Discourse Corpra about the subject of semantics

What today’s lecture is NOT about
0 computational processing of discourse, discourse parsing
of any type
0 event annotation (Martha Palmer, James Pustejovsky et
al.)
0 various forms of connectives (DRDs) in different
languages (other classes)
0 just a little about spoken corpora

What today’s lecture IS about
0 Discourse theories (RST, SDRT, PDTB, …)
0 Discourse corpora
0 A little about coreference and other discourse-related
topics
0 Overview of discourse theories from the point of view of
relations and conceptions (not connectives, they are
analyzed within other courses)

PLAN
0 general words about cohesion/coherence, large/small
discourse units, and spoken/written, elementary
discourse units, requirements to discourse structure

Processing Large
Discourse Units
M. Stede. Discourse processing, 2012.
• Logical document structure
a hierarchy of segments of the document, each of which
corresponds to a visually distinguished semantic
component of the document (Summers 1998)
Discourse Processing

Processing Large
Discourse Units

• Content zones
a continuous portion of a text document that fulfills a
functional role for the text as a whole, contributing to
the overall message or purpose, as it is characteristic for
the genre of the text (defined for each sentence)
Teufel and Moens (2002) - aimed at recovering content zones
from conference papers in CL.
• Aim, research goal
• Textual: statements about section structure
• Own: description of the authors’ work (methodology, results,
discussion)
• Background: generally accepted scientific background
• Contrast: comparison with other work
• Basis: statements of agreement with other work
• Other: description of other researchers’ work
Processing Large
Discourse Units

• Content zones
Processing Large
Discourse Units
• Topic-based segmentation
a sequence of non-overlapping text segments that
completely covers the text. Each unit consists of one or
more sentences that address a common topic
• connectives,
• word repetition,
• bridging relations, related
words,
• pronouns, ellipsis,
• lexical chains

• Content zones
Processing Large
Discourse Units
• Discourse modes (C. Smith, 2003)
narrative, information, argument (commentary), report,
description
In NLP for:
• temporal discourse
processing,
• automatic
summarization,
• information extraction,
• argumentation mining,
• genre distinctions, …
Situation entity types (A. Friedrich)
• state
• event
• generic sentence
• generalizing sentence
• general stative
• abstract entity
• report
• speech act
Annemarie Friedrich. Annotation and automatic classification of
situation entity types. 2015

• Content zones
Processing Large
Discourse Units
description
Coreference resolution

• Content zones
Processing Large
Discourse Units
description
Coreference resolution
Processing Small
Discourse Units coherence relations connectives, cue phrases
our focus
for today

cohesive elements like connectives are viewed as
important but not necessary features of discourse:
they are linguistic markers expressing the
underlying conceptual relations that are of a
cognitive nature (Sanders, 1992)
COHESION COHERENCE
In a cohesion analysis the connectivity of the discourse
is primarily tied to the explicit marking of semantic
relations. These explicit cues make a text a text. BUT:
cohesion is necessary but not sufficient condition for
the creation of text (Halliday and Hasan, 1976)
Coherence and cohesive relations are the
threads that make up a text
Figure from Taboada (2015)
after Taboada (2015)

Requirements to discourse structure
A satisfying theory of discourse structure should meet
discourse structure makes it possible
to describe the structure of all kinds of
natural texts
a theory of discourse structure should at least
generate plausible hypotheses on the role of
discourse structure in the construction of the
cognitive representation
descriptive
adequacy
psychological
plausibility
after Sanders (1992)

PLAN
All theories (?) claim that in coherent discourse, every EDU (except maybe one) should be linked to
another via some coherence relation
BUT what they differ in:
Questions that various theories are trying to answer
1)Can substructures of discourse serve as arguments of coherence relations? (Asher - yes, some theories
don’t)
2)Attachment principles - where can we attach a new information:
a)attachment to the last in a queue, to a preceding sentence (dynamic semantics)
b)long distance dependencies (RST-treebank, SDRT - prove that there are long-distance
dependencies: 35-40% of the attachments are outside the sentence and beyond the previous
sentence) - there are some principles how you can attach, graph, postions of the graph, primary
attachment portions (right frontier constraint)
c)dialogues give interesting constraints (to backwards attachment - Although I’m tired, I’m enjoying it
here)
3)How many discourse functions can a single unit have? We bought the apartment, but then we rented it
out. (N. Asher) Renting it out has two incoming adges, but we can imagine incoming edges coming from
several sources

How many discourse functions can a single unit
have?
We bought the apartment, but then we rented it out. (N. Asher)
(?)Lucka
find: an example of discourse relation where one unit is connected with different
relations to two different units.

Penn Discourse Treebank (PDTB)
Prague Dependency Treebank (PDT)
ANNODIS (French)
DiscAn Corpora
STAC
RST signalling corpus
Postdam Commentary Corpus
RST Spanish Corpus
Basque-Spanish-English parallel corpus (RST)
Discourse Graphbank
DISCOR
GECCo
French Discourse Treebank
Basque RST Treebank
Turkish Discourse Bank

Elementary Discourse Units (EDUs)
Give me your essay, please. Then you may leave the room.
When you leave the room, give me your essay.
Give me your essay, please. The best one.
John attended the lecture despite his illness.
John, although he was ill, attended the lecture.
The red car that‘s parking in front of you belongs to me.
The red car, which was brought here by my Dad, belongs to me.
embeddings
nominalizations
restrictive and non-
restrictive relative clauses
Exhausted, I called for a taxi cab.
“free adjuncts”

Elementary Discourse Unit (EDU)
A span of text, usually a clause, but in general ranging from minimally
a (nominalization) NP to maximally a sentence. It denotes a single
event or type of event, serving as a complete, distinct unit of
information that the surrounding discourse may connect to. An EDU
may be structurally embedded in another.
Abstract objects
(Asher, 1993)
Pragmatic elementary
units (Polanyi et al. 2004)
Manfred Stede. Discourse Processing. 2012
• meaning,
• discourse function,
• independent continuation
BUT!
Vary from researcher
to researcher,
depending on the level
of granularity needed

Asher’s DUs:
basic intuition – they also
exist underneath the clausal
level, and they can either be
0 titles
0 oppositions
0 relative clauses
0 some NP (e.g. principles)
[Interprovincial Pipe Line Co. said]π1 [it will delay
a proposed two-step, 830 million dollar [(US$705.6
million)]π3 expansion of its system]π2 [because
Canada’s output of crude oil is shrinking.]π4

