other of these sounds
is the one used by English speakers, not both. In this case, we say that the sounds have
complementary distribution: where one of the sounds is used, the other is never used.
Pairs like this, different sounds that never distinguish different words in a language, but which
are predictable in context are called allophones. The tonal properties of vowels in English do
not seem to be predictable in quite this way. This provides a reason to regard [t] and [th
] as
allophones of /t/, while the tonal variations of [a] in Mandarin are not allophones in English.
11.4 Looking ahead: articulatory processes
Another interesting issue comes up when we consider English dialects in which the t sound
is almost always pronounced as [R] when it occurs in the middle of a word. So for example,
for these speakers the medial consonant in the word latter has the same sound as the medial
consonant in the word ladder. It is common to transcribe both words with [læR@ô] or [læRô
"
].
But this misses something important: the words do not sound exactly the same because the
7You can hear these variants if you have web access and audio, at:
http://hctv.humnet.ucla.edu/departments/linguistics/VowelsandConsonants/Vowels%20and%20Consonants/chapter2/chinese/recording2.1.html
8In English, a slight lengthening of a simple vowel does not in itself distinguish two words. (Here we do not
mean the changing of a simple vowel into a diphthong, which would be a phonemic change.) But lengthening
simple vowels does make a difference in Serbo-Croatian. Also notice the discussion of latter and ladder below
– there it may look like vowel length is the relevant distinction, but that, we claim, is an illusion.
9This kind of proposal will get discussed later in the text – in §13.1.2 – but we need to introduce some
preliminary ideas before that discussion will make sense.
109

[æ] in ladder is regularly longer than the [æ] in latter. This shortening of a vowel is often
indicated by putting a mark over the vowel:
ladder [læRô
"
]
latter [lˇæRô
"
]
This is OK, except that this representation might lead us to miss an important generalization,
roughly:
(10) Vowels are slightly longer before voiced consonants in English.
We have seen that [d] is voiced, but [t] is not, so the spelling of the words would lead correctly
to the lengthening of the vowel in ladder but not latter. But in the phonetic transcription, we
seem to have lost a distinction which is really there. We classified [R] as voiced, but it seems
that the [R] in [læ:Rô
"
] is really a voiced [d], while the [R] in [læRô
"
] really a voiceless [t]. We will
resolve this problem with our theory of phonology, according to which the [R] in these words
arises from an underlying representation of either [t] or [d] by a process called flapping.
Flapping is one example of an articulatory process in English. Several are common: dis-
similation (carefully distinguishing two adjacent sounds), deletion (dropping a sound, such
as the first vowel in parade), epenthesis (inserting a sound, such as a [p] in the pronuncia-
tion of something as [s2mpTI ]), metathesis (reordering sounds, as in the pronunciation of
spaghetti as [p@skERi]), and progressive and regressive nasalization (spreading the nasal sound
forward or backward, respectively, marked with a tilde), as in [m˜æn]. These will be treated
more carefully within the framework of our phonological theory.
11.5 Summary
Know the phones of standard American English, as listed here and in the book (but on the
exams, sound charts like the ones here will be provided). Understand vowel and diphthong clas-
sifications front/back, high/mid/low, round/unrounded and at least roughly where each vowel
sound is made. Know the consonant classifications stop/fricative/affricate/liquid/nasal/glide,
voiced/unvoiced, and at least roughly where each consonant sound is made. Know what the
voiced flap is. Know which sounds are +coronal and which are +sonorant. Know the diacritics
for stop aspiration (as in [ph
It]), vowel shortening (as in [lˇæRô
"
]), and nasalization (as in [m˜æn]).
[1] Halle, Morris (1977) Tenseness, vowel shift, and the phonology of the back vowels in Modern
English. Linguistic Inquiry 8: 611-625.
[2] Ladefoged, Peter (1993) A Course in Phonetics. Third Edition. NY: Harcourt Brace Javonovich.
[3] Ladefoged, Peter and D.E. Broadbent (1957) Information conveyed by vowels. Journal of the
Acoustical Society of America 29: 98-104.
[4] Ladefoged, Peter and Ian Maddieson (1986) Some of the sounds of the world’s languages. UCLA
Working Papers in Phonetics 64.
[5] Liberman, A., F.S. Cooper, D.P. Shankweiler, and M. Studdert-Kennedy (1967) Perception of
the speech code. Psychological Review 74: 431-461.
[6] Perlmutter, David M. (1992) Sonority and syllable structure in American Sign Language.
Linguistic Inquiry 23: 407-442.
[7] Schatz, C.D. (1954) The role of context in the perception of stops. Language 30: 47-56.
110

Lecture 12 Phonology introduced
Introduction . . . . 111
12.1 Stop aspiriation. 111
12.2 Vowel shortening 113
12.3 Flapping. . . . . . . . 114
12.4 Nasalization . . . . 116
12.5 Summary . . . . . . . 117
It would be natural to assume that the phones we listed in the last chapter are the basic
elements of language, with words and sentences being formed just by putting the phones into
sequences. This turns out not to be right! The sounds [t] and [th
] are different, but in a
certain sense this difference does not matter in English. In English, these two phones are both
variants of the same “underlying” sound, the sound /t/. These ‘underlying’ sounds we will
call phonemes. Perhaps the phonemes (or the features defining them) are really the basic
elements of the language, and their properties can be altered when they are pronounced. This
is the kind of picture that will be developed in this chapter. We will also consider again the
point that not every sequence of phonemes can form a word. In our standard decimal numeral
system, 1111 is a perfectly good number, but not only is [kkkk] not an English word, it is not
even a possible word. Why not? When we look into the matter, we find that the arrangements
of sounds are very restricted and predictable. When we try to state what these restrictions
are, we are led almost right away to a rather complicated picture of what is going on in the
language. And this is just the beginning. You will be surprised.
12.1 Aspirated voiceless stops
We have already observed that the sounds [t] and [th
] cannot occur just anywhere. Similarly
for [k] and [kh
], and also for [p] and [ph
]. One idea is that we simply remember that pit is
pronounced [ph
It] while spit is pronounced [spIt]. This idea does not work, because it does
not account for the fact that if we make up new words, like poev and spoev, we automatically
pronounce them as [ph
ov] and [spov], respectively, even though no one has told us how they
are to be pronounced. Also, if it were just a matter of remembered pronunciations, we would
have no explanation for why (almost) all words beginning with the p sound have the aspirated
form.
An alternative idea is that there is just one basic sound, one phoneme, which we will call
/p/, which gets aspirated automatically in certain contexts and not in others. Similarly for /t/
and /k/. This would explain why we treat new words in the regular way, and why the words
already in the language are pronounced as they are. So what is the context in which stop
consonants get aspirated? It is not just beginnings of words, since the /p/ in upon [2"ph
an]
and the /t/ in retake are also aspirated. One simple idea is:
(1) English voiceless stops are aspirated syllable-initially.
We observed in class that the English voiceless stops form a natural class: we can say that
they are the +stop, −voice sounds. Consequently, the rule (1) can be expressed as follows:1
1The +aspirated feature is sometimes given the name: +spread glottis, because it involves keeping the
glottis open to allow a buildup of pressure behind the stop.
111

(stop aspiration – first try)

+stop
−voice


→
+aspirated
/[syllable
This idea has a couple of problems. In the first place, it makes the wrong prediction in
some words, like these examples that came up in class:
prey ["pô
˚
e]
tray ["tô
˚
e]
clay ["kl
˚
e]
clever ["kl
˚
Evô
"
]
Trevor ["tr
˚
Evô
"
]
The sounds [r,l] are approximants, but more specifically, they are the liquids. So in these
examples, the aspiration associated with the stop seems to just become part of the following
liquid, causing the liquid to sound less voiced – this is indicated by the small circle diacritics.
(The other approximants are [j,w]. Do they have the same property? In my dialect, they
do not seem to swallow the aspiration in the same way: [kh
wIk].) So we can restrict our
description of the aspiration context a little bit more, as follows:
(stop aspiration – second try)

+stop
−voice


→
+aspirated
/[syllable
-liquid
This is better, but it still makes the wrong prediction about words like these
happy ["hæpi]
upper ["2pô
"
]
walking ["wakI ]
For these, we could adjust our rule by adding the requirement that the stop be at the beginning
of a stressed syllable. Notice that the consonants in these examples occur in the unstressed
syllables. So we can make this adjustment to our rule:
(stop aspiration)

+stop
−voice


→
+aspirated
/[stressed
syllable
−liquid
In clear English (which is really more important!), what this rule says is:
A voiceless stop is aspirated if it appears at the beginning of a stressed syllable and is
not followed by a liquid.
Notice that this generalization does not list arbitrary sounds, but rather identifies the sounds
by the properties which seem really relevant.
112

So the basic idea here is that various words with /p/, /t/ or /k/ have predictable variations.
Let’s call the basic sounds /p/, /t/ or /k/ phonemes, indicated by slashes, to contrast them
with the predictable variants, the allophones, [ph
], [th
], [kh
]. The term we used earlier,
phone, applies to all the speech sounds.
On this view, words are associated with sequences of phonemes. These phonemes are then
pronounced in one way or another according to their context. The collection of phonemes of
‘standard’ American English may then be slightly smaller than the classification of phones,
since two different sounds, two different phones may just be alternative pronunciations of the
same underlying phoneme. In fact, phonemes can often be pronounced in many different ways.
Many different phones can represent /t/:
(2) a. [t] as in stop
b. [th
] as in top
c. [R] as in latter
d. [P] as in “button” contracted to “but’n”, or “a’las” for “atlas”
e. [t^] often the t is unrelased in fluent speech, as in “she wen’ home” – it can dissap-
pear completely!
Counting different phonetic sounds as instances of the same phoneme might make you think
that the phonemic classification of sound segments is just “coarser” than the phonetic classi-
fication. But later we will see that the classification of phonemes must also be “finer” than
the classification of phones, in a sense, since in some cases we count one phonetic sound as a
realization of different phonemes. In effect, this is what happens in flapping and various other
processes.
12.2 Vowel shortening
Another predictable variation is found in vowel shortening (p522 in the text). We mentioned
lengthening only very briefly when looking at speech sounds, but now let’s proceed as we did
with aspiration and flapping: we look for clear cases of vowel shortening to discover where it
happens and where it doesn’t. We will indicate shortening by placing a cup-like mark right
over the vowel in the phonetic representation.2
bad [bæd] bat [b˘æt]
Abe [eb] ape [˘ep]
phase [fez] face [f˘es]
leave [liv] leaf [l˘if]
tag [th
æg] tack [th
˘æk]
. . .
It could be that each of these words is just stored in the lexicon with the possibilities
for vowel lengthening indicated. But this is not right, as we can see by observing the same
lengthening in similar non-words:
2Here, we use the mark to indicate shortening [˘e] instead of the mark to indicate lengthening [e:]. Both
marks are introduced in the IPA chart on page 496.
113

gad [gæd] gat [g˘æt]
mabe [meb] mape [m˘ep]
naze [nez] nace [n˘es]
meave [miv] meaf [m˘if]
kag [kh
æg] kack [kh
˘æk]
. . .
So there is some regularity here that is not simply learned on an arbitrary word-by-word basis.
So what are the contexts in which vowels are shorter? Well, as observed in the last chapter [d
b z v g] are all +voice, while [t p s f k] are not. This suggests that vowels are shorter when they
appear before voiceless consonants. The following kind of format is often used for expressing
such a generalization:
(V-shortening – first try)
+vowel
→
+short
/


