Natural Language Processing and Language Learning

Natural Language Processing and Language Learning
To appear 2012 in Encyclopedia of Applied Linguistics, edited by Carol A. Chapelle. Wiley Blackwell
Detmar Meurers
Universität Tübingen
dm@sfs.uni-tuebingen.de
As a relatively young field of research and development started by work on cryptanalysis and
machine translation around 50 years ago, Natural Language Processing (NLP) is concerned
with the automated processing of human language. It addresses the analysis and generation
of written and spoken language, though speech processing is often regarded as a separate
subfield. NLP emphasizes processing and applications and as such can be seen as the applied
side of Computational Linguistics, the interdisciplinary field of research concerned with for-
mal analysis and modeling of language and its applications at the intersection of Linguistics,
Computer Science, and Psychology. In terms of the language aspects dealt with in NLP,
traditionally lexical, morphological and syntactic aspects of language were at the center of
attention, but aspects of meaning, discourse, and the relation to the extra-linguistic context
have become increasingly prominent in the last decade. A good introduction and overview
of the field is provided in Jurafsky & Martin (2009).
This article explores the relevance and uses of NLP in the context of language learning,
focusing on written language. As a concise characterization of this emergent subfield, the
discussion will focus on motivating the relevance, characterizing the techniques, and delin-
eating the uses of NLP; more historical background and discussion can be found in Nerbonne
(2003) and Heift & Schulze (2007).
One can distinguish two broad uses of NLP in this context: On the one hand, NLP can be used
to analyze learner language, i.e., words, sentences, or texts produced by language learners.
This includes the development of NLP techniques for the analysis of learner language by
tutoring systems in Intelligent Computer-Assisted Language Learning (ICALL, cf. Heift,
this volume), automated scoring in language testing, as well as the analysis and annotation
of learner corpora (cf. Granger, this volume).
On the other hand, NLP for the analysis of native language can also play an important role
in the language learning context. Applications in this second domain support the search for
and the enhanced presentation of native language reading material for language learners, they
provide targeted access to relevant examples from native language corpora, or they support
the generation of exercises, games, and tests based on native language materials.
1 NLP and the Analysis of Learner Language
Intelligent Language Tutoring Systems (ILTS) use NLP to provide individualized feed-
back to learners working on activities, usually in the form of workbook-style exercises as
in E-Tutor (Heift, 2010), Robo-Sensei (Nagata, 2009), and TAGARELA (Amaral & Meur-
ers, 2011). The NLP analysis may also be used to individually adjust the sequencing of the
material and to update the learner model (cf. Schulze, this volume). Typically the focus of
1

the analysis is on form errors made by the learner, even though in principle feedback can
also highlight correctly used forms or target aspects of meaning or the appropriateness of a
learner response given the input provided by an exercise.
What motivates the use of NLP in a tutoring system? To be able to provide feedback and
keep track of abilities in a learner model, an ILTS must obtain information about the student’s
abilities. How this can be done depends directly on the nature of the activity and the learner
responses they support, i.e., the ways in which they require the learner to produce language
or interact with the system. The interaction between the input given to a learner to carry out
a task and the learner response is an important topic in language assessment (cf. Bachman &
Palmer, 1996) and it arguably is crucial for determining the system analysis requirements for
different activity types in the ILTS context.
For exercises which explicitly or implicitly require the learner to provide responses us-
ing forms from a small, predefined set, it often is possible to anticipate all potential well-
formed and ill-formed learner responses, or at least the most common ones given a par-
ticular learner population. The intended system feedback for each case can then be ex-
plicitly specified for each potential response. In such a setup, knowledge about language
and learners is exclusively expressed in this extensional mapping provided offline when the
exercise is created. Sometimes targets are allowed to include regular expressions (http:
//en.wikipedia.org/wiki/Regular_expression) to support a more compact specification. The
online process of comparing an actual learner response to the anticipated targets is a sim-
ple string comparison requiring no linguistic knowledge and thus no NLP. Correspondingly,
the quiz options of general Course Management Systems (Moodle, ILIAS, etc.) can support
such language exercises just as they support quizzes for math, geography, or other subjects.
The same is true of the general web tools underlying the many quiz pages on the web (e.g.,
http://www.eslcafe.com/quiz/). Tools such as Hot Potatoes (http://hotpot.uvic.ca/) make it
easier to specify some exercise forms commonly used in language teaching, but also use the
same general processing setup without NLP.
For many types of language learning activities, however, extensionally specifying a direct
and complete mapping between potential learner input and intended feedback is not feasible.
Nagata (2009, pp. 563f) provides a clear illustration of this with an exercise taken from her
Japanese tutor ROBO-SENSEI in which the learner reads a short communicative context and
is asked to produce a sentence in Japanese that is provided in English by the system. The
learner response in this exercise is directly dependent on the input provided by the exercise
(a direct response in the terminology of Bachman & Palmer, 1996), so that a short, seven
word sentence can be defined as target answer. Yet after considering possible well-formed
lexical, orthographical, and word order variants, one already obtains 6048 correct sentences
which could be entered by the learner. Considering incorrect options, even if one restricts
ill-formed patterns to wrong particle and conjugation choices one obtains almost a million
sentences. Explicitly specifying a mapping between a million anticipated responses and their
corresponding feedback clearly is infeasible. Note that the explosion of possible learner an-
swers illustrated by Nagata already is a problem for the direct response in a constrained
activity, where the meaning to be expressed was fixed and only form variation was antici-
pated. A further dimension of potential variation in responses arises when going beyond the
analysis of language as a system (parallel to system-referenced tests in language assessment,
cf. Baker, 1989) to an analysis of the ability to use language to appropriately complete a
given task (performance-referenced).
In conclusion, for the wide range of language activities supporting significant well-formed
or ill-formed variation of form, meaning, or task-appropriateness of learner responses, it is
2

