Balancing information-structure and semantic constraints on construction choice: building a computational model of passive and passive-like constructions in Mandarin Chinese



Liu, Li and Ambridge, Ben ORCID: 0000-0003-2389-8477
(2021) Balancing information-structure and semantic constraints on construction choice: building a computational model of passive and passive-like constructions in Mandarin Chinese. Cognitive Linguistics.

[img] Text
LiuAmbridge.docx - Accepted Version
Access to this file is embargoed until 15 April 2022.

Download (12MB)

Abstract

<jats:title>Abstract</jats:title> <jats:p>A central tenet of cognitive linguistics is that adults’ knowledge of language consists of a structured inventory of constructions, including various two-argument constructions such as the active (e.g., <jats:italic>Lizzy rescued John</jats:italic>), the passive (e.g., <jats:italic>John was rescued by Lizzy</jats:italic>) and “fronting” constructions (e.g., <jats:italic>John was the one Lizzy rescued)</jats:italic>. But how do speakers choose which construction to use for a particular utterance, given constraints such as discourse/information structure and the semantic fit between verb and construction? The goal of the present study was to build a computational model of this phenomenon for two-argument constructions in Mandarin. First, we conducted a grammaticality judgment study with 60 native speakers which demonstrated that, across 57 verbs, semantic affectedness – as determined by further 16 native speakers – predicted each verb’s relative acceptability in the <jats:italic>bei</jats:italic>-passive and <jats:italic>ba</jats:italic>-active constructions, but not the Notional Passive and SVO Active constructions. Second, in order to simulate acquisition of these competing constraints, we built a computational model that learns to map from corpus-derived input (information structure + verb semantics + lexical verb identity) to an output representation corresponding to these four constructions (+“other”). The model was able to predict judgments of the relative acceptability of the test verbs in the <jats:italic>ba</jats:italic>-active and <jats:italic>bei</jats:italic>-passive constructions obtained in Study 1, with model-human correlations in the region of <jats:italic>r</jats:italic> = 0.5 and <jats:italic>r</jats:italic> = 0.3, respectively. Surprisingly, these correlations increased (to <jats:italic>r</jats:italic> = 0.75 and <jats:italic>r</jats:italic> = 0.5 respectively) when lexical verb identity was removed; perhaps because this information leads to over-fitting of the training set. These findings suggest the intriguing possibility that acquiring constructions involves forgetting as a mechanism for abstracting across certain fine-grained lexical details and idiosyncrasies.</jats:p>

Item Type: Article
Divisions: Faculty of Health and Life Sciences
Faculty of Health and Life Sciences > Institute of Population Health
Depositing User: Symplectic Admin
Date Deposited: 26 Apr 2021 09:47
Last Modified: 30 Jun 2021 11:19
DOI: 10.1515/cog-2019-0100
URI: https://livrepository.liverpool.ac.uk/id/eprint/3120321