MoL-2015-08: Modelling Syntactic and Semantic Tasks with Linguistically Enriched Recursive Neural Networks

MoL-2015-08: Mallinson, Jonathan (2015) Modelling Syntactic and Semantic Tasks with Linguistically Enriched Recursive Neural Networks. [Report]

[img]
Preview
Text (Full Text)
MoL-2015-08.text.pdf

Download (1MB) | Preview
[img] Text (Abstract)
MoL-2015-08.abstract.txt

Download (1kB)

Abstract

In this thesis a compositional distributional semantic approach, the Recursive Neural Network, is used to syntactically-semantically compose non-symbolic representations of words. Unlike previous Recursive Neural Network models which use either no linguistic enrichment or significant symbolic syntactic enrichment, I propose minimal linguistic enrichments which are both semantic and syntactic. I achieve this by enriching the Recursive Neural Networks' models with core syntactic/semantic linguistic types: head, argument and adjunct. This approach brings together formal linguistics and computational linguistics, as such I give a broad account of these theories. The syntactic understanding of the model is tested by a parsing task and the semantic understanding is tested by a paraphrase detection task. The results of these tasks not only show the benefits of linguistic enrichment but also raise further questions of study.

Item Type: Report
Report Nr: MoL-2015-08
Series Name: Master of Logic Thesis (MoL) Series
Year: 2015
Uncontrolled Keywords: logic, language
Subjects: Logic
Date Deposited: 12 Oct 2016 14:38
Last Modified: 12 Oct 2016 14:38
URI: https://eprints.illc.uva.nl/id/eprint/949

Actions (login required)

View Item View Item