necessary to abstract away from the specific string entered by the learner to more general
classes of properties by automatically analyzing the learner input using NLP algorithms and
resources. Generation of feedback, learner model updating, and instructional sequencing can
then be based on the small number of language properties and categories derived through
NLP analysis instead of on the large number of string instances they denote.
What is the nature of the learner language properties to be identified? Research on Intelli-
gent Language Tutors has traditionally focused on learner errors, identifying and providing
feedback on them. Automatic analysis can also used in an ILTS to identify well-formed lan-
guage properties to be able to provide positive feedback or record in a learner model that
a given response provided evidence for the correct realization of a particular construction,
lexical usage, or syntactic relation. All approaches to detecting and diagnosing learner errors
must explicitly or implicitly model the space of well-formed and ill-formed variation that
is possible given a particular activity and a given learner. Insights into activity design and
the language development of learners thus are crucial for effective NLP analysis of learner
errors.
Errors are also present in native language texts and the need to develop robust NLP, which
also works in suboptimal conditions (due to unknown or unexpected forms and patterns,
noise), has been a driving force behind the shift from theory-driven, rule-based NLP in the
80s and 90s to the now-dominant data-driven, statistical and machine learning approaches.
However, there is an important difference in the goal of the NLP use in an ILTS compared to
that in other NLP domains. NLP is made robust to gloss over errors and unexpected aspects of
the system input with the goal of producing some result, such as a syntactic analysis returned
by a parser, or a translation provided by a machine translation system. The traditional goal of
the NLP in an ILTS, on the other hand, is to identify the characteristics of learner language
and in which way the learner responses diverge from the expected targets in order to provide
feedback to the learner. So errors here are the goal of the abstraction performed by the NLP,
not something to be glossed over by robustness of processing.
Writer’s aids such as the standard spell and grammar checkers (Dickinson, 2006) share the
ILTS focus on identifying errors, but they rely on assumptions about typical errors made
by native speakers which do not carry over to language learners. For example, Rimrott
& Heift (2008) observe that “in contrast to most misspellings by native writers, many L2
misspellings are multiple-edit errors and are thus not corrected by a spell checker designed
for native writers.” Tschichold (1999) also points out that traditional writer’s aids are not
necessarily helpful for language learners since learners need more scaffolding than a list of
alternatives from which to chose. Writer’s aids tools targeting language learners, such as the
ESL Assistant (Gamon et al., 2009), therefore provide more feedback and, e.g., concordance
views of alternatives to support the language learner in understanding the alternatives and
choosing the right. The goal of writer’s aids is to support the second language user in writing
a functional, well-formed text, not to support them in acquiring the language as is the goal of
an ILTS.
NLP methods for the diagnosis of learner errors fall into two general classes: On the one
hand, most of the traditional development has gone into language licensing approaches
which analyze the entire learner response. On the other hand, there is a growing number of
pattern-matching approaches which target specific error patterns and types (e.g., preposition
or determiner errors) ignoring any learner response or part thereof that does not fit the pattern.
Language licensing approaches are based on formal grammars of the language to be li-
censed, which can be expressed in one of two general ways (cf. Johnson, 1994). In a validity-
based setup, a grammar is a set of rules and recognizing a string amounts to finding valid
3

derivations. Simply put, the more rules are added to the grammar, the more types of strings
can be licensed; if there are no rules, nothing can be licensed. In a satisfiability-based setup,
a grammar is a set of constraints and a string is licensed if its model satisfies all constraints
in the grammar. Thus the more constraints are added, the fewer types of strings are licensed;
if there are no constraints in the grammar, any string is licensed.
Corresponding to these two types of formal grammars, there essentially are two types of
approaches to analyzing a string with the goal of diagnosing learner errors. The mal-rule ap-
proach follows the validity-based perspective and uses standard parsing algorithms. Starting
with a standard native language grammar, rules are added to license strings which are used
by language learners but not in the native language, i.e., so-called mal-rules used to license
learner errors (cf., e.g., Sleeman, 1982; Matthews, 1992, and references therein). Given that a
specific error type can manifest itself in a large number of rules – e.g., an error in subject-verb
agreement can appear in all rules realizing subjects together with a finite verb – meta-rules
can be used to capture generalizations over rules (Weischedel & Sondheimer, 1983). The
mal-rule approach can work well when errors correspond to the local tree licensed by a sin-
gle grammar rule. Otherwise, the interaction of multiple rules must be taken into account,
which makes it significantly more difficult to identify an error and to control the interaction
of mal-rules with regular rules. To reduce the search space resulting from rule interaction,
the use of mal-rules can be limited. In the simple case, the mal-rules are only added after
an analysis using the regular grammar fails. Yet this only reduces the search space for pars-
ing well-formed strings; if parsing fails, the question remains which mal-rules need to be
added. The ICICLE system (Michaud & McCoy, 2004) presents an interesting solution by
selecting the groups of rules to be used based on learner modeling. It parses using the native
rule set for all structures which the learner has shown mastery of. For structures assumed
to currently be acquired by the learner, both the native and the mal-rules are used. And for
structures beyond the developmental level of the learner, neither regular nor mal-rules are
included. When moving from traditional parsing to parsing with probabilistic grammars, one
obtains a further option for distinguishing native from learner structures by inspecting the
probabilities associated with the rules (cf. Wagner & Foster, 2009, and references therein).
The second group of language licensing approaches is typically referred to as constraint re-
laxation (Kwasny & Sondheimer, 1981), which is an option in a satisfiability-based grammar
setup or when using rule-based grammars with complex categories related through unifica-
tion or other constraints which can be relaxed. The idea of constraint relaxation is to elim-
inate certain constraints from the grammar, e.g., specifications ensuring agreement, thereby
allowing the grammar to license more strings than before. This assumes that an error can be
mapped to a particular constraint to be relaxed, i.e., the domain of the learner error and that
of the constraint in the grammar must correspond closely. Instead of eliminating constraints
outright, constraints can also associated with weights controlling the likelihood of an anal-
ysis (Foth et al., 2005), which raises the interesting issue how such flexible control can be
informed by the ranking of errors likely to occur for a particular learner given a particular
task. Finally, some proposals combine constraint relaxation with aspects of mal-rules. Reuer
(2003) combines a constraint relaxation technique with a standard parsing algorithm modi-
fied to license strings in which words have been inserted or omitted, an idea which essentially
moves generalizations over rules in the spirit of meta-rules into the parsing algorithm.
For pattern-matching, the most common approach is to match a typical error pattern, such
as the pattern looking for of cause in place of of course. By performing the pattern matching
on the part-of-speech tagged, chunked, and sentence delimited learner string, one can also
specify error patterns such as that of a singular noun immediately preceding a plural finite
4

