German Adjectives and Degree Modification
Dissertation Project
I am working on the Syntax and Semantics of degree modification in German adjectives. What is of particular interest to me is the German modifier_viel_ (compare English ''much'') and its selectional properties. Viel seems to select for events over states in verbs, nominals, and adjectives, yet it also selects for comparatives. I believe that deriving a consistent syntactic and semantic analysis of this pattern can inform on the status of adjectives generally and contribute to the discussion on Bresnan 1973's hypothesis on English much.
Language Evolution: CENA
Collaboration with: Anderson Almeida da Silva (Universidade Federal do Delta do Parnaíba), Shigeru Miyagawa (MIT and University of São Paulo), and Vitor Nóbrega (University of Hamburg)
CENA is an emergent sign language in Piauí, Brasil, that arose out of a homesign system some 70 years ago (Almeida da Silva & Nevins 2020). We are working on a joint research project to show that (i) humans do not need an external model of language, but can instead devise one of their own (an argument already made by Goldin-Meadow and colleagues in their work on homesign, see for example Brentari & Goldin-Meadow 2017), and (ii) that a homesign system can quickly progress into a full-fledged sign language if there is a community that uses it. While similar arguments have been made on the case of Nicaraguan Sign Language, we argue that CENA is different because it was not subject to any outside influences during its genesis due to the fact that it arose in a remote village with no access to other sign languages or sign language educators.
Informative Event Extraction for Dynamic Domains
Collaboration with: Yuwei Wang, Chen Chen (University of Arizona)
The team I lead within the ToMCAT project maintains a rule based event extraction (EE) system. The domain we work in consists of spontaneous speech within a game-like context. The game setting itself is subject to change with very little notice. This leads to a very dynamic domain that does not lend itself to deep learning approaches. We are looking at ways to utilize our rule-based system to fine-tune a deep learning system. Currently, we are exploring whether NLP outputs (Dialog Act Labels, Event Labels, Sentiment Labels) of a team-dialogue are predictive of the team members' attitudes towards their team and their conversations.