The Weekly Newsletter of MIT Linguistics

Issue of Monday, April 8th, 2024

Rawski @ Rutgers colloquium

Last week, Jon Rawski (visiting Assistant Professor) gave an invited colloquium talk at Rutgers Linguistics. The talk was titled Mathematical Linguistics & Cognitive Complexity

Abstract: Mathematics is the study of structures, and mathematical linguistics studies linguistic structures. Generative linguistics put the focus squarely on grammars: finite abstractions of the highly structured and complex mental computations of linguistic phenomena. The focus of this talk is the expressive power of a grammar, which directly measures the complexity of any cognitive system which instantiates it. Two key algebraic restrictions have emerged: regularity, and transduction (mappings between finite structures). I will use these notions to describe upper and lower bounds on the weak and strong capacity of human grammars (the kinds of phenomena they cover, and the kinds structures they posit during a derivation), as well as mismatches between them. These bounds enable precise theory comparison, and I will show how many linguistic theories drastically overgenerate, using reduplication as a case study. I will then consider the expressivity of so-called ‘neural language models’.  I will show, using finite model theory and multilinear algebra, how regular transductions can be embedded into the tensor representations used by neural computation and Optimality Theory. I will then present new theoretical bounds on the capacity of language models, connecting various transformer models to classes of first-order transductions. The overall picture that emerges is that, under the lens of mathematical abstraction, linguistic complexity is indeed a window into fundamental aspects of cognition and computation.

LF Reading Group 4/10 - Shrayana Haldar (MIT)

Speaker: Shrayana Haldar (MIT)
Title:Fixing Engdahl’s Type-Shifter and Heim’s Unary Which
Time: Wednesday, April 10th, 1pm - 2pm
Location: 32-D461

Abstract: Engdahl’s (1986) account of functional readings of sentences like (1) involved having the pronoun herself bound upstairs by a covert binder E, given in (2), while having a totally impoverished trace (i.e., just “t”, without any restrictor) with a complex index; that is, “tf(x) “. The E operator also does the job of shifting the restrictor to type ⟨ee, t⟩, without which the functional reading wouldn’t be possible. Heim (2019) pointed out that having the pronoun bound upstairs like this cannot derive ϕ-feature agreement between the pronoun and the antecedent quantifier, in this case, no girl, and binding theoretic effects like *Which picture of herself1 did no girl’s1 father submit?. Motivated by this reason, she proposed to have an LF like the one in (3), where which is unary (because it attaches directly and only to the question skeleton) and polymorphic (because it needs to be able to quantify over ⟨e, e⟩-type functions). Moreover, she proposed to have the whole restrictor picture of herself is in situ, getting syntactically bound by no girl, thereby avoiding the ϕ-featural and Binding Theoretic issues.

(1) Which picture of herself1 did no girl1 submit?
      Functional reading:
      Which function, fee, that maps entities to a pictures of those entities, is such that, for no girl, xx submitted f(x)?
      Possible answer:
      Her wedding picture.
(2) 〚Ey ζg ,w = λfee . ∀x . 〚ζgx/y(f(x)) = 1
(3) [which] did no girl1 submit [picture of herself1]?
My claim in this talk is that (a) which can’t be unary after all, because, for sentences like (4) — where a functional reading is equally possible — we need LFs like (5), where which does need a restrictor upstairs, the relative clause that’s late merged/not neglected (depending on what view one subscribes to); and (b) to interpret such structures, we do need a covert morpheme very much like (2), but slightly different in that the assignment function isn’t modified in the metalanguage in the lexical entry (6).\ This covert morpheme, that I call , is necessary to shift the ⟨e, t⟩-type relative clause to an ⟨ee, t⟩-type predicate, which will make the functional reading possible.
(4) Which picture that John1 liked did he1 show no girl?
      Functional reading:
      Which function, fee, that maps entities to entities that John liked, is such that, for no girl, x, John showed x f(x)?
      Possible answer:
      The picture she hated.
(5) [which [that John1 liked]] did he1 show no girl2 [picture [f pro2]]
(6) 〚〛= λPet . λf ee . ∀x[x ∈ codom(f) → P(x) = 1]
Time permitting, I will discuss metasemantic motivations for ruling out the possibility of lexical entries like (2), while preserving the possibility of having lexical entries like (6). I will couch this is in terms of a specific limitation that semantic reconstruction has been argued to be subject to and I will show that it’s exactly the machinery that’s required for this forbidden kind of semantic reconstruction that’s also required to categorematize Engdahl’s E operator. This, I will argue, supports my claim that entries like (6) are permitted in natural language, while entries like (2) are not.

Pesetsky @ Berlin Adverbials conference

Last weekend, David Pesetsky presented a paper entitled “A double life for complement if-clauses revisited” at the Fourth International Conference on Adverbial Clauses at the Free University, Berlin. Alum (PhD 1998) Susi Wurmbrand (University of Salzburg) also presented at the conference, with a paper entitled “A syntactic approach to tense in complementation and beyond”.

Phonology Circle 4/8 - Dóra Takács (MIT)

Speaker: Dóra Takács (MIT)
Title: Vowel-zero alternations in Hungarian
Time: Monday, April 8th, 5pm - 6:30pm
Location: 32-D831

Abstract: There are a closed class of about 500 stems in Hungarian that are subject to vowel-zero alternations if they are followed by a potentially vowel-initial suffix (Siptár & Törkenczy 2000). Early proposals (Vago 1980, Törkenczy 1992, Siptár & Törkenczy 2000, Abrusan 2005) mostly focused on whether this process is a result of epenthesis, metathesis or syncope . Some patterns among the consonants in these exceptional stems have been previously noted in the literature, but these observations were not integrated in the previous analyses. In this talk I show how stem-internal vowel-zero alternation in Hungarian interacts with voicing assimilation, affrication and gemination and use these interactions to support the claim that this process is in fact a result of syncope.


The 14th Formal Approaches to South Asian Languages, FASAL 14, was held at Stony Brook University on April 4—6, 2024. MIT linguistics was well-represented by the following members of the department. Unverified rumour has it that the earthquake in New York City on April 5 has something to do with their groundbreaking research. 

  •  Gurmeet Kaur and Yash Sinha (5th year): A novel account of mixed concord: the view from Punjabi honorifics
  • Shrayana Haldar (3rd year): Threefold ambiguity between Strong Necessity, Permission, and Weak Necessity in a Benjali Modal
  • Ruoan Wang (5th year): Tiered honorification in Eastern Indo-Aryan: [Hon]-less, and also presuppositionless
  • Athulya Aravind (Faculty): Malayalam long-distance anaphor taan: a null theory (Invited Talk)