verb (e.g., in The baseball team are established.). This approach is commonly used in stan-
dard grammar checkers and e.g., realized in the open source LanguageTool (Naber, 2003;
http://www.languagetool.org), which was not developed with language learners in mind, but
readily supports the specification of error patterns that are typical for particular learner pop-
ulations.
Alternatively, pattern-matching can also be used to identify contexts patterns which are
likely to include errors. For example, determiners are a well-known problem area for certain
learners of English. Using a pattern which identifies all nouns (or all noun chunks) in the
learner response, one can then make a prediction about the correct determiner to use for this
noun in its context and compare this prediction to the determiner use (or lack thereof) in the
learner response. Given that determiners and preposition errors are among the most common
English learner errors found and that the task lends itself well to the current machine learn-
ing approaches in computational linguistics and raises the interesting general question how
much context and which linguistic generalizations are needed for predicting such functional
elements, it currently is one of the most active areas of NLP research in the context of learner
language (cf., e.g., De Felice, 2008, and references therein).
Complementing the analysis of form, for an ILTS to offer meaning-based, contextualized
activities it is important to provide an automatic analysis of meaning aspects, e.g., to deter-
mine whether the answer given by the learner for a reading comprehension question makes
sense given the reading. While most NLP work in ILTS has addressed form issues, some
work has addressed the analysis of meaning in the context of ILTS (Delmonte, 2003; Ram-
say & Mirzaiean, 2005; Bailey & Meurers, 2008) and the issue is directly related to work in
computer-assisted assessment systems outside of language learning, e.g., for evaluating the
answers of short answer questions (cf. Pérez Marin, 2007, and references therein) or in essay
scoring (cf. Attali & Burstein, 2006, and references therein). In terms of NLP methods, it
also directly connects to the the growing body of research on recognizing textual entailment
and paraphrase recognition (Androutsopoulos & Malakasiotis, 2010).
Shifting the focus from the analysis techniques to the interpretation of the analysis, just like
the activity type and the learner play an important role in defining the space of variation to
be dealt with by the NLP, the interpretation and feedback provided to the learner needs to
be informed by activity and learner modelling (Amaral & Meurers, 2008). While feedback
in human-computer interaction cannot simply be equated with that in human-human interac-
tion, the results presented by Petersen (2010) for a dialogue-based ILTS indicate that results
from the significant body of research on the effectiveness of different types of feedback in in-
structed SLA can transfer across modes, providing fresh momentum for research on feedback
in the CALL domain (e.g., Pujolà, 2001).
A final important issue arising from the use of NLP in ILTS concerns the resulting lack of
teacher autonomy. Quixal et al. (2010) explore putting the teacher back in charge of design-
ing their activities with the help of an ICALL authoring system – a complex undertaking
since in contrast to regular CALL authoring software, NLP analysis of learner language
needs to be integrated without presupposing any understanding of the capabilities and limits
of the NLP.
Learner Corpora
While there seems to have been little interaction between ILTS and learner corpus research,
perhaps because ILTS traditionally have focused on exercises whereas most learner corpora
consist of essays, the analysis of learner language in the annotation of learner corpora (cf.
Granger, 1998, this volume) can be seen as an offline version of the online analysis performed
5