Inventory and definitions of relations

Spoken vs. Written
0 Text of a single author –
he’s organizing his
thoughts together, make
it coherent as much as
he can
0 Conversation as a game of message
exchange involving a kind of signaling
game, a play with reactions:
0 X plays φ  Y decodes a message  Y
decides what signal to send in return
 X decodes a message.
very few discourse annotated
corpora for dialogues (STAC,
Italian?)

ready with general words,
now let’s continue to theories
and corpora
вложить смешную картинку и смайлик

0Halliday and Hasan – Cohesion in English
0RST – Rhetorical Structure Theory
0SDRT - Segmented Discourse Representation Theory
0PDTB – Penn Discourse Treebank
0Discourse Graphbank
0CCR – Cognitive approach to coherence relations

0 We can distinguish what is text and what is not  text has
structure
0 Text is realized by sentences but it does not have the same
structural integration among its parts as we find among the
parts of a sentence or clause
0 Investigate the resources that English has to create text
0 Cohesion – a set of explicit cues – is what makes a text a text
M.A.K.Halliday and Ruqaiya Hasan
Cohesion in English

M.A.K. Halliday & Ruqaiya Hasan. Cohesion in English. 1976
0 Reference: identity between entities
0 John love Mary. However he is afraid to kiss her.
0 Substitution: similarity between different instantiated entities of the
same type
0 These biscuits are stale. Get some fresh ones.
0 Ellipsis
0 Would you like to hear another verse? I know twelve Ø more.
0 Conjunction: logico-semantic relations between propositions
(e.g. addition, contrast, cause)
0 John love Mary. However he is afraid to kiss her.
0 Lexical cohesion (similarity between entities of the same type
based on sense relations (e.g. hyperonymy, part-whole relations)
0 Why does this little boy wriggle all the time? Girls don’t wriggle.
Types of cohesion:

German-English Contrasts in
Cohesion (GECCo)
0 based on the definition of cohesion and cohesive devices in
English by (Halliday & Hasan, 1976)
0 elaborated for a contrastive analysis of two languages
0 comparable and parallel texts in English and German (ca.
80,000 sentences)
0 various registers, including written and spoken dimensions
0 complex morpho-syntactic annotation

Rhetorical Structure Theory
Mann, W. C. and S. A. Thompson (1988). Rhetorical structure theory.
Toward a functional theory of text organization. Text 8(3), 243–281.
Taboada, Maite and William C. Mann. (2006). Rhetorical Structure Theory: Looking
back and moving ahead. Discourse Studies, 8 (3), 423-459.
http://www.sfu.ca/rst/
http://www.sfu.ca/rst/05bibliographies/

• RST is about how text works (primarily written  spoken)
• Empirical perspective, comes from corpus analysis
 Coherent texts consist of minimal
units, which are linked to each
other, recursively, through
rhetorical relations (coherence or
discourse relations )
 Coherent texts do not show
gaps or non-sequiturs.
 The resulting structure is a complete tree: no cross-dependencies.
• completedness (one schema application contains the entire text)
• connectedness (each span, except for the span that contains the entire text, is either a minimal unit or a constituent of another
schema application)
• uniqueness (each schema application contains a different set of text spans)
• adjacency (the spans of each schema application constitute one contiguous text span)
3’) Mary
loves Jonh.

31
RST: Graphical Representation
0 A horizontal line covers a
span of text (possibly made
up of further spans)
0 A vertical line signals the
nucleus or nuclei
0 A curve represents a
relation, and the direction
of the arrow, the direction
of satellite towards nucleus

• Definition of the relations are based on functional and semantic criteria, not on
morphological or syntactic signals, because no reliable or unambiguous signal for
any of the relation was found
• Different lists of the relations exist:
• Mann and Thomson – 24 relations (Cause, Contrast, Elaboration,
Restatement, Evidence, Conditions, Antithesis…)
• Later - List, Means, Preparation,
Unconditional, Unless, two Restatements
(nuclear and multinuclear), Joint (the
declared absence of a relation)

Circumstance Antithesis and Concession
Solutionhood Antithesis
Elaboration Concession
Background Condition and Otherwise
Enablement and Motivation Condition
Enablement Otherwise
Motivation Interpretation and Evaluation
Evidence and Justify Interpretation
Evidence Evaluation
Justify Restatement and Summary
Relations of Cause Restatement
Volitional Cause Summary
Non-Volitional Cause Other Relations
Volitional Result Sequence
Non-Volitional Result Contrast
Purpose
Relation names (Mann-Thompson 1988)
Other classifications are possible, both longer and shorter lists have been proposed

Relations in RST
Coherence relations between discourse segments
• asymmetric (“mononuclear”)
• one nucleus, one satellite
• symmetric (“multinuclear“)
• multiple nuclei
nucleus
satellite
nucleus nucleus
≈ hypotaxis
≈ parataxis
other schemas

Hypotactic (subordinate)
0 Sub-sentential Concession
relation
0 Concession across sentences
0 Nucleus (spans 2-3) made up of
two spans in an Antithesis
relation

36
Paratactic (coordinate)
• At the sub-sentential level (traditional coordinated
clauses)
Peel oranges, and slice crosswise.
• But also across sentences
1. Peel oranges, 2. and slice crosswise. 3. Arrange in a
bowl 4. and sprinkle with rum and coconut. 5. Chill until
ready to serve.

Relations in RST - Constraints
1. Constraints on the Nucleus
The reader may not believe N to a degree
satisfactory to the writer
2. Constraints on the Satellite
The reader believes S or will find it credible
3. Constraints on the combination of N+S
The reader’s comprehending S increases their
belief of N
4. Effect (the intention of the writer)
The reader’s belief of N is increased
A relation consists of:
1. Constraints on the Nucleus,
2. Constraints on the Satellite,
3. Constraints on the combination of
Nucleus and Satellite,
4. The Effect.
Example: Evidence
1) Darwin as a Geologist 2) He tends to be viewed now as a biologist, 3) but in his five years on the Beagle his
main work was geology, 4) and he saw himself as a geologist. 5) His work contributed significantly to the field.