-voice
+consonant


This rule makes predictions about many cases we have not considered, so it would be good
to check them! And we could also have written this as a lengthening rule – what would that
look like?
12.3 Flapping
Now we can now consider flapping, mentioned only briefly in class. Recall that the flap [R] was
introduced in the phonetics chapter with the word “butter” as a voiced alveolar consonant.
We find this sound in many words (listed here by standard spelling, not phonetically):
ladder latter utter udder
madder matter mutter hottest
soda cider pedal pedant
modify hitter outing edict
jaded edible etiquette outing
It will be good practice to begin with some simple ideas and fix them up again. A first idea
is that /t/ and /d/ are flapped only when they are “medial” consonants, flanked by vowels or
syllabic liquids.
Let’s express this first idea in our rule notation. What class includes the vowels and syllabic
liquids? – Well, they are sonorants, but that’s not what we want, since it includes nasals. But
the vowels and syllabic consonants can be syllables, so let’s call them +syllabic. The next
question is: What features distinguish /t/ and /d/? In fact, these are alveolar plosive stops:
i.e. they are picked out by the features +consonant, +alveolar, −nasal. Putting all of this
together, we can express our idea about flapping this way:
(flapping – first try)



+stop
+alveolar
−nasal



 → R/
+syllabic +syllabic
114

This rule is a good first approximation, but it is not quite right. Looking at the flapped
/t/, since the difference between [t] and [R] is easier to hear than the difference between [d]
and [R], it is easy to find counterexamples to the rule we have formulated. Here are a couple –
these are cases where we have a real [t] between vowels, one that does not get flapped (again,
listing examples by spelling):
proton neutron altitude
aptitude retail attest
mattress retool protest
protect multitude infinitude
attorney attempt attack
attentive attention detest
undertone undertake return
retroactive retire retouch
retort retain retaliate
attract fatigue eternal
material maternal pretested
We can find a similar list of /d/’s that do not get flapped:
radar ado reproduce
redo deduce residue
redraft reduction redouble
redeem podiatrist bedeck
It helps to consider minimal contrasting pairs again, cases as similar as possible, but where
only one member of the pair shows flapping:
rider [ôaIRô
"
]
radar [ôedaô], NOT: [ôeRaô]
atom [æR@m]
atomic [@th
a mIk], NOT: [@RamIk]
proton [pôoth
an], NOT: [pôoRan]
rattle [ôæR@l]
retail [ôith
el]
What is going on here? Well, there seems to be a difference in stress in each pair, which
we could regard as the difference between a stressless syllable and a syllable that receives
secondary stress. As specified on the IPA chart (p.496 of the text), using the vertical mark
above for primary stress and the vertical mark below for secondary stress, then the data is
this:
rider ["ôaIRô
"
] radar ["ôedaô]
atom ["æR@m] atomic [@"th
a mIk]
proton ["pôoth
an]
rattle ["ôæR@l] retail ["ôith
el]
It seems that flapping does not apply if the second vowel has secondary stress (as in proton or
retail), but only when the following vowel is totally unstressed (as in rattle). So we can improve
our flap rule as follows, where we now take care to mean “totally unstressed” by -stress:
115

(flapping)



+stop
+alveolar
−nasal



 → R/
+syllabic


+syllabic
-stress


This handles all of the examples listed above. The examples considered in class were a little
different, but led to a very similar rule.
12.4 Nasalization
We did not get to it in class, but when discussing the speech sounds we mentioned that vowels
sometimes acquire a nasal sound, indicated with a tilde over the vowel:
tEd t˜En nEt
hæt h˜ænd næt
Suppose the generalization is simply this: a vowel that occurs before a nasal consonant becomes
nasalized. Make sure you could represent this generalization with the rule notation we have
been using above. Is this generalization correct? Why would nasalization work this way?
What kinds of evidence could be provided to convince someone who claimed they could not
hear the difference? How could you convince someone who thought nasals could also spread
to a following vowel?
116

12.5 The new picture, and remaining questions
The new story seems intuitive, but it is surprising in a number of ways.
1. We assume that words are listed in in the lexicon not as sequences of phones, but as
sequences of phonemes. – These sounds (defined in terms of their features) are the basic
units in the sound system of a spoken language.
2. The phonemes are defined just as segments of sound with particular properties, particular
features, features which may be altered in certain contexts. So a segment with the features
of a /t/ may be altered to surface as a [R] or as a [P].
3. Rules apply to underlying segments, altering features of specific segments on the basis of
the linguistic context (which sounds are to the left and write, whether there is a word or
syllable boundary, whether there is stress, . . . )
Like every good story, this one leaves us with more puzzles.
Q1. What are the phonemes of English, and how can we defend the idea that something is a
phoneme?
We have introduced a certain strategy, and we will get more practice with it in the next
lectures.
Q2. Considering long [e] and short [˘e], the text says on p. 522:
The very fact that the appearance of [e] and [˘e] is predictable is important: it means
that the difference between the two cannot be used to distinguish words from each
other.
Why not? Is this a matter of logic?3
Or is it an empirical matter?
3For example, it is a matter of logic that if all men are mortal, then it must be the case that: if Socrates is
a man, then he’s mortal. – The opposite assumption is nonsense!
117

118

Lecture 13 Phonemes and rules of variation
At the beginning of the class, we mentioned that although language is flexible and changing
constantly, the way we speak influences our perceptions in sometimes surprising ways. Looking
at sound combinations in the last class, we see some examples of this. First, we see there are
small variations in the phoneme inventories of even “standard” English speakers, and we also
see that distinctions that do not matter in your own dialect are sometimes hard to hear. For
me, there is a clear difference between “caught” /kOt/ and “cot”/kat/, and between “paw”
/pO/ and “pa” /pa/. But for many Californians, this distinction is hard to hear. As you
practice in phonetics and phonology, you will get better at noticing a range of distinctions,
but each new language and dialect can present challenges!
The methods introduced last time are important, so let’s make them explicit here. They
can be applied when you have the relevant data, even when you are not a speaker of the
language. The phoneme basic unit of sound, and phonological rules specify how those
sounds change in context. The changed sounds, the variants which occur in one or another
context, are sometimes called allophones.
So among the phones of American English which we discussed in the first lecture, some
may be phonemes, but others may be variants, “allophones.”
The previous lecture notes list 40 or so phones for ‘standard’ American English, but once
we leave phonetics and start looking at phonological rules, the text proposes a slightly different
list of phonemes – 39 of them:
24 consonants bilabial labiodental alveolar palato- palatal velar glottal
voice dental alveolar
plosives - /p/ /t/ /Ù/ /k/
+ /b/ /d/ /Ã/ /g/
fricatives - /f/ /T/ /s/ /S/ /h/
+ /v/ / / /z/ /Z/
nasals + /m/ /n/ / /
approximant + lateral /l/
+ central /w/ /r/ /j/
11 vowels front central back back
unrounded unrounded unrounded rounded
upper high /i/ /u/
lower high /I/ /U/
upper mid /e/ /@/ /o/
lower mid /E/ /2/
low /æ/ /a/
3 diphthongs
/aI/
/aU/
/oI/
1 syllabic
consonant
/r
"
/ or /Ä/
Among the consonants, notice that there is just one phoneme for each of /t/, /k/, /r/, /l/,
and /w/ even though they have variants, and it is assumed that the flap [R] is a derived form.
119

All of the other consonants in our list of phones correspond to phonemes. (The text writes
the American [r] right side up, so I will too).
The vowel chart lists 11 simple vowels, 3 diphthongs, and 1 syllabic consonant. Comparing
this to the list of phones, we see that the /l
"
/ of the previous chapter is not listed as a phoneme,
nor is /O/. I think the /O/ is left out not because it is derived, but because it is becoming
rather rare; and the text says on p490 that /l
"
/ is left out because it will be treated as /@l/.
So this is a catalog of 39 phonemes altogether, but we have seen that this varies slightly
among English speakers. Some other languages have as few as 11 phonemes (Polynesian,
Pirahã) and some have 100 or more phonemes (e.g. the Khoisan language !Xóõ; some other
languages like the Caucasian language Ubykh have a good number of consonants).
It is commonly (but not universally) assumed that in every dialect of every language, each
word is associated with a sequence of phonemes. This picture of phonemes as the basic units,
the picture expressed in (1), raises the basic question:
(1) How do we identify the phonemes and the variants?
We did this informally when we identified some variants of /t/, /r/ and the vowels last
time, but it is useful to be explicit about the procedures. Let’s state them first, and then go
through some more examples.
On the standard view developed here, the phonemes of a language are the segments of
sound that occur in lexical entries, and this idea is captured with the following procedure:
13.1 Minimal pairs
Identifying different phonemes with minimal pairs
1. Find pairs of words that differ in a single sound. If they differ in meaning, the contrast
is phonemic. The differing sounds are different phonemes.
Complications for this method:
• Sometimes a minimal pair cannot be found, just because of accidental gaps in the lexicon.
The text (pp533,534-535) gives the English example of /Z/ and / /
So, for example, the minimal pair
[bæd] [fæd]
shows that [b] and [f] are (variants of) different underlying phonemes. And the pair
[th
Ip] [lIp]
shows that [th
] and [l] are (variants of) different underlying phonemes.
With sounds like / / and /Z/, it can be hard to find perfect minimal pairs, but we can
come close:
seizure /"siZr
"
/ neither /"ni r
"
/
adhesion /@d"hiZ@n/ heathen /"hi @n/
120