by an ILTS (Meurers, 2009).
What motivates the use of NLP for learner corpora? In contrast to the automatic analysis
of learner language in an ILTS providing feedback to the learner, the annotation of learner
corpora essentially provides an index to learner language properties in support of the goal
of advancing our understanding of acquisition in SLA research and to develop instructional
methods and materials in FLT. Corpus annotation can support a more direct mapping from
theoretical research questions to corpus instances (Meurers, 2005), yet for a reliable mapping
it is crucial for corpus annotation to provide only reliable distinctions which are replicably
based on the evidence found in the corpus and its meta-information, for which clear measures
are available (Artstein & Poesio, 2009).
What is the nature of the learner language properties to be identified? Just as for ILTS,
much of the work on annotating learner corpora has traditionally focused on learner errors,
for which a number of error annotation schemes have been developed (cf. Díaz Negrillo &
Fernández Domínguez, 2006, and references therein). There so far is no consensus, though,
on the external and internal criteria, i.e., which error distinctions are needed for which pur-
pose and which distinctions can reliably be annotated based on the evidence in the corpus
and any meta-information available about the learner and the activity which the language was
produced for. An explicit and reliable error annotation scheme and a gold standard reference
corpus exemplifying it is an important next step for the development of automatic error an-
notation approaches, which need an agreed upon gold standard for development as well as
for testing and comparison of approaches. Automating the currently manual error annotation
process using NLP would support the annotation of significantly larger learner corpora and
thus increase their usefulness for SLA research and FLT.
Since error annotation results from the annotator’s comparison of a learner response to hy-
potheses about what the learner was trying to say, Lüdeling et al. (2005) argue for making the
target hypotheses explicit in the annotation. This also makes it possible to specify alternative
error annotations for the same sentence based on different target hypotheses in multi-level
corpus annotation. However, Fitzpatrick & Seegmiller (2004) report unsatisfactory levels of
inter-annotator agreement in determining such target hypotheses. It is an open research issue
to determine for which type of learner responses written by which type of learners for which
type of tasks such target hypotheses can reliably be determined. Target hypotheses might
have to be limited to encoding only the minimal commitments necessary for error identifi-
cation; and in place of target hypothesis strings, they might have to be formulated at more
abstract levels, e.g., lemmas in topological fields, or sets of concepts the learner was trying to
express. In either case, if the target hypotheses are made explicit, the second step from target
hypothesis to error identification can be studied separately and might be realizable with high
reliability.
Returning to the general question about the nature of the learner language properties which
are relevant, SLA research essentially observes correlations of linguistic properties, whether
erroneous or not. And even research focusing on learner errors needs to identify correlations
with linguistic properties, e.g., to identify overuse/underuse of certain patterns, or measures
of language development. While the use of NLP tools trained on the native language cor-
pora is a useful starting point for providing a range of linguistic annotations, an important
next step is to explore the creation of annotation schemes and methods capturing the lin-
guistic properties of learner language (cf. Meunier, 1998; Dickinson & Ragheb, 2009; Díaz
Negrillo et al., 2010, and references therein).
In terms of using NLP for providing general measures of language development, Laufer &
Nation (1995) propose the use of Lexical Frequency Profiles (LFPs) to measure the active
6

vocabulary of learners, and a machine learning approach for predicting the levels of essays
written by second language learners based on a variety of word lists can be found in Pendar
& Chapelle (2008). Beyond the lexical level, Lu (2009) shows how the Revised D-Level-
Scale (Covington et al., 2006) encoding the order in which children acquire sentences of
increasing complexity can be measured automatically. Related measures are used in the
automated analysis of the complexity in second language writing (Lu, 2010). Finally, some
learner language specific aspects can only be identified once other levels of linguistic analysis
are annotated. For example, Hawkins & Buttery (2009) identify so-called criterial features
distinguishing different proficiency levels on the basis of part-of-speech tagged and parsed
portions of the Cambridge Learner Corpus.
2 NLP and the Analysis of Native Language for Learners
The second domain in which NLP connects to language learning derives from the need to
expose learners to authentic, native language and its properties and to given them opportunity
to interact with it. This includes work on searching for and enhancing authentic texts to be
read by learners as well as the automatic generation of activities and tests from such texts. In
contrast to the ILTS side of ICALL research covered in the first part of this article, the NLP in
the applications under discussion here is used to process native language in Authentic Texts,
hence referred to as ATICALL.
Most NLP research is developed and optimized for native language material and it is easier
to obtain enough annotated language material to train the statistical models and machine
learning approaches used in current research and development, so that in principle a wide
range of NLP tools with high quality analysis is available – even though this does not preempt
the question which language properties are relevant for ATICALL applications and whether
those properties can be derived from the ones targeted by standard NLP.
What motivates the use of NLP in ATICALL? Compared to using prefabricated materials
such as those presented in textbooks, NLP-enhanced searching for materials in resources
such as large corpora or the web makes it possible to provide on-demand, individualized
access to up-to-date materials. It supports selecting and enhancing the presentation of texts
depending on the background of a given learner, the specific contents of interest to them, and
the language properties and forms of particular relevance given the sequencing of language
materials appropriate for the learner’s stage (cf., e.g., Pienemann, 1998).
What is the nature of the learner language properties to be identified? Traditionally the
most prominent property used for selecting texts has been a general notion of readability,
for which a number of readability measures have been developed (DuBay, 2004). While
the traditional readability measures were computed by hand, the computation of most has
since been automated. Most of the measures involve only simple surface properties such as
counts of average word and sentence lengths, but sometimes also sophisticated NLP such
as morphological analysis is needed (cf. Ott & Meurers, 2010, sec. 3.2). Work in the Coh-
Metrix Project emphasizes the importance of analyzing text cohesion and coherence and
of taking a reader’s cognitive aptitudes into account for making predictions about reading
comprehension (McNamara et al., 2002). Other recent approaches replace the readability
formulas with more complex statistical classification methods for predicting the reading level
of texts (cf., e.g., Schwarm & Ostendorf, 2005; Collins-Thompson & Callan, 2005). Some
of the most advanced readability-based text retrieval methods stem from the REAP project,
which supports learners in searching for texts that are well-suited for providing vocabulary
7

