PhD positions in Northwestern

The Language and Computation Lab in the Department of Linguistics at Northwestern University, directed by Dr. Klinton Bicknell, is currently looking for Ph.D. students to join the lab. Candidates should have a strong academic background, and especially desirable skills include some experience in computer science and/or math. Interested students should apply to the Ph.D. program in the Department of Linguistics, which admits students to the department as a whole rather than to a specific lab and offers all admitted students competitive five-year funding packages not tied to a specific advisor. International applications are welcome. The department's deadline to receive applications for starting in Fall 2015 is November 30, 2014. Research in the Language and Computation Lab investigates how the human brain solves the computational problems of language comprehension, production, and acquisition. We use techniques from machine learning, computational linguistics, and statistics to build computational models of such language behaviors. We test these models by analyzing large behavioral and neuroscientific datasets, and also by gathering new empirical data, especially via eye tracking and crowdsourcing. For more information on current research, see the lab website at http://lcl.northwestern.edu/. The Department of Linguistics at Northwestern has a wealth of excellent language researchers, as does the university more broadly in the Psychology, Communication Sciences & Disorders, and Electrical Engineering & Computer Science departments. Students are highly encouraged to collaborate both within Linguistics and across departments. Students can also take advantage of the university's location in the dynamic Chicago metro area. Feel free to direct any questions to Klinton at kbicknell a_t northwestern.edu.

Two new papers: Lena Jäger et al (JML), and Logačev et al (Cognitive Science)

Two new papers have been accepted for publication by our PhD students Lena Jäger and Pavel Logačev:

Lena Jäger, Zhong Chen, Qiang Li, Chien-Jer Charles Lin, and Shravan Vasishth. The subject-relative advantage in Chinese: Evidence for expectation-based processing. Journal of Memory and Language, in press. [ DOI | .pdf ]

Chinese relative clauses are an important test case for pitting the predictions of expectation-based accounts against those of memory-based theories. The memory-based accounts predict that object relatives should be easier to process than subject relatives because, in object relatives, less linguistic material intervenes between the head noun and the gap (or verb) that it associates with. By contrast, expectation-based accounts such as surprisal predict that the less frequently occurring object relative should be harder to process than the subject relative, because building a rarer structure is computationally more expensive. Previous studies on Chinese relative clauses have the problem that local ambiguities in subject and object relatives could be confounding the comparison. We compared reading difficulty in subject and object relatives (in both subject- and object-modifications) in which the left context leads the reader to predict a relative clause structure as the most likely continuation; we validate this assumption about what is predicted using production data (a sentence completion study and a corpus analysis). Two reading studies (self-paced reading and eye-tracking) show that the Chinese relative clause evidence is consistent with the predictions of expectation-based accounts but not with those of memory-based theories. We present new evidence that the prediction of upcoming structure, generated through the probabilistic syntactic knowledge of the comprehender, is an important determiner of processing cost.

Pavel Logačev and Shravan Vasishth. A Multiple-Channel Model of Task-Dependent Ambiguity Resolution in Sentence Comprehension. Cognitive Science, 2014. Accepted pending minor revision.

Traxler et al. (1998) found that ambiguous sentences are read faster than their unambiguous counterparts. This so-called ambiguity advantage has presented a major challenge to classical theories of human sentence comprehension (parsing) because its most prominent explanation, in the form of the unrestricted race model (URM), assumes that parsing is non-deterministic. Recently, Swets et al. (2008) have challenged the URM. They argue that readers strategically underspecify the representation of ambiguous sentences to save time, unless disambiguation is required by task demands. When disambiguation is required, however, readers assign sentences full structure — and Swets et al. provide experimental evidence to this end. On the basis of their findings they argue against the URM and in favor of a model of task-dependent sentence comprehension. We show through simulations that the Swets et al. data does not constitute evidence for task-dependent parsing because it can be explained by the URM. However, we provide decisive evidence from a German self-paced reading study consistent with Swets et al.’s general claim about task-dependent parsing. Specifically, we show that under certain conditions, ambiguous sentences can be read more slowly than their unambiguous counterparts, suggesting that the parser may create several parses, when required. Finally, we present the first quantitative model of task-driven disambiguation which subsumes the URM, and show that it can explain both Swets et al.’s results and our findings.