So the previous procedure identifies variants of distinct phonemes, but the possibility of vari-
ants, the possibility that each phoneme can be altered according to its phonological context,
makes determining the actual catalog of phonemes of a language rather abstract, and so we
need a second procedure:
13.2 Phonological rules
Identifying phonemes and phonological rules
1. Identify the environments in which each sound occurs.
(It can happen that the distribution is complex, but we can begin by assuming that
adjacent sounds and boundaries are most likely to be relevant.)
2. Identify collections of sounds that never appear in the same environment:
sounds in complementary distribution.a
(We are especially interested when the
complementary sounds are related, sharing many features.)
3. If the characterization of these collections of sounds and their environments involve lists
of sounds, see whether the elements of each list fall into natural classes, so that they can
be identified by their features.
4. For each such collection, consider the hypothesis
H: the element of the collection that occurs in the widest range of environments is the
phoneme, and the other forms are derived from the phoneme by phonological rules.
Complications for this method:
• Sometimes phonological rules are optional, so the related forms will not be in a perfect
complementary distribution.
• Sometimes two sounds have complementary distributions not because one is derived
from another, but because they occur in different places for other reasons, or because
of an accidental gap in the lexicon.
The text (pp548-549) gives the English example of /h/ and / /. (See also p551 on / /)
aIn practice we sometimes encounter sounds that are almost always in different environments. For
example, the [t]/[th] contrast is nearly complementary except perhaps at the ends of words, where there is
a lot of free variation. To predictive conditioned allomorphy, it is enough to have some contexts only one
variant occurs and some where only the other occurs.
So for example, one of the first sound changes mentioned in the text is English n dentalization:
no [no] tenth [tEn”T]
annoy [@"noI] month [m2n”T]
onion ["2nj@n] panther ["pæn”Tr
"
]
This list suggests that the /n/ is dentalized just when it precedes /T/, but this list is too
short. We should check a range of data, with particular attention to sounds that are similar to
/T/: other alveolars, other fricatives. Let’s go through the steps that pin down this variation
and define it:
121

1. Identify the environments for the sounds [n n”]
n n”
o tE T
@ oI m2 T
2 j@n pæ Tr
"
2. In this data, the environments for [n] and [n”] are completely different. The [n] and [n”] are
in perfect complementary distribution, even just considering only the immediately following
sounds.
3. The sounds inolved here are very specific, and so we do not need arbitrary-looking lists
to describe what’s happening. Reflecting on how these sounds are made, it seems like an
natural rule since the dental and alveolar gestures are similar.
4. We propose the hypothesis that /n/ is a phoneme, and that [n”] is an allophone derived by
the following rule:
(n dentalization)
n →
+dental
/ T
As discussed in class, it could be that this change takes place in more contexts than our
rule says (maybe not just before Tbut before any interdental fricative?), but so far, with the
data shown above, in these notes, we see the change only in this context. So in practice,
after formulating such a hypothesis, and making it as general as possible, we collect more
data to make sure the hypothesis is right – In effect, we loop through steps 1-4 a number
of times until we are sure we have it right.
The other examples we considered last time – stop aspiration, flapping – were more com-
plicated, but the method was the same.
Chapter 12 is mainly devoted to presenting examples of this procedure, which is is sum-
marized at the bottom of p.539. 17 or so different phonological rules are discussed, some at
great length:
(p555) English t aspiration (class formulation slightly different)
(p522) English vowel shortening
(pp530,555,567) English flapping
(p527) English l devoicing
(p527) English l dentalization
(p527) English l velarization
(p545) English vowel nasalization
(p550) English alveolar place enforcement
(p552) English (optional) ae diphthongization
(p555) English preglottalization
(p564) English post-nasal t-deletion
(p566) English /aI/ raising
(p539) Maasai /k/ spirantization
(p539) Maasai post-nasal voicing
(p531) Spanish /d/ spirantization
(p559) Choctaw rhythmic lengthening
(p561) Korean stop nasalization
The important point is not to memorize this list of rules. The important thing is
to know how to use the procedures to find such rules, and what they signify.
One important thing to notice is that these rules are not necessary, so they must be learned.
Other languages do things differently.
122

13.3 Ordering the rules
There is one more wrinkle to consider: what happens when more than one rule can apply?
What we say about this matters! We can see that this matters by considering a dialect that
has the following pronunciations:
phoneme sequence phone sequence
ladder /lædr
"
/ [læRr
"
]
latter /lætr
"
/ [l˘æRr
"
]
We will get the wrong result for the word latter if we apply first flapping and then short-
ening:
/lætr
"
/
⇓ flapping
[læRr
"
]
vowel shortening does not apply (because R is voiced)
We could avoid this result if we insisted that vowel shortening applies before flapping.
The data we are trying to model here might not be trusted though, because, in most
common dialects of American English, the difference between the pronunciations of latter and
ladder is subtle. But there is a dialect of English which provides a more audible distinction
that can be used to explore these issues. In this dialect (mentioned in the text on pp566-570),
we have a diphthong [2I] heard in words like the following (more data in the text):
write [r2It] ride [raId]
tripe [tr
˚
2Ip] tribe [tr
˚
aIb]
rice [r2Is] rise [raIz]
sight [s2It] side [saId]
This difference can be seen in the spectrogram (this one from Moreton & Thomas):
Now, let’s use procedure 2 to see what’s going on with the sounds [2I] and [aI]:
123

1. Identify the environments for the sounds [2I aI]:
aI 2I
r t r d
tr
˚
p tr
˚
b
r s r z
s t s d
2. In this data, the environments for [aI] and [2I] are completely different – these sounds are
in complementary distribution
3. Looking at the consonants immediately following the vowel, the sounds [t p s] are -voice,
and the sounds [d b z] are +voice.
4. We can propose the hypothesis that /aI/ is a phoneme, and that [2I] is an allophone derived
by the following rule:
(aI-raising)
aI→ 2I/


+consonant
−voice


With this idea, we can look at what happens with flapping in this dialect, and what we find
is this very audible difference between some forms in which both flapping and aI-raising can
apply (in my dialect, this is much clearer than the latter/ladder case mentioned above and in
class):
phonemes actual phones
writer /raItr
"
/ [r2IRr
"
]
rider /raIdr
"
/ [raIRr
"
]
We get the wrong result for the word writer if flapping applies first:
/raItr
"
/
⇓ flapping
[raIRr
"
]
aI-raising cannot apply because R is voiced
We can avoid this result by insisting that aI-raising applies before flapping. So the point of this
section is: when more than one rule can apply, we need to decide which has priority. One way
to do this specifies an order in which the rules apply.
Notice how this kind of proposal complicates our picture of phonological processes. In fact,
the presentation here and in the text plays a kind of trick. When listing the environments for
the sounds [aI] and [2I], we did not include the writer/rider pair,
[r2IRr
"
] [raIRr
"
]
If we had included it, we would have noticed that [aI] and [2I] are not in complementary
distribution. So what we really did is to set this last case aside as exceptional at first, and
then explain it by proposing the raising rule.
In dialects with the vowel difference between “ladder” and “latter”, yielding respectively
124

ladder /lædr
"
/ [læRr
"
]
latter /lætr
"
/ [l˘æRr
"
]
it could appear that there is a phonemic contrast between [æ] and [˘æ], but now we see there is
the alternative option of saying that shortening occurs before flapping. (If flapping occurred
first, then since the flap is voiced, we would hear the long vowel in both cases.)
Ordering the rules also increases the complexity of our account considerably, and so some
recent work in the field explores reformulations of the theory that avoids this. You will hear
much more of this if you take more phonology.
13.4 Phonology and morphology
In the text, section 12.10 also observes that adding a prefix or suffix can change the relevant
environment for phonemes in ways that affects pronunciation. For example, adding “-able”
/@b@r/ to “note” /not/ triggers flapping. This section of Chapter 12 talks about morphology,
the study of word formation. We will discuss morphology soon, but this section is understand-
able using just the familiar understanding of words and suffixes.
Using " to mark primary stress,
note "not
notable "noR@b@l or "noR@bl
"
notation no"teS@n
We get flapping not only across stem-suffix boundaries but across word boundaries:
not a mistake "noR@mI"stek
Do we get shortening of [e] across suffix or word boundaries? Do we get dentalization of n
across suffix or word boundaries?
13.5 Phonologies vary
There is nothing necessary about the English phonological rules we have considered (n-
dentalization, vowel shortening, flapping,. . . ). Other languages can have different treatments
even of the same sounds. In Bengali, the n/n” sounds are not in complementary distribution;
on the contrary, for Bengalis, this distinction is phonemic and easy to hear. In English the /t
sounds are not in complementary distribution and are phonemic, but the d/Rsounds do have
complementary distributions and are not phonemic. In Spanish we find the opposite situation
(as discussed in the text on pp.530-531).
[pita] means ‘century plant’, while [piRa] means ’funeral pyre’.
The occurrence of d/ , on the other hand, is governed by a rule like this:
(Spanish spriantization)
d → /[+vowel]
(The conversion of stops to fricatives is often called “spirantization.”)
125