RST corpora
0 RST Discourse Treebank (Carlson et al. 2003)
0 RST Signalling Corpus (Das and Taboada 2015)
0 Potsdam Commentary Corpus
0 A corpus of 220 German newspaper commentaries annotated with different types of linguistic information, including RST
0 Stede, Manfred and A. Neumann (2014). Potsdam Commentary Corpus 2.0: Annotation for Discourse Research. Proceedings of the Language
Resources and Evaluation Conference (LREC), Reykjavik.
0 Discourse Relations Reference Corpus
0 texts from RST web site +annotated Wall Street Journal articles from the RST Discourse Treebank +review texts from the SFU Review Corpus
0 The documents in each of the subcorpora have been annotated with RSTTool. The corpus is useful to anyone interested in discourse relations,
from whatever perspective. The annotations provide rich information on what relations are more common; how they are commonly signalled;
and how relations are distributed in different genres.
0 description and the full collection are available from the Discourse Relations Reference Corpus page
(http://www.sfu.ca/rst/06tools/discourse_relations_corpus.html)
0 GUM - The Georgetown University Multilayer Corpus
0 POS, lemmas, syntax, constituent and dependency syntax, Information status (given, accessible, new)
0 Entity and coreference annotation
0 Rhetorical Structure Theory
0 Spanish RST Discourse Treebank
0 http://corpus.iingen.unam.mx/rst/
0 da Cunha, Iria, Juan Manuel Torres-Moreno and Gerardo Sierra (2011) On the development of the RST Spanish Treebank, Proceedings of the
Fifth Language and Annotation Workshop (LAW V) (pp. 1-10). Portland, OR.

RST Discourse Treebank
L. Carlson et al. 2003
Carlson, Lynn; Daniel Marcu and Mary Ellen Okurowski. RST Discourse Treebank
LDC2002T07. Web Download. Philadelphia: Linguistic Data Consortium, 2002.
• 385 WSJ articles from Penn Treebank, representing over 176,000
words of text. ~14% were double-tagged.
• Document length: 31 to 2124 words; average of 458.14 words
• Average # EDUs per document: 56.59.
• Average # words per EDU: 8.1.
• Nature of articles: general news, financial, business, cultural
reviews, editorials
• Intended users: developers of automatic text processing
systems

• Attribution: attribution, attribution-negative
• Background: background, circumstance
• Cause: cause, result, consequence
• Comparison: comparison, preference, analogy, proportion
• Condition: condition, hypothetical, contingency, otherwise
• Contrast: contrast, concession, antithesis
• Elaboration: elaboration-additional, elaboration-general-specific
• Enablement: purpose, enablement
• Evaluation: evaluation, interpretation, conclusion, comment
• Explanation: evidence, explanation-argumentative, reason
• Joint: list, disjunction
• Manner-Means: manner, means
• Topic-Comment: problem-solution, question-answer, topic-comment..
• Summary: summary, restatement
• Temporal: temporal-before, temporal-after, temporal-same-time..
• Topic Change: topic-shift, topic-drift
Relations
78 relations in 16 groups:
Tagge
rs
Units Spans Nucle
ar-ity
Rela-
tions
#
docs
Avg #
EDUs
A, E 0.98 0.90 0.84 0.76 7 57.7
A, B 1.00 0.93 0.88 0.79 5 58.2
A, C 0.95 0.84 0.78 0.68 4 116.
5
A, F 0.95 0.78 0.69 0.69 4 26.5
A, D 1.00 0.87 0.80 0.72 4 23.3
Carlson, Lynn; Daniel Marcu and Mary Ellen Okurowski. RST Discourse Treebank
LDC2002T07. Web Download. Philadelphia: Linguistic Data Consortium, 2002.
IAA

wsj_1111
(26)
(17) (18) (21)
19-20
17-21 22-26
explanation-
argumentative
consequence-s
same-unit
17-26 elaboration-additional
(16)
example
17-18
attribution
19-21
attribution
(19) (20)
24-25
purpose
(22) (23) (25)
(24)
attribution
embedded elaboration-object-
attribute-embedded
22-23
22-25
[Still, analysts don’t expect
the buy-back to significantly
affect per-share earnings in
the short term.]16 [“The
impact won’t be that
great,”]17 [said Graeme
Lidgerwood of First Boston
Corp.]18 [This is in part
because of the effect]19 [of
having to average the number
of shares outstanding,]20 [she
said.]21 [In addition,]22 [Mrs.
Lidgerwood said,]23 [Norfolk
is likely to draw down its cash
initially]24 [to finance the
purchases]25 [and thus forfeit
some interest income.]26

RST Signalling Corpus
Das, Debopam and Maite Taboada (2015) RST Signalling Corpus. LDC.
Das – Taboada, 2015
Hypothesis: There are no (or very few) ‘implicit’ relations.

RST - Present and Future
0 Annotated corpora in different languages
RST workshop
0 2007 - Collocated with the 16th Intercâmbio de Pesquisas em Linguística Aplicada (InPLA), Pontifícia
Universidade Católica de São Paulo, Brazil.
0 2009 - Named "Brazilian RST Meeting", collocated with The 7th Brazilian Symposium in Information
and Human Language Technology (STIL), in São Carlos, Brazil.
0 2011 - 3rd Workshop, "RST and Discourse Studies" was held with The 8th Brazilian Symposium in
Information and Human Language Technology (STIL), in Cuiabá, Brazil.
0 http://corpus.iingen.unam.mx/RST_2011
0 2013 - The 4th workshop "RST and Discourse Studies" was also held within STIL (2013) in
Fortaleza, Brazil.
0 http://encontrorst2013.wix.com/encontro-rst-2013
0 2015 - The 5th workshop "RST and Discourse Studies", collocated with The 31st Conference of the
Spanish Association for Natural Language Processing. Alicante, Spain.
0 https://sites.google.com/site/workshoprst2015/

0 СМАЙЛИК, ЧТО ЭТО МЫ НАКОНЕЦ ЗАКОНЧИЛИ!