and reading practice (Heilman et al., 2008). Indeed, the search for reading materials in
support of vocabulary acquisition (cf. Cobb, 2008, and references therein) is one of the most
prominent areas in this domain.
Going beyond readability, based on the insight from SLA research that awareness of language
categories and forms is an important ingredient for successful second language acquisition
(cf. Lightbown & Spada, 1999), a wide range of linguistic properties have been identified
as relevant for language awareness, including morphological, syntactic, semantic, and prag-
matic information (cf. Schmidt, 1995, p. 30). In response to this need, Ott & Meurers (2010)
discuss the design of a language aware search engine which supports access to web pages
based on a variety of language properties.
Complementing the question of how to obtain material for language learners, there are several
strands of ATICALL applications which focus on the enhanced presentation of and learner
interaction with such materials. One group of NLP-based tools such as COMPASS (Breidt &
Feldweg, 1997), Glosser (Nerbonne et al., 1998) and Grim (Knutsson et al., 2007) provides
a reading environment in which texts in a foreign language can be read with quick access
to dictionaries, morphological information, and concordances. The Alpheios project (http://
alpheios.net) focuses on literature in Latin and ancient Greek, providing links between words
and translations, access to online grammar reference, and a quiz mode asking the learner to
identify which word corresponds to which translation.
Another strand of ATICALL research focuses on supporting language awareness by au-
tomating input enhancement (Sharwood Smith, 1993), i.e., by realizing strategies which
highlight the salience of particular language categories and forms. For instance, WERTi
(Meurers et al., 2010) visually enhances web pages and automatically generates activities for
language patterns which are known to be difficult for learners of English, such as determiners
and prepositions, phrasal verbs, the distinction between gerunds and to-infinitives, and wh-
question formation. One can view such automatic visual input enhancement as an enrichment
of Data-Driven Learning (DDL). Where DDL has been characterized as an “attempt to cut
out the middleman [the teacher] as far as possible and to give the learner direct access to
the data” (Boulton 2009, p. 82, citing Tim Johns), in visual input enhancement the learner
stays in control, but the NLP uses ‘teacher knowledge’ about relevant and difficult language
properties to make those more prominent and noticeable for the learner.
A final, prominent strand of NLP research in this domain addresses the generation of exer-
cises and tests. Most of the work has targeted the automatic generation of multiple choice
cloze tests for language assessment and vocabulary drill (cf., e.g. Liu et al., 2005; Sumita
et al., 2005, and references therein). Issues involving NLP in this domain include the selec-
tion of seed sentences, the determination of appropriate blank positions, and the generation of
good distractor items. The VISL project (Bick, 2005) also includes a tool supporting the gen-
eration of automatic cloze exercises, which is part of an extensive NLP-based environment
of games and corpus tools aimed at fostering linguistic awareness for dozens of languages
(http://beta.visl.sdu.dk). Finally, the Task Generator for ESL (Toole & Heift, 2001) supports
the creation of gap-filling and build-a-sentence exercises such as the ones found in an ILTS.
The instructor provides a text, chooses from a list of learning objectives (e.g., plural, passive),
and select the exercise type to be generated. The Task Generator supports complex language
patterns and provides formative feedback based on NLP analysis of learner responses, bring-
ing us full circle to the research on ILTS we started with.
In conclusion, the use of NLP in the context of learning language offers rich opportunities,
both in terms of developing applications in support of language teaching and learning and
in terms of supporting SLA research – even though NLP so far has only had limited impact
8

on real-life language teaching and SLA. More interdisciplinary collaboration between SLA
and NLP will be crucial for developing reliable annotation schemes and analysis techniques
which identify the properties which are relevant and important for analyzing learner language
and analyzing language for learners.
SEE ALSO: Automatic Speech Recognition; Emerging Technologies for Language Learn-
ing; Intelligent Computer-Assisted Language Learning; Learner Modeling in Intelligent Com-
puter-Assisted Language Learning; Learner Corpora; Information Retrieval for Reading Tu-
tors; Technology and Language Testing; Text-to-Speech Synthesis Research
References
Amaral, L. & D. Meurers (2008). From Recording Linguistic Competence to Sup-
porting Inferences about Language Acquisition in Context: Extending the Concep-
tualization of Student Models for Intelligent Computer-Assisted Language Learning.
Computer-Assisted Language Learning 21(4), 323–338. URL http://purl.org/dm/papers/
amaral-meurers-call08.html.
Amaral, L. & D. Meurers (2011). On Using Intelligent Computer-Assisted Language Learn-
ing in Real-Life Foreign Language Teaching and Learning. ReCALL 23(1), 4–24. URL
http://purl.org/dm/papers/amaral-meurers-10.html.
Androutsopoulos, I. & P. Malakasiotis (2010). A Survey of Paraphrasing and Textual
Entailment Methods. Journal of Artiﬁcial Intelligence Research 38, 135–187. URL
http://www.jair.org/media/2985/live-2985-5001-jair.pdf.
Artstein, R. & M. Poesio (2009). Survey Article: Inter-Coder Agreement for Computational
Linguistics. Computational Linguistics 34(4), 1–42. URL http://www.mitpressjournals.
org/doi/abs/10.1162/coli.07-034-R2.
Attali, Y. & J. Burstein (2006). Automated Essay Scoring with E-rater v.2. Journal of Tech-
nology, Learning, and Assessment 4(3). URL http://escholarship.bc.edu/cgi/viewcontent.
cgi?article=1049&context=jtla.
Bachman, L. F. & A. S. Palmer (1996). Language Testing in Practice: Designing and De-
veloping Useful Language Tests. Oxford University Press.
Bailey, S. & D. Meurers (2008). Diagnosing meaning errors in short answers to reading
comprehension questions. In J. Tetreault, J. Burstein & R. D. Felice (eds.), Proceedings
of the 3rd Workshop on Innovative Use of NLP for Building Educational Applications
(BEA-3) at ACL’08. Columbus, Ohio, pp. 107–115. URL http://aclweb.org/anthology/
W08-0913.
Baker, D. (1989). Language Testing: A Critical Survey and Practical Guide. London:
Edward Arnold.
Bick, E. (2005). Grammar for Fun: IT-based Grammar Learning with VISL. In P. Juel (ed.),
CALL for the Nordic Languages, Copenhagen: Samfundslitteratur, Copenhagen Studies in
Language, pp. 49–64. URL http://beta.visl.sdu.dk/pdf/CALL2004.pdf.
Boulton, A. (2009). Data-driven Learning: Reasonable Fears and Rational Reassurance.
Indian Journal of Applied Linguistics 35(1), 81–106.
Breidt, E. & H. Feldweg (1997). Accessing Foreign Languages with COMPASS.
9