13.6 Summary
You do not need to memorize these rules, but you should be able to understand them, and
more importantly, follow the steps to their formulation:
1. the minimal pair procedure for identifying (variants of) different phonemes
2. the procedure for identifying phonemes and phonological rules
The procedures are quite simple in outline. If someone else is available to provide the relevant
data for a language you don’t know, these methods can be applied. But in real applications,
there are often complexities, some of which were mentioned in this lecture.
Even the first few steps taken in this class provides a surprising picture of how language
works. We have split the notion of “basic speech sound” that was introduced at the beginning
into three categories, roughly described this way: the basic “underlying” phonemes, the
variant phones, and of course there is a great deal of unconditioned variation by which
we mean variation in the speech sounds that is not predictable just from the linguistic context
(but rather by the size of the speaker and other non-linguistic factors). Know why we need to
say “roughly” there, and notice that the list of Standard English phones that we began with
is much too small – really there is a great deal of regular variation in speec sounds, and we
did not list the sounds [n”, r
˚
] or many others. But when we list the phonemes, we have a much
smaller collection – 40 or so. The allophonic variations can then be derived from the phonemes.
When you think of a word or phrase to say, you have a phoneme sequence like /raItr
"
/ in mind.
Then, rules apply, with some having priority over others, to make some adjustments in how
the phonemes sound when they are pronounced, so that the result might be [r2IRr
"
]:
phonemes: /raItr
"
/
phones: [r2IRr
"
]
aI-raising
flapping
126

[1] Moreton, Elliott and Erik R. Thomas (2007) Origins of Canadian Raising in voiceless-coda eﬀects:
a case study in phonologization. Laboratory Phonology 9: 37-64.
127

128

Lecture 14 Syllables and stress
So far we have seen that there may be fewer speech sounds than it seems at first, because
a single sound can have various predicable variants. The regularity we call ‘voiceless stop
aspiration’ explains why English has words like [ph
I ] but not [pI ]. This rule, and the other
rules we have considered to describe variations, seems to be related to the manner in which
these sounds are produced. The first phonology chapter in the text begins with a warning
of three complexities in phonology (p.519): (i) there is quite a lot of variation that seems to
happen “mostly for phonetic reasons;” but (ii) there are other complex distributional facts;
and (iii) sound distributions are also influenced by word-formation (morphology) and phrase-
formation (syntax). The phonology considered so far seems to fall under (i), and now we can
see that there has to be much more to phonology than that.
Returning to the example of decimal representations of numbers, we have seen that certain
sequences like [pI ] do not occur because stop aspiration will apply, but we have not considered
why there is no English word [kkk]. Does voiceless stop aspiration apply? That is, is the first k
in this sequence in a stressed syllable and not followed by a liquid? To answer this question we
need to know more about syllables and stress – nothing in this sequence looks like a syllable!
The relevance of syllables to explaining the distribution of sounds generally is easy to see.
Consider the following facts for example. English allows the word
[pI k] but not *[pInk]
[drI k] but not *[drInk]
And there is some regular pattern here that gets projected onto new words we might make up
too, since
[pE k] is possible, but not *[pEnk],
*[p2mk] is extremely odd,
*[pimk] is extremely odd,. . .
The relevance of syllables to these facts is obvious because we have other words with the wierd
sequences [nk], [mk]:
[Enkod] [Enk2mp@s] [p2mkIn]
It looks like the syllable boundaries, sometimes marked with a period, are relevant here:
[En.kod] [En.k2m.p@s] [p2m.kIn]
Chapter 13 in the text addresses these things, and we turn to them now.1
1All of chapter 13 is good reading, and provides useful background, but the crucial parts are: pp.587-593
129

14.1 Features
Before looking at syllables, the text ﬁrst observes that the phonological rules of the previous
chapters have not really taken the phonemes to be basic: rather, the phonemes have various
features in common, and the phonological rules pay attention to those features. For example,
the feature [+voice] is something that English speakers pay attention to! So the smallest unit
of analysis is really not the phoneme, but the phonological feature.
The text provides further examples to support the idea that referring to features provides
a way to describe how the phonology of a language works, and how phonologies of diﬀerent
language are similar.
14.2 Syllables
The idea that one of the natural units of speech is a syllable is familiar from traditional
grammars and dictionary entries, and we have already referred to syllables in trying to for-
mulate our phonological rules precisely. It is traditionally assumed that a syllable is formed
from zero or more consonants, followed by a vowel, and ending with another, usually shorter,
sequence of zero or more consonants.2
These three parts of a syllable are called the onset
(O), the nucleus (N) and the coda (C), respectively, with the nucleus as the only obligatory
part, and with the tree structure:
syllable σ
O
p l
R
N
æ
C
n
Here the syllable “σ” is called the root node of the tree – so the root is upside-down, the
way family trees often are. In this upside-down tree, the root has two parts, the onset and the
rime. In analogy with family trees, we call these two parts daughter nodes of the root. The
right daughter is the rime, which is in turn the mother of two more daughters: the nucleus
and the coda. In analogy with a real tree, the nodes that are furthest from the root, those
along the bottom of the tree, are sometimes called leaves, as in syntactic and morphological
trees.
Why assume that the elements of the syllable group in this way, as [onset [nucleus coda]]
rather than as [[onset nucleus] coda]? Well, one kind of argument comes from the fact that it
is quite easy to divide syllables at the onset-rime boundary. Not only is this done in rhyming
poetry, but also in language games like Pig latin. We see the same thing in “Yinglish” expres-
sions like fancy-shmancy, road-shmoad. More importantly, there are fundamental restrictions
on sound sequences which hold syllable-internally, as suggested by the *[p2mk], [p2m.kIn]
contrasts mentioned above, and discussed in more detail in the next section.
on syllables, and 597-602 on stress. – These are two loose ends from the previous lectures that we need to take
care of.
2Some prominent approaches to phonology have tried to do without syllables altogether. Among those who
accept syllables, it is a matter of controversy whether ASL has anything corresponding to a syllable structures
– perhaps it could if vowels were equated with movements, and consonants with held positions. . .
130

There are other restrictions on the structure of the English syllable. Consider the possible
onsets:
(1) Any single consonant phoneme is a possible onset, except , and maybe Z.3
(Remember that P is not counted as a phoneme here.)
(2) Only certain 2-consonant onsets are possible.
Since there are 24 consonants in our list of English phonemes, that means there are
242
=576 diﬀerent pairs of consonants. But the ones that occur in common English
words are just those given by the +’s in this table (cf. the discussion in the text on
p.591): Some of these onsets are rare! Try writ-
ing the phonetic transcriptions for the
words: equal, cute, guava, jaguar, reg-
ular, beauty, tubular. In class we also
discussed: pure, huge, muse, music
w j ô l m n p t k
p + + +
t + +
k + + + +
b + + +
d + +
g + + + +
f + + +
T + +
S +
s + + + + + + +
This chart misses words with unusual sounds (borrowings from other languages, etc.).
For example, sphere begins with the unusual onset [sf]. Clearly, [s] has special proper-
ties!
Notice that less than half of the consonants ever begin a complex onsets. Never Ù, Ã,
v, , Z, m, n, , l, r, w, j.
(3) The number of diﬀerent 3-consonant sequences is 243
=13,824. But in onsets, there are
even fewer 3-consonant possibilities than there were 2-consonant possibilities!! I count
just these 9: Write, for example, the phonetic tran-
scriptions for the words: pewter, spew,
splash, fume, ambiguity, Gwen, regular,
fume, thwart.
w j ô l m n
sp + + +
st + +
sk + + + +
See if you can think of any I missed. Again we see dramatically the special properties
of [s].
(4) (Certain other onsets appear in words borrowed from other languages.)
Why are the possible onsets and codas so restricted? We will return to this question.
A simple strategy for specifying the structure of most (but not all!) English syllables is
this: First, each syllabic phone (each vowel and syllabic consonant) is a nucleus. Second, we
have the preferences (p592 in the text):
(a) prefer syllables without onsets or codas (“open syllables”),
3Maybe we should count English as allowing an initial Z for names like Dr. Zhivago. – It is hard to draw a
sharp line between borrowings from other languages and sound sequences that naturally occur in English.
131

but if there are consonants around, the preference is to put them into the onsets:
(b) prefer syllables with onsets,
as long as this does not yield an onset that the language disallows.
When consonants occur between two vowels, then, we typically prefer to associate the con-
sonants with the onset of the second syllable. In other words, each onset should include the
longest possible sequence of consonants preceding a nucleus. Finally, any remaining conso-
nants must be codas of the preceding nuclei. For obvious reasons, this idea is sometimes called
the “onsets before codas” rule; what it amounts to is: “maximize onsets.”
So, for example, the word construct
/k2nstô2kt/
gets parsed into two syllables (shown by the dot in the last step) this way:
k2nstô2kt sequence to syllabify
k [N
2] nstô[N
2] kt identify nuclei
[O
k] [N
2] n[O
stô][N
2] kt maximize onsets
[O k] [N 2][C n].[O stô][N 2][C kt] other consonants in codas
This last line shows with brackets the same thing that can be drawn with the tree:
word
σ
O
k
R
N
2
C
n
σ
O
s t ô
R
N
2
C
k t
This rule works properly for many words (try matron, atlas, enigma), but it does not seem
to provide quite the right account of words like apple or gummy or happy. The ﬁrst syllable
of apple is stressed, and it sounds like it should include the consonant. Cases like these are
called ambisyllabic: a consonant is ambisyllabic if it is part of a (permissible) onset but
immediately follows a stressed lax (-ATR) vowel. For the word happy, the text presents a
structure on page 588 in which the [p] sound is both the coda of the ﬁrst syllable and the onset
of the second one – there’s just one [p] sound but it’s ambisyllabic:4
word
σ
O
h
R
N
æ
C
σ
O
p
R
N
i
Notice that this is not a tree! It’s not a tree because two of its branches “grow back together”
– the /p/ has two mothers! That’s not the way trees work! Unfortunately, that seems to be
what happens with ambisyllabic consonants.
4The tree for the ambisyllabic /hæpi/ in the text on p588 does not show the onsets, nuclei and codas,
because these are not introduced until the next page – p589.
132