Segmented Discourse
Representation Theory (SDRT)
0 Comes from formal semantics, initially from formal sentence
semantic and then systematically extended to discourse
0 interested in ‘semantic scopes’ – some sort of bijection between one
structure to another between the relations
0 Recursive Structure (SDRS) - constituted by discourse units linked
by discourse relations where a discourse unit is either an elementary
discourse unit (EDU) or a complex discourse unit (CDU)
0 DUs can be embedded one into another (EDUs)
0 CDUs may not partially overlap
Asher, N. and A. Lascarides (2003). Logics of conversation. Cambridge University Press

Segmented Discourse
A - is a set of labels (units)
Last - is a label A (intuitively last clause)
F - is a formula which assigns each member of
A a member of a formula of the SDRS language
SDRS =
<A,F,Last>
A discourse structure in SDRT:

Segmented Discourse
0 Graph representation - An SDRS can be seen as a
directed acyclic graph where each DU is a vertex
and
0 Directed labelled edges for rhetorical
relations
0 Directed unlabelled edges link CDUs to their
content.
0 Two kind of relations:
0 Coordinating, e.g. Result, Narration (drawn
horizontally)
0 Subordinating, e.g. Elaboration, Explanation,
Constrast (drawn vertically) after Polanyi 1985

Segmented Discourse
Principles of natural selection. The theory of
natural selection, such as it was initially
described by Charles Darwin, is based on
three principles: 1. Principle of variation 2.
Principle of adaptation 3. Principle of
heredity

Segmented Discourse
0 Right frontier constraint (RFC) - A discourse constituent must be attached on the
right frontier of the ongoing discourse.
0 Given a tree or a graph, an attachment cannot jump to a constituent on the left of the current
one.
Polanyi (1985), Webber (1988) Asher (1993), Asher - Lascarides (2003)
a. John had a great evening last night.
b. He had a great meal.
c. He ate salmon.
d. He devoured lots of cheese.
e. He then won a dancing competition.
f. It (the salmon) was a beautiful pink.
It was a
beautiful pink.

Segmented Discourse
• RFC requires that new discourse constituents
cannot be integrated anywhere in the tree
representing preceding discourse but only in
a well-defined area situated at the Right
Frontier of the tree.
• The constituents situated in this area are said
to be open for attachment while all the others
are said to be closed (Polanyi, 1988; Asher &
Lascarides, 2003).
• Constituents that are targets of a coordinating
relation are not available for further
attachment
From Vergez-Couret et al. 2012

RST SDRT PDTB
based on formal
semantics
tries to capture the intensions
of the author as they are
judged by the human annotator
based on lexically
grounded relations
long-distance attachments
are possible
are possible
are not possible
Trees Graph No assumption on text structure

SDRT corpora
0 DISCOR
Brian Reese, Julie Hunter, Nicholas Asher, Pascal Denis and Jason Baldridge. Reference
Manual for the Analysis and Annotation of Rhetorical Structure. 2007
0 ANNODIS
Afantenos S. D., Asher N., Benamara F., Bras M., Fabre C., Ho-Dac L.-M., Le Draoulec A.
Muller P., Péry-Woodley M.-P., Prévot L., Rebeyrolle J., Tanguy L., Vergez-Couret M., Vieu
L. (2012). An empirical resource for discovering cognitive principles of discourse
organization: the ANNODIS corpus. LREC 2012, Istanbul, Turkey, July 2012.
0 GEOPO
Lydia-Mai Ho-Dac : La position initiale dans l'organisation du discours : une exploration
en corpus, thèse de doctorat, Université de Toulouse-Le Mirail, novembre 2007
0 STAC
Stergos Afantenos Eric Kow Nicholas Asher Jérémy Perret. Discourse parsing for multi-
party chat dialogues. Proceedings of EMNLP-2015, pp. 928–937, Lisbon.

DISCOR (Discourse Structure and
Coreference Resolution)
0 Goal: to test hypotheses about the interaction between
discourse structure and the resolution of anaphoric links.
0 Data: MUC and ACE corpora (because already have
coreference)
0 14 relations
(cf. 78 in RST corpus)

DISCOR (Discourse Structure and
Coreference Resolution)
0 ca. clauses, but not if embedded, e.g. do not segment wh-clause in
Privately held Arnold, which had about $750 million in billings and $90.7 million in
revenue last year, handles advertising for such major corporations as McDonald’s Corp., ...
0 nonrestrictive relative clauses introduce EDUs, unless doing so results in a
discontinuous EDU
0 Many, but certainly not all, cases of syntactic subordination introduce a new
elementary discourse unit. Complements of verbs of communication, for
example, introduce EDUs (say, note, announce, etc).
0 appositions are EDUs
Brian Reese, Julie Hunter, Nicholas Asher, Pascal Denis and Jason Baldridge. Reference Manual for the Analysis and Annotation of
Rhetorical Structure. 2007

ANNODIS - (ANNOtation DIScursive)
0 project 2007-2010 – texts annotation and parsing
0 people: Nicholas Asher, Farah Benamara, Philippe Muller, Laure Vieu, Stergos Afantenos,
etc.
0 an annotated corpus of French written texts (news, wikipedia, linguistic papers, reports)
for the study of discourse organization – 86 stories, 687,000 words,
0 3 years of annotation
0 3188 Elementary Discourse Units (EDU) and 1395 Complex Discourse Units (CDU) linked
by 3355 rhetorical relations (e.g. contrast, elaboration, result, attribution, etc.)
0 text annotated with various discourse phenomena
0 bottom-up approach: applying a compositional and logical model of discourse organization
(SDRT)
0 top-down approach: starts from the text as a whole and focuses on the identification of
configurations of cues signalling higher-level text segments, in an attempt to address the interplay
of continuity and discontinuity within discourse (annotation of Enumerative Structures and
Topical Chains)
0 annotation tool Glozz (will be shown by the STAC corpus)

alternation
attribution
background
comment
continuation
contrast
entity-elaboration (e-elab)
elaboration
explanation
flashback
frame
goal
meta-relation
narration
parallel
result
temploc
• 17 relations
• were chosen because they are more
or less common to all the theories
of discourse, or correspond to well-
defined subgroups in fine-grained
theories
• The intermediate level of
granularity was chosen as a
compromise between
informativeness and reliability of
the annotation process. It
corresponds to the level chosen in
the PDTB, and a coarse-grained
RST.
ANNODIS - (ANNOtation DIScursive)
Muller P., Vergez M., Prevot L., Asher N., Benamara F., Bras M., Le
Draoulec A. & Vieu L. (2012). Manuel d'annotation en relations de
discours du projet Annodis, Carnets de grammaire 21, CLLE-ERSS.
DISCOR
French

SDRT corpora - STAC
0 running project: 2011-
2016
0 ca 1100 negotiation
dialogues, short texts
(in English!) – much larger
corpus than ANNODIS:
0 refined annotation
tool Glozz
CDU
0 inter-annotator agreement better than for ANNODIS (basically
because turns are annotated as EDU without further division)