Machine Translation 12(1–2), 153–174. URL http://www.springerlink.com/content/
v833605061168351/fulltext.pdf. Special Issue on New Tools for Human Translators.
Burstein, J. & C. Leacock (eds.) (2005). Proceedings of the Second Workshop on Building
Educational Applications Using NLP. Ann Arbor, Michigan: Association for Computa-
tional Linguistics. URL http://aclweb.org/anthology/W/W05/#0200.
Cobb, T. (2008). Necessary or nice? The role of computers in L2 reading. In L2 Reading
Research and Instruction: Crossing the Boundaries, Ann Arbor: University of Michigan
Press.
Collins-Thompson, K. & J. Callan (2005). Predicting reading difficulty with statistical lan-
guage models. Journal of the American Society for Information Science and Technology
56(13), 1448–1462.
Covington, M. A., C. He, C. Brown, L. Naçi & J. Brown (2006). How complex is that
sentence? A proposed revision of the Rosenberg and Abbeduto D-Level Scale. Computer
Analysis of Speech for Psychological Research (CASPR) Research Report 2006-01, The
University of Georgia, Artificial Intelligence Center, Athens, GA. URL http://www.ai.uga.
edu/caspr/2006-01-Covington.pdf.
De Felice, R. (2008). Automatic Error Detection in Non-native English. Ph.D. thesis, St
Catherine’s College, University of Oxford.
Delmonte, R. (2003). Linguistic knowledge and reasoning for error diagnosis and feedback
generation. CALICO Journal 20(3), 513–532. URL http://purl.org/calico/delmonte-03.
html.
Dickinson, M. (2006). Writer’s Aids. In K. Brown (ed.), Encyclopedia of Language and
Linguistics, Oxford: Elsevier. 2 ed.
Dickinson, M. & M. Ragheb (2009). Dependency Annotation for Learner Corpora. In Pro-
ceedings of the Eighth Workshop on Treebanks and Linguistic Theories (TLT-8). Milan,
Italy. URL http://jones.ling.indiana.edu/~mdickinson/papers/dickinson-ragheb09.html.
DuBay, W. H. (2004). The Principles of Readability. Costa Mesa, California: Impact Infor-
mation. URL http://www.impact-information.com/impactinfo/readability02.pdf.
Díaz Negrillo, A. & J. Fernández Domínguez (2006). Error Tagging Systems for Learner
Corpora. Revista Española de Lingüística Aplicada (RESLA) 19, 83–102. URL http:
//dialnet.unirioja.es/servlet/fichero_articulo?codigo=2198610&orden=72810.
Díaz Negrillo, A., D. Meurers, S. Valera & H. Wunsch (2010). Towards interlanguage POS
annotation for effective learner corpora in SLA and FLT. Language Forum 36(1–2), 139–
154. URL http://purl.org/dm/papers/diaz-negrillo-et-al-09.html. Special Issue on Corpus
Linguistics for Teaching and Learning. In Honour of John Sinclair.
Fitzpatrick, E. & M. S. Seegmiller (2004). The Montclair electronic language database
project. In U. Connor & T. Upton (eds.), Applied Corpus Linguistics: A Multidi-
mensional Perspective, Amsterdam: Rodopi. URL http://chss.montclair.edu/linguistics/
MELD/rodopipaper.pdf.
Foth, K., W. Menzel & I. Schröder (2005). Robust parsing with weighted constraints. Natural
Language Engineering 11(01), 1–25.
Gamon, M., C. Leacock, C. Brockett, W. B. Dolan, J. Gao, D. Belenko & A. Klementiev
(2009). Using Statistical Techniques and Web Search to Correct ESL Errors. CAL-
ICO Journal 26(3), 491–511. URL http://research.microsoft.com/pubs/81312/Calico_
10