14.3 Syllables 1: feature agreement
We introduced syllables using the *[p2mk], [p2m.kIn] contrast. That is, across syllable bound-
aries we find certain consonant combinations that seem to be impossible syllable-internally.
Let’s consider these more carefully.
The text suggests on p588: “in English, a nasal followed by a non-coronal stop (p,b,k,g)
. . . is obligatorily homorganic with the stop when the two are in the same syllable.”
Remembering that the non-coronal stops = labial stops + velar stops. And the nasals =
labial m + velar + alveolar n. So considering all 12 possible non-coronals+nasal combina-
tions, the generalization above tells us that 8 are impossible syllable-internally:
mp mb *mk *mg
* p * b k g
*np *nb *nk *ng
This explains the *[p2mk], [p2m.kIn] contrast. The latter allows /m/ before /k/ because
a syllable boundary intervenes. Similarly for “drainpipe”, “gunpoint”, “unpronounceable”,
“incredible”, “ingrown”, and many others.
The text also suggests (p588) “A parallel observation .. when [sequences of obstruents
(plosive stops, fricatives)] occur in the same syllable, the entire cluster must have the same
voicing value.”
For example, we have [k2bz] and [k2ps] but neither *[k2pz] nor *[k2bs]. But across a
syllable boundary we find [b] next to [s] in “absurd” [@b.sr
"
d], and “Hudson” [h2d.s@n], for
example. We also have words like these:
[glImpst] glimpsed
[tEmpts] tempts
[EnstI kts] instincts
Here the nasals are voiced, and occur in consonant clusters that include voiceless segments.
So consonant clusters containing obstruents need not always have the same voicing value, only
actually adjacent obstruents. But there are also words like this,
[TaUz@ndTs] thousandths
However, maybe this transcription, with [dT], is not accurate; when native American speakers
pronounce this, does it really have a [d] or [R] in it? Maybe not.
A more restricted version of this idea is sometimes considered:5
The Voice Agreement Principle: Obstruent sequences at the end of an English word
cannot differ with respect to voicing.
Another possibly related question is: why do we have
skits [skIts] trailed [treld] but never *[szIts] *[tdeld]?
5This principle and the next one are from the text, pp.612ff, but this is from the the next chapter –
not assigned! Here it is enough to notice just that various versions of these simple ideas are being explored in
our attempt to get the facts exactly right.
133

The Not-Too-Similar Principle: Obstruent sequences cannot differ only in voicing.
These and other regularities should be explored more carefully, but we will have to leave that
for another time (like a good class in phonology!)
14.4 Syllables 2: the Sonority Principle
Of the infinitely many possible consonant combinations, only a tiny fraction occur. And the
regularities mentioned in the previous section explain only a small part of this.
One other idea, one that we will formulate just roughly here, excludes a much larger range
of combinations than the generalizations given above. This idea is based on the idea that there
are degrees of sonority. Listing sounds in order of increasing sonority we get an order like the¶¶
following:
The Sonority Hierarchy:
−sonorant (=obstruent) +sonorant
stops affricates fricatives nasals liquids glides vowels
Very roughly, sonority corresponds to the amplitude (i.e. volume) of the speech sound. The
onsets and codas in English seem to respect this ordering according to the following principle:6
Sonority principle (SP): onsets usually rise in sonority towards the nucleus, and codas
fall in sonority away from the nucleus.
This accounts for the impossibility of words with onsets like rtag, while allowing trag. And
it accounts for the impossibility of words with codas like gatr while allowing words like gart.
Similar sonority hierarchies play this kind of role in other human languages too, though there
is significant variation in exactly what onsets and codas each language includes.
The SP reveals the syllable as a kind of cycle in the rising and falling sonority of human
speech, as Leonard Bloomfield proposed quite a long time ago:
In any succession of sounds, some strike the ear more forcibly than others: differ-
ences of sonority play a great part in the transition effects of vowels and vowel-like
sounds. . . In any succession of phonemes there will thus be an up-and-down of
sonority.. . Evidently some of the phonemes are more sonorous than the phonemes
(or the silence) which immediately precede or follow. . . Any such phoneme is a
crest of sonority or a syllabic; the other phonemes are non-syllabic. . . An utterance
is said to have as many syllables (or natural syllables) as it has syllabics. The ups
and downs of syllabification play an important part in the phonetic structure of all
languages. (Bloomfield, 1933 p120)
14.5 Stress (briefly!)
We won’t give much attention to stress in this class, but it is clear to everyone that in English
and many other languages, some syllables are more prominent, more stressed than others. We
6The text provides slightly more restricted observations, saying for example (p591): “. . . English, like many
other languages, does not allow sonorant-obstruent sequences at the beginning of a word. . . ”
134

have been indicating this with stress marks: a high mark to indicate primary stress on the
following syllable, and a low mark to indicate secondary stress on the following syllable:7
@sIm.@"le.S@n
The text proposes another representation, using a “pile” of grid marks to indicate the stress
of each syllable:
x
x x
x x x x x
@ sIm @ le S@n
Here we pile up grid marks according to the following rules:
o. each syllable has a grid mark x,
a. syllables with more stress have more grid marks, and
b. we use no more grid marks than necessary.
These rules can handle even more complex cases, like the phrase maintain assimilation in
which the second word has more stress on its most stressed syllable than the first word does.
The main idea here is simply that there are different levels of stress in words, phrases and
sentences.
These are ways of representing the stress in a word or phrase, but what decides where the
stress should go in the first place? It turns out that the rules of English stress assignment are
not simple, and they vary with dialects, so we will just observe a couple of things:
1. The text says (p598): “In English, stressed syllables – whether they carry main or subsidiary
stress – are chiefly identified by the vowel qualities they allow: vowels such as [æ], [a], [E],
[O], [U], or [u], [i] or [2] are permitted only under stress.”
By “stress”, this passage cannot mean just “main stress” though, as we see from many
examples we have already seen in the text, like these:
(p598) [i] in nuclear ["nUkliÄ] (p572) [i] in lucky ["l2ki]
(p572) [E] in extract [Ek"strækt] (p572) [a] in Exxon ["Eksan]
(p572) [E i] in mentality [mEn"tælIRi] (p535) [æ] in adhesion [æd"hiZ@n]
(p535) [a] in automatic [aR@"mæRIk]
In any case, it is clear that vowel quality and stress are related, as we see also in the
following generalization (one that was mentioned earlier). . .
2. A syllable is said to be light if its rime consists of just one short (-ATR) vowel, with no
coda; otherwise, it is heavy.8
In these terms, we can observe that in English:
Stressed syllables must be heavy (though not all heavy syllables are stressed.)
Since monosyllabic nouns and verbs are typically stressed, we see that this last idea
was already mentioned when we pointed out earlier that monosyllabic nouns and verbs
in English cannot end in lax vowels: we do not have nouns like [sI], [sE], [sæ], [sU].
3. English stress assignment varies with the “syntactic category” of a word.
7The [m] in this example immediately follows the stressed lax vowel [I] so it is really “ambisyllabic” in the
sense mentioned on page 132 of these notes, and shown by the tree for happy on page 588 of the text.
8As discussed on page 108 of the notes for Lecture 11, the short (-ATR), lax vowels of our local Californian
English are [I E æ U].
135

a. Examples. the verb digest, as in Did you digest that chapter?,
has stress on the second syllable [daI"ÃEst];
the noun digest, as in Let’s get the “Reader’s digest” version!,
has stress on the first syllable ["daIÃEst].
A similar thing happens in pairs like ábstract/abstráct, éscort/escórt, súrvey/survéy,
tórment/tormént, cónvict/conv´ıct.
b. Tendencies: (Burzio) In American English most unsuffixed words respect the follow-
ing rules, where a ’superheavy’ syllable is one that has either has a long vowel (the long
[+ATR] vowels are [e i a o u]) and a coda, or else a vowel and a two consonant coda:
• In nouns, a heavy penultimate syllable is stressed if there is one, and otherwise the
antepenult is stressed if there is one.
– for example, agenda /a"dZEnd@/, but America /@"merIk@/
• In verbs, if the final syllable is superheavy it gets stressed, and otherwise the penulti-
mate syllable gets stress.
– for example, prevent /pri"vEnt/, but imagine /I"mædZIn/
In recent surveys of the stress systems of various languages, among languages that assign
a single primary stress per word, penultimate and final stresses are fairly common, but initial
stress is found more often (Hyman 1977; see also Gordon 2004):
number of languages percentage
initial 114 37.3
final 97 31.7
penultimate 77 25.2
peninitial 12 3.9
antepenultimate 6 2
14.6 Reflecting on the big picture: Speech perception
Reflecting on the whole picture of phones, phonemes, syllables, and stress, it may seem simple
enough (?), but it makes the language understanding task seem quite amazing. What we hear
is nothing like a sequence of sounds that correspond 1 for 1 with the phonemes of words.
Rather, each sound in each word affects and is affected by neighboring sounds in complicated
ways, and the words are all run together. Thinking of the basic sequence of phonemes as a
row of Easter eggs, pouring forth in our speech at a rate of about 3 per second, the famous
linguist Charles Hockett described the speech understanding problem this way:
Imagine a row of Easter eggs carried along a moving belt; the eggs are of various sizes
and various colors, but not boiled. At a certain point, the belt carries the row of eggs
between the two rollers of a wringer, which quite effectively smash them and rub them
more or less into each other. The flow of eggs before the wringer represents the series of
impulses from the phoneme source; the mess that emerges from the wringer represents
the output of the speech transmitter. At a subsequent point, we have an inspector
whose task it is to examine the passing mess and decide, on the basis of the broken and
unbroken yolks, the variously spread out albumen and the variously colored bits of shell,
the nature of the flow of eggs which previously arrived at the wringer. (Hockett 1955,
p.210)
136