Discourse Corpra about the subject of semantics

- domain level acts
offer : I’ll give you 2 clay for a rock
counteroffer : How about 2 clay for a wheat ?
accept : OK, it’s a deal.
refusal : I don’t think so.
has-resource : I have wheat
strategic comment : joel fancies a bit of your clay
other (non relevant for negotiation)
- relational rhetorical annotation
Two-level annotation:
• non-treelike structures certainly exist
• long distance crossing dependencies
reflected in parsers output
STAC

References
0 Asher, Nicholas (1993). Reference to Abstract Objects in Discourse. Dordrecht: Kluwer.
0 Asher, Nicholas and Lascarides, Alex (2003). Logics of Conversation. Cambridge: Cambridge U.P.
0 Asher, Nicholas . Discourse Parsing for Text and Dialogue [2013]. MELODI. IRIT, CNRS
0 Asher, Nicholas (with Antoine Venant, Philippe Muller, Pascal Denis, Stergos Afantenos, Farah Benamara). Annotating and
learning models of discourse structure [2015]
0 Stergos D. Afantenos and Nicholas Asher [2010] Testing SDRT’s Right Frontier . Proceedings of Coling 2010, pages 1–9,
0 Jacques Jayez. [2008]. The Right Frontier Constraint. ENS-LSH and L 2 C2
0 Lascarides, A. and N. Asher [2007] Segmented Discourse Representation Theory: Dynamic Semantics with Discourse Structure,
in H. Bunt and R. Muskens (eds.) Computing Meaning: Volume 3, pp87--124, Springer.
0 Lascarides, A. Semantics and Pragmatics of NLP Segmented Discourse Representation Theory. SPNLP. (slides)
0 Nicholas Asher, Laure Vieu. Subordinating and coordinating discourse relations. Lingua 115 (2005), 591–610.
0 Philippe Muller, S. Afantenos, Pascal D., N, Asher. Constrained decoding for text-level discourse parsing [2012], Proceedings of
COLING 2012: Technical Papers, pages 1883–1900,
0 Polanyi, Livia (1985). A theory of discourse structure and discourse coherence. Proceedings of the 21st Meeting of the Chicago
Linguistics Society.
0 Webber, Bonnie (1988). Discourse deixis and discourse processing. Technical Report MS-CIS-86-74, Department of Computer
and Information Science, University of Pennsylvania.

with anaphora and bridging
0 Asher, Nicholas and Lascarides, Alex (1997). Bridging.
Journal of Semantics
0 Cornish, Francis (1999). Anaphora, Discourse and
Understanding. Evidence
from English and French. Oxford: Oxford U.P.

Manuals, etc.
0 Brian Reese, Julie Hunter, Nicholas Asher, Pascal Denis
and Jason Baldridge. [2007] DISCOR: Reference
Manual for the Analysis and Annotation of Rhetorical
Structure (Version 1.0)

0Halliday and Hasan – Cohesion in English
0RST – Rhetorical Structure Theory
0SDRT - Segmented Discourse Representation Theory
0PDTB – Penn Discourse Treebank and PDTB-like
0Discourse Graphbank
0CCR – Cognitive approach to coherence relations

Jerry R. Hobbs’ theory
Hobbs, Jerry R. 1985. On the coherence and structure of discourse. Technical Report 85-37,
Center for the Study of Language and Information (CSLI), Stanford, CA.
Hobbs, Jerry R. 1985. Literature and Cognition., CSLI, 1990
Theory of discourse coherence
Theory of discourse interpretation
• … hold between segments of a discourse
• … are defined in terms of propositions that can be inferred from the assertions of
discourse segments (in terms of formal logic)
• The assertion of a clausal discourse segment is, roughly, what is predicated by
the main verb
Coherence relations, e.g. contrast,
elaboration, parallel…

Jerry R. Hobbs’ theory
Occasion Relation
Cause
Evaluation Relation
Ground-Figure
----------- Explanation Relations
Expansion Relations
----------- Parallel
----------- ----------- Elaboration
----------- Exemplification:
----------- Contrast
----------- Violated Expectation
Parallel:
Infer p(al, a2, ...) from the assertion of So and p(bl, b2, ...) from
the assertion of S1, where ai and bi are similar, for all i.
Set stack A empty and set link variable P to T.
p is set, al is stack A, a2 is the value empty, bl is link variable P, b2 is the value T
Elaboration: Infer the same proposition P from the
assertions of S0 and S1.
go(Agent: you, Goal: x, Path: First St., Measure: y) for some x and y.
go (Agent: you, Goal: A St., Path: First St., Measure: 3 blks)
Go down First Street. Just follow First Street three
blocks to A Street.
If we assume that x is A Street and y is 3 blocks, then the two are
identical and serve as the proposition P in the definition.

0 motivated by Hobbs
0 vs. RST: polemics to tree structures
0 135 texts annotated with coherence relations (WSJ and AP Newswire)
0 DUs – mainly clauses
0 delimiting discourse segments (because, and, for example, periods, semicolons,
commas, etc.)
0 treat attributions (John said that . . .) as discourse segments
0 Discourse Segment Groupings
0 groups: e.g. attributed to the same source, topically related, might also be subgroups,
consisting of several discourse segments each (making a partially hierarchical
structure for the text)
0 groups were allowed to partially overlapped, but this was not used by annotators
Discourse Graphbank
Wolf, F. and Gibson, E.: 2005, Representing discourse coherence: A corpus based study,
Computational Linguistics 31(2), 249–287.

Discourse Graphbank
Occasion Relation
Cause
Evaluation Relation
Ground-Figure
Explanation Relations
Expansion Relations
Background
Parallel
Elaboration
Exemplification:
Contrast
Violated Expectation
Temporal sequence
Cause – Effect
Condition
Elaboration
Example
Generalization
Similarity
Contrast
Violated Expectation
Attribution
Same
asymmetrical (directed)
symmetrical (undirected)
following (Mann and Thompson
1988; Marcu 2000)
RELATIONS

Discourse Graphbank
(ce=Cause-Effect; attr=Attribution; elab=Elaboration; sim=Similarity.)
0. Farm prices in October edged up 0.7% from September
1. as raw milk prices continued their rise,
2. the Agriculture Department said.
3. Milk sold to the nation's dairy plants and dealers averaged $14.50 for
each hundred pounds,
4. up 50 cents from September and up $1.50 from October 1988,
5. the department said.
discourse structure is complex and requires
a representation in terms of chain graphs
REPRESENTATION

DGB vs RST (or graphs vs. trees)
Markus Egg and Gisela Redeker, How Complex is Discourse Structure? LREC 2010
• crossed dependencies
• 41.22% of the segments have
multiple parents (W&G 2005)
DGB: Egg and Redeker:
distinguish the complexity inherent in the data and the
one arising from specific design choices in W&G’s
annotation (on 14 texts from DGB)
Result: Many non-tree-like
dependencies may be converted to
tree-like ones.

