Brian W. Dillon
Associate Professor, Department of Linguistics, UMass Amherst
In the sentence Mary likes her, English speakers intuitively understand that her cannot refer to Mary - it's someone else. These Principle B effects occur in many languages. But why do languages have this constraint? And how do comprehenders so effortlessly exclude the unacceptable interpretation, even though they're likely thinking about Mary when they encounter the pronoun? Our research group - BUMASA (Binding at UMASs) - is exploring questions like these in English, Tagalog, and Turkish using both comprehension methods and production methods. Initial results suggest that English speakers have a 'default assumption' of disjoint reference: When processing transitive predicates like like, they presume that the two entities involved in the event are distinct.
Outcomes
Keshev, M., Wagers, M., & Dillon, B. (submitted). Predictability of the retrieval site does not modulate interference: Evidence from reflexive attraction.
Bakay, Ö., Akkus, F. & Dillon, B. (2023). The role of structural cues and recency in processing a pre-verbal anaphor in Turkish. Poster presented at the 36th Annual Conference on Human Sentence Processing (HSP), University of Pittsburgh
Bakay, Ö., Burnsky, J., Keshev, M., Asatryan, M., Johnson, K. & Dillon, B. (2023). Predicting disjoint reference: offline and online evidence. Poster presented at the 36th Annual Conference on Human Sentence Processing (HSP), University of Pittsburgh
Burnsky, J., Keshev, M., Asatryan, M., Hlachova, B., Johnson, K. & Dillon, B. (2022, March). Look Away! An Object is Coming. Poster presented at the 35th annual Human Sentence Processing Conference, 2022, University of California, Santa Cruz.
Pizarro-Guevara, J., & Dillon, B. (2022). What Tagalog can teach us: The influence of word order in reflexive processing. Proceedings of AFLA XXVIII (The Twenty-Eighth Meeting of the Austronesian Formal Linguistics Association).
Kush, D., & Dillon, B. (2021). Principle B constrains the processing of cataphora: Evidence for syntactic and discourse predictions. Journal of Memory and Language, 120, 104254. [Preprint] [OSF]
To what degree can garden path effects be reduced to effects of word-level predictability (surprisal)? And do deep neural network models process garden path ambiguities in a way that is similar to humans? With Tal Linzen, our group is addressing these questions by building the Syntactic Ambiguity Processing (SAP) Benchmark. The SAP benchmark is a large, open dataset of reading time data on a range of syntactic ambiguities, using both self-paced reading and eye-tracking-while-reading techniques. The early results of our project suggest that garden path effects are unlikely to (just) reflect word predictability, and that at present DNNs do a relatively poor job in capturing the processing difficulty human readers experience on these sentences.
Outcomes
Huang et al. Surprisal does not explain syntactic disambiguation difficulty: Evidence from a large-scale benchmark
Arehalli, S., Dillon, B. & Linzen, T. (2022). Syntactic Surprisal from Neural Models Predicts, but Underestimates, Human Processing Difficulty From Syntactic Ambiguities. CoNLL 2022. Recipient of a Distinguished Paper Award - Congrats Suhas!
Arehalli, S., Dillon, B. & Linzen, T. (2022). Syntactic Surprisal from Neural Language Models tracks Garden Path Effects. Poster presented at the 35th annual Human Sentence Processing Conference, 2022, University of California, Santa Cruz.
Huang, K.-J., Arehalli, S., Kugemoto, M., Muxica, C., Prasad, G., Dillon, B., & Linzen, T. (2022). SPR mega-benchmark shows surprisal tracks construction- but not item-level difficulty. Talk presented (virtually) at the 35th annual Human Sentence Processing Conference, 2022, University of California, Santa Cruz (virtual).
Arehalli, S., & Dillon, B.. (2022). Neural Nets and Language Processing. Minicourse presented at the Linguistic Society of America's Annual Meeting 2022, Washington, DC.
Human memory is prone to error and interference, even at very short time scales. One way this impacts language use is the errors that users make in forming agreement dependencies, both in comprehension and production. With Aya Meltzer-Asscher and Maayan Keshev, our group is investigating these types of memory errors in comprehension in Hebrew and English. In this work we ask: How do comprehenders update noisy representations of past input as they encounter new material? And does rich agreement marking in languages like Hebrew help to combat memory interference by providing redudant cues to the features of items in memory? This work is supported by the National Science Foundation (NSF #2146798) in collaboration with the Binational Science Foundation.
Outcomes
Dillon, B. & Keshev, M. (to appear). Syntactic dependency formation in sentence processing: A comparative perspective. Chapter to appear in Polinsky, Corver, and Barbiers (eds.), The Cambridge Handbook of Comparative Syntax.
Keshev, M., Cartner, M., Neal, A., Meltzer-Asscher, A., & Dillon B. (2022). Dynamic encoding of agreement features and its effects on interference. Poster presented at the 35th annual Human Sentence Processing Conference, 2022, University of California, Santa Cruz.