This might exaggerate our difficulties slightly. One thing that the phonological constraints
and “smearing” or “spreading” (i.e. assimilation) effects like nasalization provide is a kind
of redundancy. This is suggested in the text on p.522. With this redundancy, we can do
perfectly well even if we miss a bit here and there. The linguist Steve Pinker puts it this way:
Thanks to the redundancy of language, yxx cxn xndxrstxnd whxt x xm wrxtxng xvxn
xf x rxplxcx xll thx vxwxls wxth xn “x” (t gts lttl hrdr f y dn’t vn kn whr th vwls
r). In the comprehension of speech, the redundancy conferred by phonological rules can
compensate for some of the ambiguity in the sound wave. For example, a listener can
know that “thisrip” must be this rip and not the srip because the English consonant
cluster sr is illegal.
14.7 A question
To: E Stabler <stabler@ucla.edu>
Subject: Ling 20 Question
I have a question about the book. On page 583 it lists [+consonantal]
as a feature value for /a/. I am assuming this is a typographical
error, but I wanted to ask you about it nonetheless.
Right, a typo!
14.8 Summary
Phonemes are sound segments defined by features. Words are given by the sequence of
phonemes in them, but these features may be altered by phonological rules (such as stop
aspiration, vowel shortening, flapping), which apply in a certain order to the sequence of
sound segments.
Sequences of phonemes are organized into syllables. Each syllable has a nucleus, which
combines with an optional coda to form the rime, and an optional preceding onset. We drew
the parts of a syllable with a tree that has its root at the top. Every node in a tree is either
a mother or a leaf. Know how to syllabify English words and how to draw the syllables with
trees. And we will use trees a lot to show the parts of many other things later.
(The funny situation of ambisyllabicity, where a single sound is part of two different
syllables, is sometimes drawn with a structure that is not a tree because the shared consonant
has two mothers! We will not worry about this complication in this class.)
The parts of the syllable must conform to size limits and sonority patterns. We saw that the
things that can occur in the parts of English syllables are restricted. We considered restrictions
on nasal+non-coronal clusters, and on obstruent clusters as examples. At a higher level of
abstraction, more approximately, we also notice the sonority principle SP. We want to aim
for a story that provides a more accurate account of these generalizations.
We didn’t say much about stress in this class. We can observe quickly that stress patterns
can depend on whether a word is a verb or a noun. – And we will have much more to say
about these “syntactic categories” later. The tendencies for English stress on page 136 should
be understandable (but you don’t need to memorize them).
137

Though it’s only briefly mentioned in the text, it’s useful to understand the replies to
Hockett, mentioned in the last section of these notes, explaining why things are not as bad as
they might seem from the point of view of figuring out what phonemes you are hearing.
Finally, we have at least the outlines of an explanation for why no word of English could
have the phonetic representation [kkkk]. At least, we see that since this has no vowels, it
has no syllable nuclei, and hence no syllables. Furthermore, it violates the Not-too-similar
principle.
[1] Bloomfield, Leonard (1933) Language. Chicago: University of Chicago Press.
[2] Burzio, Luigi (1994) Principles of English Stress. Cambridge: Cambridge University Press.
[3] Chomsky, Noam and Morris Halle (1968) The Sound Pattern of English. NY: Harper and Row.
[4] Gordon, Matthew (2004) A factorial typology of quantity-insensitive stress. Natural Language
and Linguistic Theory 20: 491-552.
[5] Hayes, Bruce (1982) Extrametricality and English stress. Linguistic Inquiry 13: 357-394.
[6] Hockett, Charles (1955) A Manual of Phonology. International Journal of American Linguistics,
Memoir 11.
[7] Hyman, Larry (1977) On the nature of linguistic stress. In L. Hyman, ed., Studies in Stress and
Accent, University of Southern California, pp. 37-82.
[8] Kenstowicz, Michael (1994) Phonology in Generative Grammar. Cambridge, Massachusetts:
Blackwell.
[9] Ohala, John J. (1990) The phonetics and phonology of aspects of assimilation. In J. Kingston and
M. Beckman, eds., Papers in Laboratory Phonology I: Between the grammar and physics of speech.
Cambridge: Cambridge University Press.
[10] Pinker, Steven (1994) The Language Instinct. NY: Morrow.
[11] Steriade, Donca (1999) Alternatives to the syllabic interpretation of consonantal phonotactics.
Proceedings of the 1998 Linguistics and Phonetics Conference. Oxford: Karolinum Press.
138

Lecture 15 Review
We ﬁnally prepare for a comprehensive exam about the whole picture of language presented
in the class.
phones phonemes roots&affixes
words
relations
phrases
reorderingsreasoning, semantic
This section of the notes contains materials distributed in the last class (with some typos
removed!), followed by answers to the questions (0-21) that we discussed in that class. Finally,
a couple of paragraphs summarize at the end.
139

Phonetics and phonology. Phonemes (typically identified by ‘minimal pairs’):
manner voice place
1. /p/ spit plosive stop − labial
2. /t/ stuck plosive stop − alveolar
3. /Ù/ chip plosive stop affricate − alveopalatal
4. /k/ skip plosive stop − velar
5. /b/ bit plosive stop + labial
6. /d/ dip plosive stop + alveolar
7. /Ã/ jet plosive stop affricate + alveopalatal
8. /g/ get plosive stop + velar
9. /f/ fit fricative − labiodental
10. /T/ thick fricative − interdental
11. /s/ sip fricative − alveolar
12. /S/ ship fricative − alveopalatal
13. /h/ hat fricative − glottal
14. /v/ vat fricative + labiodental
15. / / though fricative + interdental
16. /z/ zap fricative + alveolar
17. /Z/ azure fricative + alveopalatal
18. /m/ moat nasal stop + labial
19. /n/ note nasal stop + alveolar
20. / / sing nasal stop + velar
21. /w/ weird central approximant + labiovelar
22. /j/ yet central approximant + palatal
23. /l/ leaf lateral approximant + alveolar
24. /ô/ reef central approximant + retroflex
25. /ô
"
/ or /Ä/ or /@ô/ bird central approximant + retroflex
tongue body
height
tongue body
backness
lip
rounding
tongue root
tense (+ATR)
or lax (−ATR)
1. /i/ beat high front unrounded +
2. /I/ fit high front unrounded −
3. /u/ boot high back rounded +
4. /U/ book high back rounded −
5. /E/ let mid front unrounded −
6. /o/ road mid back rounded +
7. /2/ shut low back unrounded −
8. /e/ ate mid front unrounded +
9. /æ/ bat low front unrounded −
10. /a/ pot low back unrounded +
11. /@/ roses mid back unrounded −
12. /aI/ lies dipthong
13. /aU/ crowd dipthong
14. /oI/ boy dipthong
liquids = /l,ô,ô
"
/, glides=/j,w/
coronals = dental, alveolar and alveopalatal stops, fricatives, affricates, liquids, and alveolar nasals
sonorants = vowels,glides,liquids,nasals, obstruents=non-sonorants
Rules deriving allophonic variations. For example, considering e.g. [no] vs. [m2n”T], and
[@"noI] vs. ["pæn”Tr
"
]:
(n dentalization) n →
+dental
/ T
Rules of syllable structure (universal). σ → (Ons) Rime, Rime → Nuc (Coda)
140

Phonotactics. In English, in any syllable, nasal-stop clusters are homorganic, and obstruent
clusters agree in voicing.
Morphology. Right hand head rule, and specific rules for affixes. E.g. N → -er / [V ]
Syntax. constituency tests: substitution for word level categories, substitution by a pro-
noun, substitution by do or do so, phrasal preposing, coordination, sentence fragments
TP → DP T’ NP → AP NP
T’ → T NegP NP → NP PP
T’ → T VP NP → NP CP
NegP → Neg VP VP → AdvP VP
DP →



(D) NP
Name
Pronoun



VP → VP PP
NP → N (PP) AP → AdvP AP
VP → V (DP) (



PP
CP
VP



)
PP → P (DP)
AP → A (PP)
CP → C’
C’ → C TP
AdvP → Adv
(Tricky detail: English be acts like an auxiliary, even when it’s the main verb.)
Semantics: [[Name T’]]=true just in case [[Name]]∈[[T’]]
[[No NP T’]]=true just in case [[NP]]∩[[T’]]=∅
[[Every NP T’]]=true just in case [[NP]]⊆[[T’]]
[[AP NP]]=[[AP]]∩[[NP]] just in case AP is intersective
(1) Ds are conservative in English (except only and phrases containing it)
(2) NPIs are OK if they are c-commanded by a decreasing DP or other negative element
(A) Reflexive pronouns are OK if they are c-commanded by an antecedent DP in same TP
(B) Non-reflexive pronouns cannot be c-commanded by an antecedent DP in same TP
(C) Other DPs (like John or every student) do not have antecedents
• D is conservative = [D N T’] entails [D N is/are N that T’]
• NPI = negative polarity item like ever, one bit, at all
• DP is decreasing = [DP VP1] entails [DP VP2] whenever [[V P2]] ⊆ [[V P1]]
• A c-commands B = A is a sister or aunt of B (in the syntactic tree)
• A is the antecedent of B = B gets its reference from A
141

Notation:
x ∈ A x is an element of set A
A ∩ B the intersection of A and B, that is,
the set of things that are in both A and B
A ⊆ B A is a subset (or equal to) B
|A| the number of things in set A
Universals We can now to understand these claims all human languages:
(1) Every human language is infinite (has infinitely many declarative sentences)
(2) Every human language distinguishes vowels and consonants,
and among the consonants, distinguishes obstruents from sonorants. Vowels are the
syllabic nuclei (with diff possibilities for Ons,Coda)
(3) Every human language has transitive and intransitive sentences, but the constituents
(Subject Object Verb) occur in different orders:
SOV (Turkish,Japanese,Navajo,Hopi,Mojave,Burmese,Somali,Walbiri)
SVO (English,Czech,Mandarin,Thai,Vietnamese,Indonesian)
VSO (Welsh,Irish,Tahitian,Chinook,Squamish)
rare: VOS (Malagasy,Tagalog,Tongan), OVS (Hixkaryana), OSV ?
These are summarized in Joseph Greenberg’s (1963) famous claim G1:
G1. In declarative sentences with nominal subject and object, the dominant order is almost
always one in which the subject precedes the object.
G2. In languages with prepositions, the genitive almost always follows the governing noun,
while in languages with postpositions it almost always precedes.
G3. Languages with dominant VSO order are always prepositional.
G4. With overwhelmingly greater than chance frequency, languages with normal SOV order
are postpositional.
• pay
he
t’antata
bread-acc
mikhu-n
eat-s
mantekilla-wan
butter-with
(Quechua)
G11. Inversion of statement order so that verb precedes subject occurs only in languages
where the question word or phrase is normally initial. This same inversion occurs in
yes-no questions only if it also occurs in interrogative word questions.
G12. If a language has a dominant order VSO in declarative sentences, it always puts inter-
rogative words or phrases first in interrogative word questions; if it has dominant order
SOV in declarative sentences, there is never such an invariant rule. (In Greenberg’s
sample, these languages put question words first: Berber, Finnish, Fulani, Greek, Guarani,
Hebrew, Italian, Malay, Maori, Masai, Mayan, Norwegian, Serbian, Welsh, Yoruba, Zapotec.)
G20. When any or all of the items (demonstrative, numeral, and descriptive adjective) pre-
cede the noun, they are always found in that order. If they follow, the order is either
the same or its exact opposite.
• yu-le
that
m-tu
person
m-moja
one
m-refu
tall
a-li-y-e-ki-soma
he-past-rel-it-read
ki-le
that
ki-tabu
book
ki-refu
long
(Swahili)
‘that one tall person who read that long book’
142