CCR (Cognitive approach to
coherence relations)
coherence relations are considered as cognitive entities
(coherence relations and their linguistic marking affect the
cognitive representation of discourse, e.g. text understanding,
they are psychological entities rather than merely an analytic
tool)
Sanders, T. J. M., W. P. M. Spooren, and L. G. M. Noordman (1992). Toward a taxonomy of
coherence relations. Discourse Processes 15, 1–35.
Proved by series of psycholinguistic experiments (Spooren, 1989;
Haberlandt, 1982; Sanders 1986, e.g. linguistic marking appears to
lead to faster processing of the following discourse segment).

CCR (Sanders et al.)
Investigating coherence relations
relational account language-based account
0 Focus on the meaning of
the relation and not on
the meaning of each
specific segment
0 Classification in terms of
cognitive primitives
0 Focus not so much on
relations themselves but
rather on a study of
linguistic devices that are
used to signal relations
explicitly

Although he worked hard, he failed the exam.
Four cognitive primitives
discourse segment S1 discourse segment S2
proposition P proposition Q
expresses
expresses

0 basic operation (Additive –
Temporal – Causal /
Conditional)
originally (Sanders et al. 1992) only additive (weakly connected)
vs. causal (strongly connected) were distinguished
She got wet and her friend got wet too.
P Q
P & Q
She got wet because it rained.
P
Q
P  Q
There was a lot of rain. Later, storms came in.
P Q
P, later Q
From Sanders 2015 (Fribourg)

Conditional)
0 source of coherence
(semantic/ pragmatic)
P
Q
She is not at home, because her car is not there
P
Q
A relation is objective (semantic) when both segments
happen in the real world and are thus facts. The speaker’s
opinion is not reflected in the relation.
A relation is subjective (pragmatic) if one or both segments express
an opinion, argument, claim or conclusion of the speaker.
illocutions
locutions

Conditional)
0 source of coherence (semantic/
pragmatic)
0 order of segments
(basic/non-basic)
Q
P
P
Q
A relation with a basic order has an antecedent,
followed by a consequent. The antecedent is the
cause of the argument, the consequent is the
consequence or the claim.
A relation with a non-basic order has a consequent which precedes
the antecedent. The cause or the claim thus precedes the cause of
the argument.

Conditional)
0 source of coherence (semantic/
pragmatic)
0 order of segments (basic/non-
basic)
0 polarity (positive/negative)
S1=Q
A relation is positive if the two discourse segments S1 and S2
function in the basic operation as antecedent (P) and
consequent (Q) respectively.
S2=P
not-S2=Q
not-S1=P
A relation is negative if not S1 and S2, but their
negative counterparts, not-S1 and not-S2, function
in the basic operation.

References
0 Sanders, T.J.M. & Knott, A. (1998). The classification of coherence
relations and their linguistic markers: An exploration of two
languages. Journal of pragmatics, 30, (pp. 135-175) (41 p.).
0 Ted Sanders. On the Cognitive and Text-Analytic Status of
Coherence Relations.

DiscAn corpora
Towards a discourse annotation system for Dutch language corpora
Sanders, T.J.M., Vis, K. & Broeder, D. (2012). Project notes on the Dutch project DiscAn. Eighth Joint ACL - ISO
Workshop on Interoperable Semantic Annotation Pisa.
0 DiscAn project – integrating existing corpora of Dutch discourse
phenomena in the CLARIN infrastructure
0 set of corpus analyses has been standardized (both in terms of raw data --
the texts -- and analyses) and opened up for further scientific research
0 text have been annotated for discourse phenomena during last ca 15
years at several universities in the Netherlands and Belgium
0 format: Excel tables, doc files, SPSS files etc.  brat?
0 visualization ANNIS

DiscAn subcorpora
0 Degand 2.2: compiled for a study of the causal connectives
"aangezien", "want" and "omdat" in Dutch news, 143 cases from a
Dutch newspaper (NRC Handelsblad from 1994)
0 PanderMaatSanders: causal connectives "daardoor", "daarom" and
"dus" in Dutch news
0 Persoon Corpus: causal connectives "want" and "omdat" in Dutch
spontaneous conversations, from the Corpus of Spoken Dutch
0 SandersSpooren Corpus: causal connectives "want" and "omdat" in
several types of discourse
Degand, L. (2001). Form and function of causation: A theoretical and empirical investigation of causal
constructions in Dutch. Leuven: Peeters.
Persoon, I., Sanders, T., Quené, H. & A. Verhagen (2010). Een coördinerende omdat-constructie in gesproken
Nederlands? Tekstlinguïstische en prosodische aspecten. Nederlandse Taalkunde, 15, 259-282.

DiscAn subcorpora
From Sanders et al .2012

DiscAn Annotation
Four annotated categories:
0 Polarity (negative/positive)
0 Basic operation
0 source of coherence (objective/subjective)
0 order
Rel
causal
non-causal
conditional
non-conditional
additive
temporal
BUT! Temporal and non-causal negative relations do not differ in source of
coherence, because they are an objective representation of reality by nature.
BUT! Order for additive relations is not marked,
because they are symmetric
Temporal relations can have a chronological order (basic order)
and non-chronological (non-basic). Two events can also happen
simultaneously. In this case, order is not annotated.

0 Halliday and Hasan – Cohesion in English
0 RST – Rhetorical Structure Theory
0 SDRT - Segmented Discourse Representation
Theory
0 PDTB – Penn Discourse Treebank and PDTB-like
0 Discourse Graphbank
0 CCR – Cognitive approach to coherence relations

0 How to compare the theories / approaches /
structures / corpora …?
0 Can we translate from one corpus to another,
thus extending the range of data available for
performing automated tasks ?