published.pdf;https://www.calico.org/a-757-Using%20Statistical%20Techniques%
20and%20Web%20Search%20to%20Correct%20ESL%20Errors.html.
Granger, S. (ed.) (1998). Learner English on Computer. London; New York: Longman.
Granger, S. (this volume). Learner Corpora. In C. A. Chapelle (ed.), The Encyclopedia of
Applied Linguistics. Technology and Language, Wiley-Blackwell.
Hawkins, J. A. & P. Buttery (2009). Using Learner Language from Corpora to Profile Levels
of Proficiency – Insights from the English Profile Programme. In Studies in Language Test-
ing: The Social and Educational Impact of Language Assessment, Cambridge: Cambridge
University Press.
Heift, T. (2010). Developing an Intelligent Language Tutor. CALICO Journal 27(3),
443–459. URL https://calico.org/a-811-Developing%20an%20Intelligent%20Language%
20Tutor.html.
Heift, T. (this volume). Intelligent Computer-Assisted Language Learning. In C. A.
Chapelle (ed.), The Encyclopedia of Applied Linguistics. Technology and Language,
Wiley-Blackwell.
Heift, T. & M. Schulze (2007). Errors and Intelligence in Computer-Assisted Language
Learning: Parsers and Pedagogues. Routledge.
Heilman, M., L. Zhao, J. Pino & M. Eskenazi (2008). Retrieval of Reading Materials for
Vocabulary and Reading Practice. In Proceedings of the Third Workshop on Innovative
Use of NLP for Building Educational Applications (BEA-3) at ACL’08. Columbus, Ohio:
Association for Computational Linguistics, pp. 80–88. URL http://aclweb.org/anthology/
W08-0910.
Johns, T. (1994). From printout to handout: Grammar and vocabulary teaching in the con-
text of data-driven learning. In T. Odlin (ed.), Perspectives on Pedagogical Grammar,
Cambridge: Cambridge University Press, pp. 293–313.
Johnson, M. (1994). Two ways of formalizing grammars. Linguistics and Philosophy 17(3),
221–248.
Jurafsky, D. & J. H. Martin (2009). Speech and Language Processing: An Introduction to
Natural Language Processing, Computational Linguistics, and Speech Recognition. Upper
Saddle River, NJ: Prentice Hall, second ed. URL http://www.cs.colorado.edu/~martin/slp.
html.
Knutsson, O., T. C. Pargman, K. S. Eklundh & S. Westlund (2007). Designing and developing
a language environment for second language writers. Computers & Education 49(4), 1122
– 1146.
Kwasny, S. C. & N. K. Sondheimer (1981). Relaxation Techniques for Parsing Grammat-
ically III-Formed Input in Natural Language Understanding Systems. American Journal
of Computational Linguistics 7(2), 99–108. URL http://portal.acm.org/citation.cfm?id=
972892.972894.
Laufer, B. & P. Nation (1995). Vocabulary Size and Use: Lexical Richness in L2 Written
Production. Applied Linguistics 16(3), 307–322. URL http://applij.oxfordjournals.org/
content/16/3/307.abstract.
Lightbown, P. M. & N. Spada (1999). How languages are learned. Oxford: Oxford
University Press. URL http://books.google.com/books?id=wlYTbuCsR7wC&lpg=
PR11&ots=_DdFbrMCOO&dq=How%20languages%20are%20learned&lr&pg=PR11#
11

v=onepage&q&f=false.
Liu, C.-L., C.-H. Wang, Z.-M. Gao & S.-M. Huang (2005). Applications of Lexical Informa-
tion for Algorithmically Composing Multiple-Choice Cloze Items. In Burstein & Leacock
(2005), pp. 1–8. URL http://aclweb.org/anthology/W05-0201.
Lu, X. (2009). Automatic measurement of syntactic complexity in child language acquisition.
International Journal of Corpus Linguistics 14(1), 3–28.
Lu, X. (2010). Automatic analysis of syntactic complexity in second language writing. In-
ternational Journal of Corpus Linguistics 15(4), 474–496.
Lüdeling, A., M. Walter, E. Kroymann & P. Adolphs (2005). Multi-level error annotation
in learner corpora. In Proceedings of Corpus Linguistics. Birmingham. URL http://www.
corpus.bham.ac.uk/PCLC/Falko-CL2006.doc.
Matthews, C. (1992). Going AI: Foundations of ICALL. Computer Assisted Language
Learning 5(1), 13–31.
McNamara, D. S., M. M. Louwerse & A. C. Graesser (2002). Coh-Metrix: Automated Cohe-
sion and Coherence Scores to Predict Text Readability and Facilitate Comprehension. Pro-
posal of Project funded by the Ofﬁce of Educational Research and Improvement, Reading
Program. URL http://cohmetrix.memphis.edu/cohmetrixpr/archive/Coh-MetrixGrant.pdf.
Meunier, F. (1998). Computer Tools for Interlanguage Analysis: A Critical Approach. In
G. Sylviane (ed.), Learner English on Computer, London and New York: Addison Wesley
Longman, pp. 19–37.
Meurers, D. (2009). On the Automatic Analysis of Learner Language: Introduction to
the Special Issue. CALICO Journal 26(3), 469–473. URL http://purl.org/dm/papers/
meurers-09.html.
Meurers, D., R. Ziai, L. Amaral, A. Boyd, A. Dimitrov, V. Metcalf & N. Ott (2010).
Enhancing Authentic Web Pages for Language Learners. In Proceedings of the 5th
Workshop on Innovative Use of NLP for Building Educational Applications (BEA-5)
at NAACL-HLT 2010. Los Angeles: Association for Computational Linguistics. URL
http://purl.org/dm/papers/meurers-ziai-et-al-10.html.
Meurers, W. D. (2005). On the use of electronic corpora for theoretical linguistics. Case
studies from the syntax of German. Lingua 115(11), 1619–1639. URL http://purl.org/dm/
papers/meurers-03.html.
Michaud, L. N. & K. F. McCoy (2004). Empirical Derivation of a Sequence of User Stereo-
types for Language Learning. User Modeling and User-Adapted Interaction 14(4), 317–
350. URL http://www.springerlink.com/content/lp86123772372646/.
Naber, D. (2003). A Rule-Based Style and Grammar Checker. Master’s thesis, Universität
Bielefeld. URL http://www.danielnaber.de/publications.
Nagata, N. (2009). Robo-Sensei’s NLP-Based Error Detection and Feedback Generation.
CALICO Journal 26(3), 562–579. URL https://www.calico.org/a-762-RoboSenseis%
20NLPBased%20Error%20Detection%20and%20Feedback%20Generation.html.
Nerbonne, J. (2003). Natural language processing in computer-assisted language learning. In
R. Mitkov (ed.), The Oxford Handbook of Computational Linguistics, Oxford University
Press. URL http://www.let.rug.nl/~nerbonne/papers/nlp-hndbk-call.pdf.
Nerbonne, J., D. Dokter & P. Smit (1998). Morphological Processing and Computer-Assisted
Language Learning. Computer Assisted Language Learning 11(5), 543–559. URL http:
12

