Monthly Archives: November 2017

Lab Meeting (11/13/2017): Milena Rabovsky (Freie Universität Berlin)

For those who might be interested, Milena Rabovsky, a Postdoctoral Researcher (Marie Curie fellow) at the Neurocomputation and Neuroimaging Unit of Freie Universität Berlin (Germany), will present on work she conducted with McClelland at Stanford during Amitai Shenhav’s lab meeting. Details of her talk are as follows:

N400 amplitudes as change in a probabilistic representation of meaning: A neural network model

The N400 component of the event-related brain potential has aroused much interest because it is thought to provide an online measure of meaning processing in the brain. Yet, the underlying process of meaning construction remains incompletely understood. In the talk, I will present a computationally explicit account of this process and the emerging representation of sentence meaning. We simulate N400 amplitudes as the change induced by an incoming stimulus in an implicit and probabilistic representation of meaning captured by the hidden unit activation pattern in a neural network model of sentence comprehension, and propose that the process underlying the N400 also drives implicit learning in the network. We account for a broad range of empirically observed N400 effects which have previously been difficult to capture within a single integrated framework (Rabovsky, Hansen, & McClelland, bioRxiv).

More information about the speaker can be found here.

LingLang Lunch (11/10/2017): Hadas Kotek (New York University)

Hadas Kotek’s main research focus is on generative syntax and its interaction with formal semantics. Her recent research topics include the syntax and semantics of wh-questions, Association with Focus, relative clauses, ellipsis, and comparative and superlative quantifiers. For more information, her website is here.
 

Which QuD (joint work with Matt Barros)

Sluicing is ellipsis in a question, leaving only a wh-phrase overt (Ross 1969), e.g.: Sally called someone, but I don’t know who. Recent work on the identity conditions underlying the licensing of sluicing has converged on the need for a semantic approach to ellipsis licensing, where the sluiced question must be congruent to a Question under Discussion (QuD, Roberts 1996; e.g. Ginzburg and Sag, 2000; AnderBois, 2011; Weir, 2014; Barros, 2014; Kotek and Barros, To appear). In this talk, we address problems of over-generation predicted by this account, stemming from a more general concern: what is the source of QuDs, and how are they constrained?.

Adopting the notion of strategy trees and super/sub-questions from Rojas-Esponda (2014) (cf Büring 2003, Roberts 2012), we propose that sluicing is licensed by the most recently raised QuD in the discourse, and not by its super- or sub-questions, nor by unrelated QuDs. We show how this proposal accounts for several test-cases that are problematic for traditional approaches, including cases of sprouting (1), Dayal and Schwarzschild’s (2010) Antecedent Correlate Harmony generalization (2-3), and contrast sluicing (4). This approach furthermore provides a natural explanation for Barker’s (2014) ‘Answer Ban:’ the observation that the antecedent clause must not resolve, or even partially resolve, the issue raised by the sluiced interrogative.

(1) Sally left, but I don’t know {why, when, in which car, with whom, …}
(2) Joan ate a donut.
a. * Fred doesn’t know what.
b.    Fred doesn’t know which donut.
(3) Joan ate something.
a.    Fred doesn’t know what.
b. * Fred doesn’t know which donut.
(4) A:    Did(n’t) Mary call [Jack]F?
B: # I don’t know who.