Comparative attempts
0 Taboada, M. (2015) Rhetorical relations are relations of coherence: What discourse
coherence means, and how we can find it. Conference of the COST Action TextLink:
Structuring Discourse in Multilingual Europe . Louvain-la-Neuve, Belgium. January
2015. (Plenary presentation)
0 Rehbein, Ines, Merel Scholman and Vera Demberg. Annotating discourse relations in
spoken language: A comparison of the PDTB and CCR frameworks. DiSpoL meeting of
TextLink, Saarbrucken, September, 2015.
0 Anja Nedoluzhko, Ekaterina Lapshinova, Kerstin Kunz: PDT vs. GECCo (2015: LAW-IX
NAACL, TextLink meetings in Louvain-la-Neuve and Saarbrucken, LREC subm.)
0 Venant –Asher – Muller – Pascal – Afantenos: Expressivity and comparison of models of
discourse structures, Sigdial 2013.
Creating common standard
0 ISO standard for the annotation of semantic relations in discourse
0 applying Sanders’ dimensions (April 2015, Fribourg)

Taboada, M. (2015) Rhetorical relations are relations of coherence: What discourse
coherence means, and how we can find it. Conference of the COST Action TextLink: Structuring
Discourse in Multilingual Europe . Louvain-la-Neuve, Belgium. January 2015.
https://www.rese
archgate.net/publ
ication/25430212
1_Coherence_relat
ions_Towards_a_g
eneral_specificatio
n
Bateman, J. and
Rondhuis, K.J.
(1997) 'Coherence
Relations:
Towards a
General
Specification' ,
Discourse
Processes 24: 3-
49

0 200 400 600 800
coreference
bridging/lexical cohesion
ellipsis
DSDs
PDT GECCo
Anja Nedoluzhko, Ekaterina
Lapshinova, Kerstin Kunz: PDT vs.
GECCo (2015)
Prague
Dependency
Treebank (Prague)
German-English
Contrasts in Cohesion
(Saarland Uni)

Rehbein, Ines, Merel Scholman and Vera Demberg. Annotating discourse relations in spoken
language: A comparison of the PDTB and CCR frameworks. DiSpoL meeting of TextLink,
Saarbrucken, September, 2015.
0 idea: Use CCR (Cognitive approach to Coherence Relations, Sanders,
Spooren & Noordman 1992) as an intermediate language between different
frameworks
0 question: To what extent can PDTB relations be analysed consistently using
CCR dimensions?
0 2 annotators analyzed 1197 relations independently using PDTB 3.0 and
CCR, respectively
0 Annotations mapped onto each other to investigate consistency of
relation meanings across theories

Rehbein, Ines, Merel Scholman and Vera Demberg. Annotating discourse relations in spoken
language: A comparison of the PDTB and CCR frameworks. DiSpoL meeting of TextLink,
Saarbrucken, September, 2015.
0 Overall, 69% of the PDTB relations were
consistently categorized as
belonging to the target CCR class.
0 Analysis of random sample of 50
disagreements: 48% of disagreements
due to differences between the theories.
0 Other disagreements due to difference in
segmentation or interpretation
of relation (14%) and to annotation errors
(38%)

0 Use CCR as an intermediate language
Ted Sanders, Let's try to make annotation systems communicate – towards a systematic approach of
coherence relations . Fribourg, 2015.
from Sanders et al. 1992
from Sanders et al. 2015

Ted Sanders, Let's try to make annotation systems communicate – towards a systematic approach of
coherence relations . Fribourg, 2015.
from Sanders, 2015

immediately
• Attribution (π1,[π2, π3, π4])
Restatement(π2, π3)
Explication(π2, π4)
Venant –Asher – Muller – Pascal – Afantenos: Expressivity and comparison of
models of discourse structures, Sigdial 2013.
[Interprovincial Pipe Line Co. said]π1 [it will delay a proposed two-step, 830 million
dollar [(US$705.6 million)]π3 expansion of its system]π2 [because Canada’s output of
crude oil is shrinking.]π4
If RST is interpreted indirectly, we should be
able to describe how the unpacking of the
real arguments in the tree structure works!!!
interested
in relations
interested
in scopes
Mixed Nuclearity Principle
• NS relations only transmit nucleus argument to
a parent relation.
• Restatement(π2, π3) ∧ Explanation(π2, π4) ∧
Attribution(π1, π2)

0 Comparisons require a language expressive enough to express semantic scopes
for all discourse theories and interpretative mappings from the different
structures into this language.
0 Venant et al. 2013 propose a language (formal semantic description) in which all
theories can be described. Then we can look and compare the commonalities
and restrictions, incl. decoding one from the other
0 Immediate interpretation of a RS-Tree can be decoded back to a SDRS (it gives
lots of CDUs)
Venant –Asher – Muller – Pascal – Afantenos: Expressivity and comparison of
models of discourse structures, Sigdial 2013.
CDU
EDU
EDU EDU
0 Relaxed Nuclearity Principle  RS Tree
can be decoded into a set of SDRSs
0 The same with dependency trees
0 DP and RST-tress are both underspecified
version of a fully specified structure 

0 Once you choose one of these theories you get the
corpus that looks completely different than another
corpus in another theory. RST corpus looks
completely different than PDTB
0 What we’re interested in is the semantics.

Rashmi Prasad and Harry Bunt. Semantic Relations in Discourse: The Current State of ISO
24617-8. Proceedings of the 11th Joint ACL-ISO Workshop on Interoperable Semantic Annotation
(ISA-11). 2015
compared approaches
created an ISO, very close to
current PDTB 3.0 version
0 Types of objects connected by discourse relations,
0 explicit connective/implicit/AltLex
0 representation of discourse structure
0 semantic description of discourse relations
0 pragmatic variants of discourse relations
0 hierarchical classification of discourse relations
0 representation of (a)symmetry of relations
0 relative importance of arguments for text meaning/structure
RST
DGB
PDTB
SDRT
CCR
H&H
ISO 24617-8