//urd.let.rug.nl/nerbonne/papers/call_fr.pdf.
Ott, N. & D. Meurers (2010). Information Retrieval for Education: Making Search En-
gines Language Aware. Themes in Science and Technology Education. Special issue on
computer-aided language analysis, teaching and learning: Approaches, perspectives and
applications 3(1–2), 9–30. URL http://purl.org/dm/papers/ott-meurers-10.html.
Pendar, N. & C. Chapelle (2008). Investigating the Promise of Learner Corpora: Method-
ological Issues. CALICO Journal 25(2), 189–206. URL https://calico.org/html/article_
689.pdf.
Petersen, K. (2010). Implicit Corrective Feedback in Computer-Guided Interaction: Does
Mode Matter? Ph.D. thesis, Georgetown University. URL http://apps.americancouncils.
org/transfer/KP_Diss/Petersen_Final.pdf.
Pienemann, M. (1998). Language Processing and Second Language Development: Process-
ability Theory. Amsterdam: John Benjamins.
Pujolà, J.-T. (2001). Did CALL Feedback Feed Back? Researching Learners’ Use of Feed-
back. ReCALL 13(1), 79–98.
Pérez Marin, D. R. (2007). Adaptive Computer Assisted Assessment of free-text students’
answers: an approach to automatically generate students’ conceptual models. Ph.D. thesis,
Universidad Autonoma de Madrid.
Quixal, M., S. Preuß, B. Boullosa & D. García-Narbona (2010). AutoLearn’s authoring
tool: a piece of cake for teachers. In Proceedings of the NAACL HLT 2010 Fifth Work-
shop on Innovative Use of NLP for Building Educational Applications. Los Angeles, pp.
19–27. URL http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.170.6551&rep=
rep1&type=pdf.
Ramsay, A. & V. Mirzaiean (2005). Content-based support for Persian learners of English.
ReCALL 17, 139–154.
Reuer, V. (2003). Error recognition and feedback with Lexical Functional Grammar. CAL-
ICO Journal 20(3), 497–512. URL https://www.calico.org/a-291-Error%20Recognition%
20and%20Feedback%20with%20Lexical%20Functional%20Grammar.html.
Rimrott, A. & T. Heift (2008). Evaluating automatic detection of misspellings in German.
Language Learning and Technology 12(3), 73–92. URL http://llt.msu.edu/vol12num3/
rimrottheift.pdf.
Schmidt, R. (1995). Consciousness and foreign language: A tutorial on the role of atten-
tion and awareness in learning. In R. Schmidt (ed.), Attention and awareness in foreign
language learning, Honolulu: University of Hawaii Press, pp. 1–63.
Schulze, M. (this volume). Learner Modeling in Intelligent Computer-Assisted Language
Learning. In C. A. Chapelle (ed.), The Encyclopedia of Applied Linguistics. Technology
and Language, Wiley-Blackwell.
Schwarm, S. & M. Ostendorf (2005). Reading Level Assessment Using Support Vector
Machines and Statistical Language Models. In Proceedings of the 43rd Annual Meeting of
the Association for Computational Linguistics (ACL’05). Ann Arbor, Michigan, pp. 523–
530. URL http://aclweb.org/anthology/P05-1065.
Sharwood Smith, M. (1993). Input enhancement in instructed SLA: Theoretical bases. Stud-
ies in Second Language Acquisition 15, 165–179.
Sleeman, D. (1982). Inferring (mal) rules from pupil’s protocols. In Proceedings of ECAI-82.
13

Orsay, France, pp. 160–164.
Sumita, E., F. Sugaya & S. Yamamoto (2005). Measuring Non-native Speakers’ Proﬁciency
of English by Using a Test with Automatically-Generated Fill-in-the-Blank Questions. In
Burstein & Leacock (2005), pp. 61–68. URL http://aclweb.org/anthology/W05-0210.
Toole, J. & T. Heift (2001). Generating Learning Content for an Intelligent Language Tu-
toring System. In Proceedings of NLP-CALL Workshop at the 10th Int. Conf. on Artiﬁcial
Intelligence in Education (AI-ED). San Antonio, Texas, pp. 1–8.
Tschichold, C. (1999). Grammar checking for CALL: Strategies for improving foreign lan-
guage grammar checkers. In K. Cameron (ed.), CALL: Media, Design & Applications,
Exton, PA: Swets & Zeitlinger.
Wagner, J. & J. Foster (2009). The effect of correcting grammatical errors on parse prob-
abilities. In Proceedings of the 11th International Conference on Parsing Technologies
(IWPT’09). Paris, France: Association for Computational Linguistics, pp. 176–179. URL
http://aclweb.org/anthology/W09-3827.
Weischedel, R. M. & N. K. Sondheimer (1983). Meta-rules as a Basis for Processing
Ill-formed Input. Computational Linguistics 9(3-4), 161–177. URL http://aclweb.org/
anthology/J83-3003.
Suggested Readings
Swartz, M. L. & M. Yazdani (eds.) (1992). Intelligent tutoring systems for foreign language
learning: The bridge to international communication. Berlin: Springer.
Holland, V., J. Kaplan & M. Sams (eds.) (1995). Intelligent language tutors. Theory shaping
technology. Hillsdale: Lawrence Erlbaum Associates, Inc.
Garside, R., G. Leech & A. McEnery (eds.) (1997). Corpus Annotation. Harlow: Addison
Wesley Longman.
14

Natural Language Processing and Language Learning

More Related Content

What's hot (20)

Similar to Natural Language Processing and Language Learning (20)

More from antonellarose (8)

Recently uploaded (20)

Natural Language Processing and Language Learning