Some review questions:
(0) Draw complete trees for the following sentences, showing movements with arrows:
(a) The students will have been complaining about the bad jokes!
(b) She knows which foreign students the class has inspired.
(c) Several short questions usually suffice.
(d) The midterm did not confuse most students.
You should know how to use constituency tests to defend these trees, how to list the rules used
by each tree, and how to list the lexical items (spelling, category, arguments) in each tree.
(1) The flap [R] is a phone of English T F
(2) No voiced stops are phonemes of English T F
(3) /t/ is an allophone of the English phoneme [R] T F
(4) In English, there is a vowel that is sometimes pronounced [s] T F
(5) In English, there is a fricative that is sometimes pronounced [s] T F
(6) In English, [s] is sometimes a syllable T F
(7) In English, there is a word that is sometimes pronounced [s] T F
(8) In English syllable onsets with 2 consonants, the 1st is usually more sonorant T F
(9) Every English affix consists of 1 or more complete syllables T F
(10) There are more than 10 phonemes in English T F
(11) More than 20 different phoneme sequences can be syllable onsets in English T F
(12) More than 100 different phoneme sequences can be syllables in English T F
(13) More than 100000 different phoneme sequences can be the pronunciations of English determiners D T F
(14) More than 100000000 different phoneme sequences can be the pronunciations of English phrases T F
(15) Complements are always obligatory, never optional T F
(16) English wh-movement moves phrases with any number of morphemes T F
(17) English wh-movement can move a phrase out of a preposition T F
(18) [[sings]] is always a subset of [[sings beautifully]] T F
(19) [[sings and dances]] is always a subset of [[sings or dances]] T F
(20) More than 100 human languages are SV0 T F
(21) Consider this (slightly simplified) grammar of Malagasy, a language spoken by more
than 20 million speakers, mainly in Madagascar, an island off the east coast of Africa.
It has 4 vowels: /a,i,e,u/ and these 20 consonants.
labial alveolar velar glottal
obstruent stop -voice p t,⁀ts,Ù k
+voice b d,⁀dz,⁀dr g
fricative -voice f s h
+voice v z
sonorant +nasal m n
-nasal l,R
a. In this language, syllable onsets have a single consonant. Codas usually are empty
but sometimes also have a single consonant. Which of the following are not possible
Malagasy words:
minan⁀dri bararantla landihazu misuÙ mamangliha
b. In this language, one syllable roots get primary stress on their only syllable; two-
syllable and longer get stress on penultimate uless they end in [ka, tra, na] in which
case they get stress on their antepenultimate syllable if there is one:
mandeha ‘to go’ be ‘big’
lena ‘wet’ karama ‘salary’
vohitra ‘hill’ fotsi (spelled ‘fotsy’) ‘white’
c. Allophones. In this language, after a stressed syllable, /a, u, i/ are reduced to [@].
(V reduction) [a,u,i] → @/"σ. . .
With this rule, how are the 6 forms in (21b) pronounced?
143

d. Phonotactics. No vowel after the final stressed vowel in a word has the sound [e].
Which (if any) of the examples in (21b-21c) are impossible for this reason?
e. Syntax. Phrase structure rules. For sentences here, assume same as English, ex-
cept for coordination:
α → Coord α Coord α (for α=D,V,N,A,P,C,Adv,VP,NP,AP,PP,AdvP,TP,CP),
where the two Coordinators must be the same.
Movement rules. For the sentences here, let’s assume that all sentences (even ones
with NegP) have affix hopping but to prefix (not suffix) position, and that (after
affix hopping, if it applies) VP or NegP always moves to attach to CP (in the same
way wh-phrases do in English).
Lexical items.
word meaning category arguments
n- past T {VP, NegP}
m- pres T {VP, NegP}
tsy not Neg
ijinja cut V DP DP
ahotra fear V DP DP
an-tsirambina carelessly Adv
ny the D NP
vary rice N
mpamboly farmer N
inona what DP
izy he/she DP
na or Coord
Draw a complete tree for the following sentence, showing all movements with ar-
rows:
N- ijinja vary an-tsirambina ny mpamboly
past- cut rice carelessly the farmer
‘The farmer carelessly cut rice’
f. Paul (2005) argues that na inona na inona is an NPI in the following sentence:
Tsy m- atahotra na inona na inona izy
not pres- fear or what or what he/she
‘he/she fears nothing’
Draw the tree and explain whether respects the NPI requirement we proposed in
class (and listed on the review page)
144

Answers to the previous review questions, except that it is not easy for me to
draw the arrows indicating the movements in the syntactic trees – Make sure you can do that
for all the trees shown here!
(0a) The students will have been complaining about the bad jokes!
CP
C’
C TP
DP
D
the
NP
N
students
T’
T
V
will
T
-s
VP
V VP
V
have
VP
V
been
VP
VP
V
complaining
PP
P
about
DP
D
the
NP
AP
A
bad
NP
N
jokes
(0b) She knows which foreign students the class has inspired.
CP
C’
C TP
DP
she
T’
T VP
V
V
know
T
-s
CP
DP
D
which
NP
AP
A
foreign
NP
N
students
C’
C TP
DP
D
the
NP
N
class
T’
T
V
have
T
-s
VP
V VP
VP
V
inspired
DP
145

(0c) Several short questions usually suﬃce.
CP
C’
C TP
DP
D
several
NP
AP
A
short
NP
N
questions
T’
T VP
AdvP
Adv
usually
VP
V
V
suﬃce
T
PRES
(0d) The midterm did not confuse most students.
CP
C’
C TP
DP
D
the
NP
N
midterm
T’
T
V
DO
T
-ed
NegP
Neg
not
VP
V
confuse
DP
D
most
NP
N
students
(1) The flap [R] is a phone of English T F
(2) No voiced stops are phonemes of English T F
(3) /t/ is an allophone of the English phoneme [R] T F
(4) In English, there is a vowel that is sometimes pronounced [s] T F
(5) In English, there is a fricative that is sometimes pronounced [s] T F
(6) In English, [s] is sometimes a syllable T F
(7) In English, there is a word that is sometimes pronounced [s] T F
(8) In English syllable onsets with 2 consonants, the 1st is usually more sonorant T F
(9) Every English affix consists of 1 or more complete syllables T F
(10) There are more than 10 phonemes in English T F
(11) More than 20 different phoneme sequences can be syllable onsets in English T F
(12) More than 100 different phoneme sequences can be syllables in English T F
(13) More than 100000 different phoneme sequences can be the pronunciations of English determiners D T F
(14) More than 100000000 different phoneme sequences can be the pronunciations of English phrases T F
(15) Complements are always obligatory, never optional T F
(16) English wh-movement moves phrases with any number of morphemes T F
(17) English wh-movement can move a phrase out of a preposition T F
(18) [[sings]] is always a subset of [[sings beautifully]] T F
(19) [[sings and dances]] is always a subset of [[sings or dances]] T F
(20) More than 100 human languages are SV0 T F
146

(21) Consider this (slightly simplified) grammar of Malagasy, a language spoken by more
than 20 million speakers, mainly in Madagascar, an island off the east coast of Africa.
It has 4 vowels: /a,i,e,u/ and these 20 consonants.
labial alveolar velar glottal
obstruent stop -voice p t,⁀ts,Ù k
+voice b d,⁀dz,⁀dr g
fricative -voice f s h
+voice v z
sonorant +nasal m n
-nasal l,R
a. In this language, syllable onsets have a single consonant. Codas usually are empty
but sometimes also have a single consonant. Which of the following are not possible
Malagasy words:
minan⁀dri bararantla landihazu misuÙ mamangliha
b. In this language, one syllable roots get primary stress on their only syllable; two-
syllable and longer get stress on penultimate (next to last) unless they end in [ka,
tra, na] in which case they get stress on their antepenultimate syllable (one before
the next to last) if there is one:
man."de.ha ‘to go’ "be ‘big’
"le.na ‘wet’ ka"ra.ma ‘salary’
"vo.hi.tra ‘hill’ "fot.si (spelled ‘fotsy’) ‘white’
(Actually, the rule I gave does not specify a stress for lena, so the right thing to
do is to note this fact. I should have been explicit in saying that a bisyllabic word
ending in [ka, tra, na] gets penultimate stress.)
c. Allophones. In this language, after a stressed syllable, /a, u, i/ are reduced to [@].
(V reduction) [a,u,i] → @/"σ. . .
With this rule, how are the 6 forms in (21b) pronounced?
man."de.h@ "be
"le.n@ ka"ram@
"vo.h@.tr@ "fot.s@
d. Phonotactics. No vowel after the final stressed vowel in a word has the sound
[e].
Which (if any) of the examples in (21b-21c) are impossible for this reason? ALL
OK.
147

e. Syntax. Phrase structure rules. For sentences here, assume same as English,
except for coordination:
α → Coord α Coord α (for α=D,V,N,A,P,C,Adv,VP,NP,AP,PP,AdvP,TP,CP),
where the two Coordinators must be the same.
Movement rules. For the sentences here, let’s assume that all sentences (even ones
with NegP) have affix hopping but to prefix (not suffix) position, and that (after
affix hopping, if it applies) VP or NegP always moves to attach to CP (in the same
way wh-phrases do in English).
Lexical items.
word meaning category arguments
n- past T {VP, NegP}
m- pres T {VP, NegP}
tsy not Neg
ijinja cut V DP DP
ahotra fear V DP DP
an-tsirambina carelessly Adv
ny the D NP
vary rice N
mpamboly farmer N
inona what DP
izy he/she DP
na or Coord
Draw a complete tree for the following sentence, showing all movements with ar-
rows:
N- ijinja vary an-tsirambina ny mpamboly
past- cut rice carelessly the farmer
‘The farmer carelessly cut rice’
Draw the arrows in the tree!
CP
VP
VP
V
T
n-
V
ijinja
DP
NP
N
rice
AdvP
Adv
an-trisrambina
TP
DP
D
ny
NP
N
mpamboly
T’
T VP
148