Constraints are placed on the semantic nature of
arguments rather than on their syntactic form. An
argument of a discourse relation must denote a certain
type of abstract object. The ISO scheme remains neutral
on this issue and does not specify any constraints on the
extent or adjacency of argument realizations ISO 24617-8
PDTB 2.0 - sentences (or where were connectives), PDTB
3.0 - also intrasentential, incl. free adjuncts, VP-
coordination without connectives, etc.
PDTB
embedded, min. clause
SDRT
sentences
H&H
A span of text, usually a clause, but in general
ranging from minimally a (nominalization) NP
to maximally a sentence.
RST
=RST
DGB
text segments,
minimally clauses
CCR
Types of objects connected by discourse relations

as it is
ISO 24617-8
explicit, implicit, AltLexes
PDTB
?
SDRT
explicit
H&H
all elements having a connecting function
RST-signaling
-
DGB
?
CCR
Explicit connectives/Implicit/AltLex
explicit
PDT

ISO takes a pre-theoretical stance involving low-level
annotation of discourse; individual relations can then
be annotated further to project a higher-level tree or
graph structure, depending on one’s theoretical
preferences
ISO
24617-8
relations along with their arguments are annotated
without being combined to form a structure that
encompasses the entire text
PDTB
allow directed acyclic
graphs that allow for
multiple parents, but
not for crossing
SDRT
relations along with their arguments
are annotated without being combined
to form a structure that encompasses
the entire text
H&H
a tree representation to subsume the
complete text of the discourse
RST
allow general graphs
that allow multiple
parents and crossing
DGB
relations along with their
arguments are annotated without
being combined to form a structure
that encompasses the entire text
CCR
Representation of Discourse Structure

in ‘informational’ terms
ISO
24617-8
in ‘informational’ terms, i.e., in terms of the content of the
arguments
PDTB
in ‘informational’
terms, i.e., in terms of
the content of the
arguments
SDRT
in ‘informational’ terms, i.e., in terms of
the content of the arguments
H&H
in terms of the intended effects on the
hearer/reader
RST
in ‘informational’
terms, i.e., in terms of
the content of the
arguments
DGB
in ‘informational’ terms, i.e., in
terms of the content of the
arguments
CCR
Semantic Description of Discourse Relations

semantic-pragmatic distinction, for all relation types
However, the ISO scheme does not encode this
distinction on the relation, but on the arguments of the
relation
ISO
24617-8
semantic-pragmatic, for some relation types
PDTB
content -- metatalk
SDRT
internal-external distinction
H&H
semantic-pragmatic, for some relation types
(volutional - non volutional) subject matter
relations vs. presentational relations
RST
do NOT distinguish
pragmatic
DGB
semantic-pragmatic distinction, for
all relation types
Subjective – Objective
CCR
Pragmatic Variants of Discourse Relations

a ‘flat’ set of core relations
ISO
24617-8
yes, 3 levels
PDTB
?
SDRT
no, only basic groups
H&H
yes, 2 levels, big groups: Causal-conditional,
Contrastive, Additive
RST
?
DGB
no, 4 criteria + additional features
CCR
Hierarchical Classification of Discourse Relations
yes, 2 levels
PDT

annotations abstract over the linear ordering for
argument realizations, since this is not a semantic
distinction. Instead, asymmetry is represented by
specifying the argument roles in the definition of each
relation.
ISO
24617-8
Argument spans are named Arg1 and Arg2 according to
syntactic criteria, including linear order, and the
asymmetrical relations are defined in terms of the Arg1
and Arg2 labels (for example, the relation Cause:Reason
has Arg2 as the cause and Arg1 as the effect, while the
relation Cause:Result has Arg1 as the cause and Arg2 as
the effect.
PDTB
yes
SDRT
no relations
H&H
yes
RST
yes
DGB
The argument span ordering is one of the ‘cognitive’
primitives underlying the scheme. The relation Cause-
Consequence captures the ‘basic’ order for the semantic
causal relation, with the cause appearing before the
effect, whereas the relation Consequence-Cause is used
for the reversed order of the arguments.
CCR
Representation of (a)symmetry of Relations
direction of the relation
arrows in dependency
trees
PDT

the relative role of arguments for the text (meaning or
structure) as a whole is not represented directly
ISO
24617-8
Argument spans are named Arg1 and Arg2 according to
syntactic criteria, including linear order, and the
asymmetrical relations are defined in terms of the Arg1
and Arg2 labels (for example, the relation Cause:Reason
has Arg2 as the cause and Arg1 as the effect, while the
relation Cause:Result has Arg1 as the cause and Arg2 as
the effect.
PDTB
yes, classifies a relation as ‘subordinating’ or
‘coordinating’, depending on what structural
configuration the arguments create in the discourse
graph (Asher and Vieu, 2005)
SDRT
no arguments
H&H
yes (nuclear-satellite)
RST
yes, ‘dominance’, deriving
a single assertion from a
discourse relation that
connects two segments,
and distinguishing
relations in terms of how
this single assertion
should be derived. In
subordinating relations,
in particular, the
assertion associated with
the relation is obtained
from the ‘dominant’
segment, as specified in
the relation definitions.
DGB
no
CCR
Relative Importance of Arguments for Text meaning/structure
direction of the relation
arrows in dependency
trees
PDT

Acknowledgements
0 I used the materials published in articles and slides by
Manfrede Stede, Maite Taboada, Annemarie Friedrich, Lynn
Carlson, Daniel Marcu, Ines Rehbein, Merel Scholman, Vera
Demberg, Nicholas Asher, Jacques Jayez, Alex Lascarides,
Markus Egg, Gisela Redeker, Ted Sanders, Merel Scholman,
etc.
0 Many thanks to my discourse colleagues (Eva Hajičová,
Lucie Poláková, Pavlína Jínová, Šárka Zikánová, Jiří
Mírovský, etc.)
0 Grant ...

References
0 Marianne Vergez-Couret, Laurent Prévot et Myriam Bras, «
How Different Information Sources Interact in the
Interpretation of Interleaved Discourse: The Case of Two-
Step Enumerative Structures », Discours [En ligne], 11 |
2012, mis en ligne le 23 décembre 2012, consulté le 16
décembre 2015. URL : http://discours.revues.org/8743 ;
DOI : 10.4000/discours.8743

распечатать и иметь там
0 RST manual
0 Wolf, F. and Gibson, E.: 2005, Representing discourse
coherence: A corpus based study, Computational
Linguistics 31(2), 249–287 (DGB)
0 M. Scholman – annotation manual for DiscAn

Discourse Corpra about the subject of semantics

More Related Content

Similar to Discourse Corpra about the subject of semantics (20)

Recently uploaded (20)

Discourse Corpra about the subject of semantics