Two new papers: Hanne et al, and Sorensen and Vasishth

We just submitted two new papers:

1. A tutorial on fitting Bayesian linear mixed models using Stan:
   pdf: http://www.ling.uni-potsdam.de/~vasishth/pdfs/SorensenVasishthJMP.pdf
   abstract: With the arrival of the R packages nlme and lme4, linear mixed models (LMMs) have come to be widely used in psychology, cognitive science, and related areas. In this tutorial, we provide a practical introduction to fitting LMMs in a Bayesian framework using the probabilistic programming language Stan. Although the Bayesian framework has several important advantages, specifying a Bayesian model requires quite a lot of background knowledge compared to frequentist tools like lme4. This tutorial provides the necessary background through two detailed examples of self-paced reading studies with repeated measures. One is a two-condition design, and the other a 2×2 factorial design. These two examples can easily be extended to more complex factorial designs. The data and code associated with this tutorial are available as a supplement.
2. Sentence comprehension in aphasia: Eye-tracking reveals delayed morphological cue integration and late parsing commitments
   http://www.ling.uni-potsdam.de/~vasishth/pdfs/Hanne-Burchert-DeBleser-Vasishth-JNL-submitted.pdf
   abstract: Comprehension of non-canonical sentences can be difficult for individuals with aphasia (IWA). It is still unclear to which extent morphological cues like case-marking or verb inflection may influence IWA’s performance or even help to override deficits in sentence comprehension. Until now, studies have mainly used offline methods to draw inferences about syntactic deficits and, so far, only a few studies have looked at online syntactic processing in aphasia. We investigated sentence processing in German IWA by combining an offline (sentence-picture matching) and online (eye-tracking in the visual-world paradigm) method. Our goal was to determine whether IWA are capable of using inflectional morphology (number-agreement markers on verbs and case markers in noun phrases) as a cue to sentence interpretation. We report results of two visual-world experiments using German reversible SVO and OVS sentences. In each study, there were eight IWA and 20 age-matched controls. Experiment 1 targeted the role of case-morphology, while Experiment 2 looked at processing of number-agreement cues at the verb in case-ambiguous sentences. IWA showed deficits in using both types of morphological markers as a cue to non-canonical sentence interpretation and the results indicate that in aphasia, processing of case-marking cues is more vulnerable as compared to verb-agreement morphology. However, the online data revealed that IWA are in principle capable of successfully computing morphological cues, but the integration of morphological information is delayed as compared to age-matched controls. Furthermore, we found striking differences between controls and IWA regarding subject-before-object parsing predictions. While in case-unambiguous sentences IWA showed evidence for early subject-before-object parsing commitments, they exhibited no straightforward subject-first bias in case-ambiguous sentences, although controls did so for ambiguous structures. IWA delayed their parsing decisions in case-ambiguous sentences until unambiguous morphological information, such as a subject-verb-number-agreement cue, was available. We attribute the differential results for processing of case and agreement markers to differences in the degree of reliability of both morphological cues. We ascribe our findings for erroneous processing of case-unambiguous sentences in aphasia to late parsing commitments and failures in integrating case cues on time. For processing of case-ambiguous sentences in aphasia, we suggest that IWA adopt a wait-and-see strategy and make parsing commitments only when the agreement cue at the verb prompts a particular sentence structure. Our results for IWA further point to deficits in predictive processes during sentence comprehension.

