Week of Nov 10

Johnson at the Harvard Linguistics Colloquium Talk Series

The third talk for Harvard Linguistics Colloquium Talk Series of Fall 2025 was given by Kyle Johnson (UMass Amherst) on Friday November 7. The talk was titled 'A vintage syntax and fashion-forward semantics for theta-roles'. Thank you so much for the great talk, Kyle! 

Milenković published in Glossa

We’re excited to announce that Aljoša Milenković, a 5th-year PhD student in our department, has just published an article in Glossa! His paper, 'Against strict stratum-internal transparency: Within-stratum countershifting in Gallipoli Serbian', offers fresh insights into the discussion on Stratal OT. Congratulations to Aljoša on this well-deserved success! Read more about his work here: https://www.glossa-journal.org/article/id/23107/.

GSAS Workshop in IE & Historical Linguistics

We are excited to announce the third GSAS Indo-European & Historical Linguistics Workshop talk of Fall 2025, this Friday, November 14! Paul Russell will be presenting research on Friday November 14 @ 5 PM in Boylston 335. 

Speaker: Paul Russell (Harvard University)

Time: Friday November 14 @ 5 PM EST

Location: Boylston 335 (Linguistics department, third floor)

Title: 'Down among the marsupials: pretonic phonology in insular Celtic languages'

Abstract: While the patterns of reduction and syncope after the initial stress accent in Goidelic and around the penultimate accent in Brittonic languages are relatively well understood, what happens in pretonic position is less well understood; there is an assumption of reduction of various kinds but problems arise in particular with prepositions and preverbs especially in contact with articles and pronominal elements. Part of the problem is that the phonology of these reductions is not clear and the evidence is thin and ambiguous, perhaps because we have not been taking all the evidence into account. This paper attempts to clarify the issues, if not to provide any answers.

We look forward to seeing you all there!

LangCog 

The next LangCog meeting will be Tuesday, 11/11 from 5:30-7:00pm, in William James Hall, Room 1550. The speaker is Ankana Saha (Harvard University), and the title and abstract of her talk can be found below. You can find the schedule for the remainder of the semester on the LangCog website. Food will be provided, as always. 

Title: From Humans to Language Models: Testing the Boundaries of Pragmatic Understanding

Abstract: How do we keep track of who or what is being talked about as discourse unfolds? Languages employ a range of referential expressions in anaphora, such as pronouns (he), definites (the boy), and demonstratives (that boy), but how discourse structure constrains their use remains a central question for theories of meaning. Focusing on definites and demonstratives in anaphoric contexts, this work investigates the pragmatic principles that govern their acceptability. Although anaphora is often assumed to offer limited insight into the distinction between definites and demonstratives, I argue that it in fact provides a powerful diagnostic of their contrasting pragmatic profiles. Using cross-linguistic experimental data from English, Turkish, and Bangla, I show that demonstratives in anaphora are uniquely sensitive to subtle information-structural contrasts, unlike definites. This empirical pattern brings to the fore the distinctive pragmatic signature of demonstratives and establishes a new experimental framework for testing competing theories of definite reference, particularly in languages whose definiteness systems are complex or debated, as shown for Mandarin and German. Extending this line of inquiry, I examine whether these discourse-level constraints on reference are computationally tractable in artificial language systems. Nineteen large language models (LLMs) were tested to evaluate how model size and architecture affect discourse sensitivity. While larger models closely mirrored human behavior in tracking the discourse-structural constraints on demonstratives, smaller models did not. These findings provide the first systematic assessment of discourse-pragmatic competence in LLMs, showing that the fine-grained pragmatic conditions underlying demonstrative use are not easily learnable from surface-level statistics alone and emerge only in more advanced models.