f. Paul (2005) argues that na inona na inona is an NPI in the following sentence:
Tsy m- atahotra na inona na inona izy
not pres- fear or what or what he/she
‘he/she fears nothing’
Draw the tree and explain whether respects the NPI requirement we proposed in
class (and listed on the review page)
CP
NegP
Neg
tsy
VP
V
T
m-
V
atahota
DP
Coord
na
DP
inona
Coord
na
DP
inona
TP
DP
izy
T’
T NegP
Draw the arrows!
Yes, the DP [na inona na inona] is c-commanded by the Neg [tsy], since it is
contained in the VP sister of Neg.
15.1 Summary summary
Since the chart of speech sounds and list of syntax rules given on pages 140-141 of these lecture
notes will be provided with the exam, you do not need to memorize them, but you need to
understand what they mean!
There might be a question about how you figure out what the phonemes are, and about
the rules that derive allophones. There is likely to be a question about constituency tests in
syntax. And there will probably be questions about the semantics of determiners, about how
antecedents c-command pronouns and reflexives (or negative elements c-command negative
polarity items), and maybe even a question about when determiners are conservative.
Also you should know the very basics from Lecture 1. In our language use we see a kind of
“creativity,” which Chomsky thinks we may never really understand, even when we understand
the “productivity” and “compositionality” in language. Make sure you know what that means.
And there may be a question about how many phonemes there are in English (39), and about
how many syllables (a lot, but finitely many), words, phrases, and sentences (∞) there are.
149

[1] Chomsky, Noam (1968) Language and Mind. NY: Harcourt Brace Javonovich.
[2] Greenberg, Joseph H. (1963) Some universals of language with particular reference to the order of
meaningful elements. In J.H. Greenberg, ed., Universals of Language. Cambridge, Massachusetts:
MIT Press.
[3] Greenberg, Joseph H. (1966) Language Universals, with special reference to feature hierarchies.
The Hague: Mouton.
[4] Hawkins, John A. (1983) Word Order Universals. NY: Academic Press.
[5] Hawkins, John A. (1990) A parsing theory of word order universals. Linguistic Inquiry 21: 223 –
262.
[6] Steele, Susan (1978) Word order variation. In J.H. Greenberg, ed., Universals of Human Language.
Stanford: Stanford University Press.
150

Index
θ-roles, 43
accusative case, 50
acoustic perception, 101
adjectives (A), 26
adjectives (A), substitution test for, 27
adverbs (Adv), 26
adverbs (Adv), substitution test for, 27
affix hopping, 62
affixes, 5
affricates, 105
agent, θ-role, 43, 47
agglutinating language, 19
aI-raising, 124
allophone, 109, 119
alphabet
morphemic, 102
phonemic, 102
syllabic, 102
alveolar ridge, 102
ambiguity
morphological, 25
syntactic, 68
ambisyllabic consonants, 132
American Sign Language, ASL, 1, 106, 130
analytic language, 19
Anderson, Stephen R., 20
arguments, in syntax, 39, 43, 45
aspiration of stops, 111, 112
assimilation, 137
ATR feature, 108
auxiliary verbs, 53
Baker, Mark, 20
Barwise, Jon, 10
Basciano, Bianca, 26
Bickel, Balthasar, 20
Bickerton, Derek, 25
binding, of pronouns, 93
Bloomfield, Leonard, 134, 138
Booij, Geert, 31
Broadbent, D.E., 101, 110
Burmese, 142
Burzio, Luigi, 136, 138
c-command, 92, 94
c-selection, 46
Carnap, R., 82, 87
case, of DPs, 50
Caucasian, 120
Ceccagno, Antonella, 26
children, in a tree, 44
Chinese, 102
Chinese, Mandarin, 109, 142
Chinook, 142
Chomsky, Noam, 4, 10, 52, 59, 138, 150
coda, of syllable, 130
complementizers, introduced, 28
complements, in syntax, 36, 43, 46
compositionality, 3
compounds, 5
compounds, in morphology, 12
conservative, determiners, 89
consonants, basic features, 105
constituency tests, in syntax, 30
Cooper, F.S., 110
coordination, 33
coronals, 106
creative aspect of language, 4
Czech, 142
daughter node, in a tree, 130
dental, 102
determiners (D), 26
determiners (D), substitution test for, 27
Di Sciullo, Anna-Maria, 20
diphthongs, basic features, 107
direct object, 39
dissimilation, 110
distributional properties, and syntactic cate-
gory, 27
DO, not an aux, 64
DO-support, 64
Downing, Pamela, 31
151

embeeded, 48
English, 142
entailment, 81
epenthesis, 110
experiencer, θ-role, 43
Fabb, Nigel, 13, 20
Fijian, 19
flapping, 109, 114
flexible aspects of language, 5
Fodor, J.A., 82, 87
Fodor, J.D., 82, 87
formants, 100
Frege, Gottlob, 3, 10, 82
fricatives, 105
fusional language, 19
Garrett, M.F., 82, 87
genitive case, 50
glides, 106
glottis, 102
Gordon, Matthew, 138
Greenberg, Joseph H., 150
grid marks, representing stress, 135
Halle, M., 108
Halle, Morris, 20, 110, 138
Haspelmath, Martin, 20
Hawkins, John A., 150
Hayes, Bruce, 138
head-final principle, 15
heads, in syntax, 36, 39, 46
heavy syllable, 135
Hindi, 108
Hockett, Charles, 136, 138
homorganic
nasal-stop, 133
Hopi, 142
Huang, Yi Ting, 10
Hyman, Larry, 138
idioms, 17
indirect object, 40
Indonesian, 142
intersective adjectives, 85
intransitive verbs, 46
Irish, 142
isolating language, 19
Jackendoff, Ray S., 25
Japanese, 102, 142
katakana, 102
Keenan, Edward L., 90
Kenstowicz, Michael, 138
Khoisan, 120
labial, 102
Ladefoged, Peter, 101, 109, 110
Lapata, Maria, 31
lax vowels, 108
leaf node, in a tree, 130
Levi, Judith, 25
Liberman, A., 101, 110
Lieber, Rochelle, 20
light syllable, 135
liquids, 106
long vowels, 108
Maddieson, Ian, 109, 110
Marantz, Alec, 10, 20
meaning postulate, 82
metathesis, 110
modal verbs, 53
modal verbs, introduced, 27
modifiers, in syntax, 36
Mojave, 142
Moreton, Elliott, 123
morphemes, 6
morphological atoms, 18
Morris, Joanna, 10
mother node, in a tree, 130
nasal, 102, 105
nasalization, 110, 116, 122, 137
Navajo, 142
negative adjectives, 85
negative polarity items (NPIs), 90, 92
Nichols, Johanna, 20
node, in a tree, 130
nominative case, 50
nouns (N), 26
NPI licensing rule, 92, 93
nucleus, of syllable, 130
O’Grady, William, 20
objective case, 50
obstruents, 106
onset, of syllable, 130
onsets, possible in English, 131
open syllables, 131
P¯an.ini, 1
152

Packard, Jerome, 26
palate, 102
part of speech, 11, 12
patient, θ-role, 43
Perlmutter, David M., 110
Perry, John, 10
Peters, Stanley P., 59
phoneme, 112
phonemes, of standard American English, 119
phones, 105, 108, 109
phonological rules, 119
phrase structure rules, 44
Pima, Uto-Aztecan language, 17
Pinker, Steven, 10, 137, 138
pluralization, and ±voice in English, 99
Polynesian, 120
polysynthetic language, 19
postposing, 31, 33
prefixes, 5
preposing, 31, 33
prepositions (P), 26
productivity of language, 3
compounding, 24
morphology, 23
pseudoclefts, 42
Pylkkänen, Liina, 10
Quechua, 19
Quine, Willard van Orman, 82
referential use of pronouns, 93
rhyming poetry and rime, 130
right (hand) head rule, 15
rime, of syllable, 130
Ritchie, Robert W., 59
root, in a tree, 130
roots, 5
s-selection, 43
Schatz, C.D., 101, 110
segments, speech, 117
selection, see c- and s-selection, 46
Selkirk, Elisabeth O., 21
semantic atoms, 6
sentence fragments, 33
Serbo-Croatian, 109
Shankweiler, D.P., 110
short vowels, 108
sign languages, 1
sisters, in a tree, 43, 46
Somali, 142
sonorants, 106
sonority hierarchy, 134
sonority principle (SP), 134
spectrograph, 100
speech perception, 136
Spencer, Andrew, 21
spirantization, 125
Squamish, 142
Stabler, Edward, 21
Stavi, Jonathan, 90
Steele, Susan, 150
Steriade, Donca, 138
Stockall, Linnaea, 10
stops, 105
stress assignment, 135
Studdert-Kennedy, M., 110
subcategorization, of heads, 46
subject-auxiliary inversion, see also T-to-C move-
ment, 57
subject-verb agreement, 50
suffixes, 5
Swahili, 142
syllabification, in English, 131
syllable structure, 130
syllable, light and heavy, 135
syntactic atom, 12, 27
syntactic categories, introduced, 11
syntax, definition, 26
T-to-C head movement, 66
Tahitian, 142
tense vowels, 108
Thai, 142
Thomas, Erik R., 123
to, as T, 63, 77
tone languages, 109
topicalization, 31, 33
transitive verbs, 46
tree structures, introduced, 130
Turing, Alan, 10
Turkish, 142
Ubykh, 120
universals, 142
V-to-T head movement, 66
velar, velum, 102
verbs (V), 26
Vietnamese, 142
vowel shortening, 113
vowels, basic features, 107
153

vowels, tense and lax, 108
Walbiri, 142
Welsh, 142
wh-movement, 57
Williams, Edwin, 20
word, deﬁnition, 6
Yinglish, 130
Zipf, George K., 10
Zweig, Eytan, 10
Zwicky, Arnold M., 21
154

Introduction to linguistics

More Related Content

What's hot (20)

Viewers also liked (12)

Similar to Introduction to linguistics (20)

Recently uploaded (20)

Introduction to linguistics