Two new papers

We have two new papers:

1. Jaeger et al 2014 (under review): The subject-relative advantage in Chinese relatives: Evidence for expectation-based processing, pdf
   abstract: Chinese relative clauses are an important test case for pitting the predictions of expectation-based accounts against those of memory-based theories. The memory-based accounts predict that object relatives should be easier to process than subject relatives because, in object relatives, less linguistic material intervenes between the head noun and the gap (or verb) that it associates with. By contrast, expectation-based accounts such as surprisal predict that the less frequently occurring object relative should be harder to process than the subject relative, because building a rarer structure is computationally more expensive. Previous studies on Chinese relative clauses have the problem that local ambiguities in subject and object relatives could be confounding the comparison. We compared reading difficulty in subject and object relatives (in both subject- and object-modifications) in which the left context leads the reader to predict a relative clause structure as the most likely continuation; we validate this assumption about what is predicted using production data (a sentence completion study and a corpus analysis). Two reading studies (self-paced reading and eye-tracking) show that the Chinese relative clause evidence is consistent with the predictions of expectation-based accounts but not with those of memory-based theories. We present new evidence that the prediction of upcoming structure, generated through the probabilistic syntactic knowledge of the comprehender, is an important determiner of processing cost.
2. Hofmeister and Vasishth 2014: Distinctiveness and encoding effects in online sentence comprehension, pdf
   abstract: In explicit memory recall and recognition tasks, elaboration and contextual isolation both facilitate memory performance. Here, we investigate these effects in the context of sentence processing: targets for retrieval during online sentence processing of English object relative clause constructions differ in the amount of elaboration associated with the target noun phrase, or the homogeneity of superficial features (text color). Experiment 1 shows that greater elaboration for targets during the encoding phase reduces reading times at retrieval sites, but elaboration of non-targets has considerably weaker effects. Experiment 2 illustrates that processing isolated superficial features of target noun phrases — here, a green word in a sentence with words colored white — does not lead to enhanced memory performance, despite triggering longer encoding times. These results are interpreted in the light of the memory models of Nairne 1990, 2001, 2006, which state that encoding remnants contribute to the set of retrieval cues that provide the basis for similarity-based interference effects.

Teaching a course in Paris

I'm teaching a course in Paris starting next week: http://www.labex-efl.org/?q=fr/node/269

Paper submitted: The importance of reading naturally

Here is a new paper from our lab: http://www.ling.uni-potsdam.de/~vasishth/pdfs/MetznerEtAl2014P600.pdf Abstract: Traditionally, a lot of ERP research on language processing uses auto-paced, word-by-word presentation. What is the difference in ERP effects between word-by-word presentation and natural reading? We investigated this question by conducting two experiments; event-related brain potentials were recorded while participants read either word-by-word or naturally. In the natural reading experiment, eye movements were recorded concurrently with the brain potentials. Comprehension difficulty was induced by syntactic (gender mismatch) or semantic violations (implausible association of adjective and noun) that occurred either sentence-medially or sentence-finally. This design allowed us to investigate whether recovery from unexpected input differs in the two modalities. Both the word-by-word and natural reading experiments showed the classical N400 and P600 effects seen in syntactic and semantic violations, replicating published results. However, compared to word-by-word presentation, natural reading led to higher accuracy in judging well-formedness of sentences; and in natural reading, trials with regressive eye movements led to higher accuracy than trials without regressive eye movements. Interestingly, in the sentence-medial violations, a P600 effect was observed when regressions occurred, suggesting that a regressive eye movement event is associated with a recovery process. By contrast, when no regression occurred, no effect was seen in the brain potentials sentence-medially, implying that no recovery process was triggered. In sentence-final position, both violations led to N400 and P600 effects in regression trials, and a sustained negativity in no-regression trials, the latter suggesting that the reader may have detected the violation but was willing to tolerate it. In sum, we show that natural reading leads to higher comprehension accuracy than word-by-word presentation. Additionally, the separation of regression and no-regression trials revealed the strategic choices made by the comprehension system. Thus, although the relative simplicity of the word-by-word presentation paradigm offers clear advantages, important aspects of language comprehension can perhaps be better studied using natural reading.

Vasishth Lab at Ohio State, 2014

We had a very successful trip to the CUNY sentence processing conference at Ohio State; two talks (Paul Metzner and Titus von der Malsburg; and Pavel Logacev), and six posters.