Saved in:
Bibliographic Details
Main Authors: Lo, Chun Hei, Lam, Wai, Cheng, Hong, Emerson, Guy
Format: Preprint
Published: 2023
Subjects:
Online Access:https://arxiv.org/abs/2309.08325
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866929240010653696
author Lo, Chun Hei
Lam, Wai
Cheng, Hong
Emerson, Guy
author_facet Lo, Chun Hei
Lam, Wai
Cheng, Hong
Emerson, Guy
contents Functional Distributional Semantics (FDS) models the meaning of words by truth-conditional functions. This provides a natural representation for hypernymy but no guarantee that it can be learnt when FDS models are trained on a corpus. In this paper, we probe into FDS models and study the representations learnt, drawing connections between quantifications, the Distributional Inclusion Hypothesis (DIH), and the variational-autoencoding objective of FDS model training. Using synthetic data sets, we reveal that FDS models learn hypernymy on a restricted class of corpus that strictly follows the DIH. We further introduce a training objective that both enables hypernymy learning under the reverse of the DIH and improves hypernymy detection from real corpora.
format Preprint
id arxiv_https___arxiv_org_abs_2309_08325
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle Distributional Inclusion Hypothesis and Quantifications: Probing for Hypernymy in Functional Distributional Semantics
Lo, Chun Hei
Lam, Wai
Cheng, Hong
Emerson, Guy
Computation and Language
Functional Distributional Semantics (FDS) models the meaning of words by truth-conditional functions. This provides a natural representation for hypernymy but no guarantee that it can be learnt when FDS models are trained on a corpus. In this paper, we probe into FDS models and study the representations learnt, drawing connections between quantifications, the Distributional Inclusion Hypothesis (DIH), and the variational-autoencoding objective of FDS model training. Using synthetic data sets, we reveal that FDS models learn hypernymy on a restricted class of corpus that strictly follows the DIH. We further introduce a training objective that both enables hypernymy learning under the reverse of the DIH and improves hypernymy detection from real corpora.
title Distributional Inclusion Hypothesis and Quantifications: Probing for Hypernymy in Functional Distributional Semantics
topic Computation and Language
url https://arxiv.org/abs/2309.08325