Research

My research programme asks how the parser navigates and represents sentential structure in real time upon encountering an ambiguous string in the context of certain other linguistic and extra-linguistic information. In other words, how do we use contextual information or world knowledge during parsing?

Reliability and familiarity

Prosodic boundaries may indicate syntactic constituent boundaries, but they are produced with high variability depending on many complex factors. When we speak out of breath, or if we shout across a loud room, or when we’re uncertain, or in a brainfog, or reading, or if we have a speech or language impairment… those boundaries may not occur in the most informative locations. How hard is it to accommodate these anomalous productions? If we’re expecting anomalous prosodic boundaries, do we use prosodic information in the same way as if we’re expecting typical or syntactically informative prosodic boundaries?

It is not clear whether or not the parser will use such information in its initial representations. If the parser only takes prosodic boundaries into account when other information has already been processed, then unusual or inconsistent prosodic boundaries in structurally ambiguous sentences won’t change the parser’s behavior while it builds an initial representation of a sentence. However, if the parser uses prosodic information early on to determine syntactic constituents, then unusual or inconsistent prosodic boundaries could confuse or mislead the parser and create notable slow-downs associated with garden-path effects.

This is particularly useful if we’re trying to understand how (or if!) we adapt our real-time comprehension to the world around us. If this line of research bears fruit, it could be useful in the training of people who care for those with speech or language impairments. Since certain impairments (e.g., aphasia, apraxia) can make understanding the speaker challenging (especially if the listener is unfamiliar with the speaker’s patterns), I hope my research can inform how we become familiar with new and anomalous speech patterns.

Gender and categories

In coreference resolution between a pronoun and a name, there is often an assumption that gender is a formal feature of the [English] word, thus is processed with other lexical/grammatical features. Furthermore, there is an increase in recognition of nonbinary genders and nonbinary pronouns (‘neopronouns’). It is not clear how one’s exposure to nonbinary genders and neopronoun use will influence real-time, early long-distance dependency formation. On one hand, pronouns are a conservative closed class in English, so it may be difficult to influence ingrained processing behavior associated with coreference dependency formation. On the other hand, meaningful and consistent exposure to people with non-binary genders and people who use so-called neopronouns may be sufficient to influence the lexical category and the way in which the earliest stages syntactic representation are built. Here is my AMLaP2017 poster that supports this latter possibility. I’ve also written a squib that describes how I suggest researchers approach questions concerning the human gender space and how we talk about human gender (explicitly and implicitly).

As a spin-off of this topic, I am also examining the possibility of a multidimensional “gender space” in which gender identities and personal names might be mapped. The objective of this is to provide more accurate and inclusive tools to researchers looking to examine the effects of linguistic behavior as a function of gender identity (e.g., one’s own identity or the encoding of another individual’s identity).

plot of shared (masc/fem) baby names (1996-2000)
Preliminary analysis of distribution of “shared” names registered to male and female babies in England and Wales from 1996 to 2000.

 

Advertisements