Week of Nov 18
Ross at EMNLP
Our fifth-year PhD student Hayley Ross presented her paper 'Is artificial intelligence still intelligence? LLMs generalize to novel adjective-noun pairs, but don’t mimic the full human distribution', co-authored with Kathryn Davidson and Najoung Kim (BU), at the GenBench workshop of the EMNLP conference in Miami this weekend. Her paper also won the best paper award of the workshop! Huge congrats, Hayley!
Milenković at Harvard GSAS Indo-European & Historical Linguistics Workshop
The fourth GSAS Indo-European & Historical Linguistics Workshop talk this semester was given by our fourth-year PhD student Aljoša Milenković on Friday Nov 15. The title of the talk was '(Un)expected immobile accent in some Slavic infinitives'. Thanks for the great talk, Aljoša!
Harvard Linguistics Colloquium Talk Series
Matthew Hewett will be giving a talk as part of the Harvard Linguistics Colloquium Talk Series this Friday Nov 22 at 12pm. Please see the details below. There will also be a joint dinner reception with the Indo-European & Historical Linguistics Workshop on the same day, held in the department lounge. Hope to see many of you there!
Title: The A-/Ā-distinction is categorial: the view from A-resumption in Arabic
Speaker: Matthew Hewett, UPenn
Time: Friday Nov 22nd @ 12pm
Location: The Fong Auditorium, Boylston Hall (1st floor)
Abstract: The A-/Ā-distinction—a distinction between dependency types like raising-to-subject and wh-movement—has been a mainstay of modern syntactic theorizing since Chomsky (1981). Yet, despite widespread acceptance of the empirical correlates of this distinction, its theoretical underpinning remains a matter of debate. In this talk, I offer a novel perspective on understanding the A-/Ā-distinction by examining resumptive A-dependencies in Arabic (on which see Doron and Heycock 1999, 2010 and Sellami 2024). A-resumption challenges traditional positional approaches to the A-/Ā-distinction that predict a strict bipartition among position types (i.e. A-positions vs. Ā-positions, see Chomsky 1981, Mahajan 1990) because, as I show for the first time, A-resumption simultaneously displays a mixture of A- and Ā-properties. For instance, A-resumption feeds agreement with T (= A-property) and skips over intervening DPs and finite clause boundaries (= Ā-properties). Moreover, A-resumption challenges recent approaches which derive A- and Ā-properties from the mechanics of movement (e.g. van Urk 2015, Safir 2019) because A-resumption involves binding without movement.
Instead, I develop an analysis of A-resumption and the A-/Ā-distinction based on the following two claims about structure-building features:
(1) There is a distinction between A- and Ā-features (see also van Urk 2015) which reduces to the category of the constituent targeted by that feature: A-features target (features of) DPs, Ā-features target (features of) QPs (in the sense of Cable 2010).
(2) There is a distinction between the featural triggers for external Merge and internal Merge (Hewett 2023; Elbourne 2024).
A-resumption is derived by externally merging a DP (and not a QP) in a non-thematic position and binding a pronominal variable in its scope. I demonstrate that the properties of A-movement, A-resumption, and Ā-movement are correctly predicted by this account—a success arguably not shared by its alternatives. From a broader perspective, my analysis supports the conclusion that Merge is feature-driven (i.e. driven by features of lexical items, see Adger 2003; Müller 2010; Zyman 2018) rather than free (see Chomsky 2015; Collins 2017; Safir 2019), shedding light on the nature of syntactic structure building.
Harvard GSAS Indo-European & Historical Linguistics Workshop
The fifth GSAS Indo-European & Historical Linguistics Workshop talk of the Fall 2024 semester will take place this Friday! Sólveig Hilmarsdóttir will be presenting research on Friday November 22 @ 5 PM in Boylston 335. The talk will be followed by a reception in the department lounge. Please find all details below.
Speaker: Sólveig Hilmarsdóttir (University of Cambridge)
Time: Friday November 22 @ 5 PM EST
Location: Boylston 335 (Linguistics department, third floor)
Title: No country for old-minō? On the different forms of the Latin future imperative
Abstract: In addition to the present imperative, Latin had a so-called ‘future imperative’ (sometimes called ‘imperative II’ or the ‘imperative in -to’). See example (1) where the future imperative redito takes place after the present imperative abi (transl. by de Melo 2011):
(1) Abi quaeso hinc domum ... Redito huc circiter meridie.
‘Please go home ... Return here around midday’ (Pl.Mos.578–9).
I will briefly outline the function of this verbal category before focusing in detail on an aspect of deponent/passive usage of future imperatives in Latin. Throughout the history of the Latin language, we see a competition between the three different singular endings that could be used for 2/3p. deponent/passive future imperatives: -tō, -tor and -minō; the last of which has been referred to by Ringe as ‘one of the stranger products of “analogical change”’ (Ringe 2007:301). By comprehensively examining the usage and distribution of these endings in Latin texts, I will provide new suggestions about their chronology and register. Further, I will consider how the absence of evidence may in itself be telling for language change. I do so by showing how Latin authors largely avoided using the future imperative of deponents and passives altogether, and how eventually, all three forms disappeared.
LangCog
The next LangCog meeting of the semester will be on Tuesday, November 19th from 5:30-7:00pm. It will take place in William James Hall, Room #1550. The speaker is Adele Mortier (MIT), and the title and abstract of her talk can be found below. Food will be available at the meeting, and you can find the schedule for the remainder of the semester on the LangCog website.
Speaker: Adele Mortier
Title: How do Large Language Models process scalar Hurford Disjunctions?
Abstract: Disjunctions featuring mutually compatible disjuncts generally appear odd [Hurford, 1974, Singh, 2008]. For instance, #Mary owns a dog or a poodle is odd, because owning a poodle contextually entails owning a dog. Oddness has been argued to disappear when one of the disjuncts can be pragmatically strengthened to contradict the other [Gazdar, 1979]. For instance, the sentence Mary ate some or all of the biscuits appears fine because some can be understood to implicate not all. However, it also seems that these pragmatic “rescue” processes are not always available [Singh, 2008]. For instance, swapping the order of the disjunct in the previous sentence still leads to oddness: #Mary ate all or some of the biscuits, suggesting a pragmatic enrichment of some cannot take place in that case. Are statistical models of language sensitive to such linear asymmetries? And, if so, do they connect them to their model of the logical properties of the disjuncts? In this presentation, I propose to leverage a recent dataset for scalar implicatures [Jeretic et al., 2020] to tentatively answer these questions. I explore if LLMs (i) capture the infelicity of Hurford Disjunctions involving scalar items, in terms of surprisal (ii) identify an asymmetry between the two possible linear orders (again, in terms of surprisal) and (iii) exhibit a correlation between surprisal measures and compatibility/entailment score assigned to the relation between their disjuncts.