Tuesdays 10-1 in 32-D461
Professor F. Davis
Course Description
Seminar on current models and techniques in computational linguistics. Instruction provided in implementing and using models, including model training, evaluation, and interpretation. No background in programming or machine learning is assumed.
I explore how the insights of dynamic treatments of anaphora can be captured in a two-dimensional framework with a classical logic, where the "dynamic" effects of indefinites are parcelled off to a second dimension. The classicality of the resulting system means that it avoids well-known problems that face dynamic semantics from double negation and disjunction, while the dynamicness of the second dimension allows it to still capture the central insight of dynamic semantics: that indefinites can “bind" subsequent definites in an open-ended manner. The paper this is based on can be found here.
Indefinites exhibit two remarkable properties: they can take scope out of islands and they can be strengthened in certain configurations to convey universal meanings. In some cases, indefinites taking scope out of islands creates just the configurations required for their universal strengthening. We argue that these cases include donkey anaphora sentences — their meanings can be seen to emerge from an interplay of the two remarkable properties of indefinites.
Since the advent of event-anchored modal backgrounds, it is a matter of debate what the exact source of world quantification in certain structures is, with potential candidates ranging from embedding attitude verbs to complementizers and other peripheral elements in the embedded clause. One of these elements is mood, for which there are two current approaches in Romance: (i) mood as a world pronoun with presuppositional \phi-features (Schlenker 2005, Romero 2024) and (ii) mood as the locus of world quantification (Portner & Rubinstein 2020, Alonso-Ovalle, Menéndez-Benito & Rubinstein 2024).
To weigh in on this debate, the present talk examines modally interpreted Relative Clauses in Spanish like (1), where subjunctive mood leads to a purpose interpretation absent in its indicative counterpart (2):
(1) Ana le compró a Pedro una radio que le distrajera en el hospital.
Ana CL bought to Pedro a radio that him entertain.subj at the hospital
‘Ana bought Pedro a radio in order to entertain him at hospital’.
(2) Ana le compró a Pedro una radio que le distraerá en el hospital.
Ana CL bought to Pedro a radio that him entertain.ind at the hospital
‘Ana bought Pedro a radio that will entertain him at hospital’.
First, we will develop an analysis of mood in relative clauses within the world-pronoun approach (i). Then, we will examine an alternative analysis of the construction within the world-quantification approach (ii), due to Alonso-Ovalle et al. (2024). Finally, we will compare the two approaches and suggest that the world-pronoun approach (i) makes more accurate predictions.
A growing body of evidence suggests that whether or not attitude predicates may combine with interrogative complements is determined at least in part by some of their semantic properties. Zooming in, Uegaki & Sudo 2019 have claimed that non-veridical preferential predicates such as “hope" and “fear” do not combine with interrogative complements, but this empirical generalization as well as its proposed explanation have been challenged by a number of counterexamples (White 2021; Uegaki 2022), whose properties—beyond the fact that they feature embedded interrogatives—remain ill understood. In this talk, I will identify two types of counterexamples to the generalization, i.e., negative preferential predicates and question-oriented preferential predicates; and explore a theoretical account of their compatibility with interrogative complements. I will furthermore discuss how fine-grained cross-linguistic predictions of the account can be tested using the Crosslinguistic Database for Combinatorial and Semantic Properties of Attitude Predicates (Özyıldız, Qing, Roelofsen, Romero & Uegaki 2023).
This talk investigates when and how children figure out the clause types of their language and their canonical function, that is, the speech act that these clauses are typically used for. We hypothesize that the mapping between clause type and speech act is a crucial precursor to the acquisition of syntactic features that emerge before age two, and that infants figure out form and function in tandem, with their growing knowledge of each mutually constraining the other. I will report on some experiments suggesting that children have linked interrogatives with questions and declaratives with assertions by 18 months, and on corpus analyses and computational modeling of child-directed speech showing how learners can draw from pragmatic and syntactic information and the way they interact to figure out this mapping. (joint work with D. Goodhue, J. Lidz, E. Swanson, and Y. Yang)
