Presentation
Search Abstracts | Symposia | Slide Sessions | Poster Sessions | Poster Slams
Deriving syntactic structure from novel noun-noun compounds
Poster C9 in Poster Session C, Friday, October 7, 10:15 am - 12:00 pm EDT, Millennium Hall
Phoebe Chen1, Arianna Zuanazzi1, David Poeppel1; 1New York University
Noun-noun compounds (NNCs, e.g., ‘orange juice’) are expressions generated from two root nouns (generally, modifier: ‘orange’, and head: ‘juice’) that act as one complex noun. While known NNCs’ meaning is learned, the interpretation of novel NNCs (e.g., ‘coffee tension’) relies on deriving the relation between the word constituents in the absence of syntactic cues. Previous studies investigating relational NNCs (i.e., NNCs where the nouns have a predicating relation), exclusively focused on the categories of relations that link the meanings of the constituent nouns (e.g., location relation: ‘mountain cloud’; cause relation: ‘college headache’). In our study, we adopted an approach that specifically investigates the syntactic relations between nouns in novel relational NNCs. In NNCs, the head is the subject of the underlying ‘sentence’, which can be in the active voice (e.g., ‘headache tension’ is: ‘tension that causes a headache’) or in the passive voice (e.g., ‘coffee tension’ is: ‘tension that is caused by coffee’). Here, we asked whether lexical-semantic features of NNCs (i.e., differences in frequency, agentivity and imageability between the constituents, and semantic similarity) predict their specific syntactic interpretation (i.e., active/passive). To address this question, we created a stimulus set consisting of 438 novel NNCs that span the distributions of the above linguistic features. In an online behavioral study, we collected free interpretations of our NNCs dataset from 147 participants (~20 participants for each NNC), along with RTs and eye-movements. Participants’ free interpretations show that agentivity and imageability are significant predictors of active/passive voice. In particular, the higher the agentivity difference between modifier and head, the more likely the underlying sentence is built in the passive voice. In addition, when both the head and modifier have high imageability, the underlying sentence is more likely built in the passive voice, suggesting that perceptual properties might also guide syntactic interpretation. None of the other lexical-semantic features were significant predictors of the syntactic relations. The RTs and eye-movement results show that participants spent more time and made more eye-movements between the two words when building the underlying structure in the passive voice. We interpret these results in light of the ‘agent-first’ strategy in sentence processing, which suggests that sentential subjects tend to be interpreted as agents: we hypothesize that heads of NNCs (which are subjects of the underlying sentence) are initially interpreted as agents, and only later reanalyzed as patients/themes, when the modifier has relatively higher agentivity compared to the head. Overall, our findings demonstrate that syntactic relations between the nouns in NNCs are determined by constituents’ agentivity and imageability. Furthermore, behavioral results suggest that the thematic role of the head is reanalyzed in the context of passive interpretation.
Topic Areas: Syntax, Meaning: Combinatorial Semantics