Presented at the American Association for Applied Linguistics Annual Conference, Denver, Colorado; 21 March 2009.
This colloquium brought together researchers interested in both applied and computational linguistics. The six presentations discussed a variety of current topics and showcased a diversity of computer applications by highlighting some of the challenges and opportunities that result from cross-disciplinary research.
Trude Heift and Anne Rimrott (Simon Fraser University) began the colloquium with their presentation ‘Computational and pedagogical design challenges of intelligent language tutoring systems (ILTSs)’, which centered around issues of task design in computer-assisted language learning (CALL). Their study on task-induced variation of interlanguage reported on longitudinal data collected from second language (L2) learners working on three task types in the E-Tutor (www.e-tutor.org), an ILTS for German. Study results showed systematic variation within language learners at the syntactic level across form-focused and meaning-focused tasks. Their findings prompt suggestions for the choice and design of task types in CALL and their impact on L2 acquisition.
The presentation by Detmar Meurers (University of Tübingen) and Luiz Amaral (University of Massachusetts) focused ‘On the annotation and use of learner language corpora’. Data collected in learner corpora in principle can help validate generalizations about language acquisition and provide a broad empirical basis for the development of new hypotheses and theories. The authors argued for the creation of learner corpora stemming from a variety of contexts and tasks, the linguistic annotation of learner corpora to support effective querying for patterns discussed in second language acquisition (SLA) research, and the importance of high annotation quality and how it can be achieved. They identified a need for more interdisciplinary collaboration between applied and computational linguistics to develop adequate annotation schemes for learner language gold standard corpora and automatic annotation methods for interlanguage.
Jill Burstein's (Educational Testing Service) presentation entitled ‘Pedagogically-motivated natural language processing technology’ discussed E-rater® and Text Adaptor (http://www.ets.org/research/erater.htm), two technologies that evolved through rich collaboration between computational and applied linguists and teachers. E-rater® development began within the field of natural language processing (NLP), with a modest set of NLP-based tools that extracted content, syntactic, and discourse information from student essays to capture a range of text characteristics that reflect writing quality. Since its initial development, new features have been added, always keeping in mind the goal of enriching the system's representation of the writing construct. With regard to Text Adaptor's NLP-based development, adaptations created by teachers and evaluated by experts opened new doors to NLP research with respect to text quality measures and text appropriateness for students at different levels of English proficiency.
Michael Gamon's (Microsoft) talk ‘The MSR ESL Assistant: System and usage’ provided an overview of the Microsoft Research ESL Assistant (www.eslassistant.com), a prototype web-based proofing tool. The system's architecture consists of a combination of error correction modules that target specific error types and utilize machine-learned components and heuristic rules. A language model component filters out spurious suggestions. Data sets indicate that users can reliably identify and apply good suggestions. Users also tend to take advantage of spurious suggestions that survive filtering. Although incorrect, these false flags often highlight the neighborhood of some dysfluency in the sentence. Once the users’ attention has been drawn to this general vicinity, it is often possible for them to manually correct the sentence.
Kathleen McCoy's (University of Delaware) presentation was entitled ‘Applications of linguistics to communication aids for people with disabilities’. People who are unable to speak might use a device to select what they want to say. A major issue in such systems is the speed of communication, which can be increased by predicting the next word the user is typing in order to save keystrokes. Traditional statistical methods base their prediction on the previous few words. This talk discussed ways to improve these methods by incorporating information about the broad topic of the conversation into the statistical models. Pragmatic issues involving tradeoffs between speed versus accuracy were also explored.
The final presentation, ‘Statistical semantics’, by Walter Kintsch (University of Colorado), focused on the CI-2 model and its application to Summary Street (http://www.pearsonkt.com/prodSSt.shtml), a writing program used in Colorado schools. Current statistical models of semantics (e.g., Latent Semantic Analysis, Topic Model) describe what is stored in long-term memory. For instance, word meanings are generated in working memory from long-term memory traces that record our experience with words. The dual-memory CI-2 model describes how this information is used to construct sentence meanings. It distinguishes between a gist level and an explicit level and incorporates syntactic information about how words are used. The construction of meaning is conceptualized as feature sampling from the explicit memory traces, with the constraint that the sampling must be contextually relevant both semantically and syntactically.
The colloquium concluded with a general discussion that emphasized the need for more collaborative research efforts between computational and applied linguists. It was the first of its kind at AAAL and, given similar events that have taken place at other CALL-related and SLA-related conferences, we hope that the success of this colloquium has also sparked interest in bringing together these two communities at future AAAL conferences. We would like to thank the conference organizers for making this event possible.
