Nouns Are Vectors, Adjectives Are Matrices: Representing Adjective-noun Constructions in Semantic Space
M. Baroni, and R. Zamparelli. Proceedings of the 2010 Conference on Empirical Methods in Natural Language Processing, page 1183--1193. Stroudsburg, PA, USA, Association for Computational Linguistics, (2010)
Abstract
We propose an approach to adjective-noun composition (AN) for corpus-based distributional semantics that, building on insights from theoretical linguistics, represents nouns as vectors and adjectives as data-induced (linear) functions (encoded as matrices) over nominal vectors. Our model significantly outperforms the rivals on the task of reconstructing AN vectors not seen in training. A small post-hoc analysis further suggests that, when the model-generated AN vector is not similar to the corpus-observed AN vector, this is due to anomalies in the latter. We show moreover that our approach provides two novel ways to represent adjective meanings, alternative to its representation via corpus-based co-occurrence vectors, both outperforming the latter in an adjective clustering task.
%0 Conference Paper
%1 baroni2010nouns
%A Baroni, Marco
%A Zamparelli, Roberto
%B Proceedings of the 2010 Conference on Empirical Methods in Natural Language Processing
%C Stroudsburg, PA, USA
%D 2010
%I Association for Computational Linguistics
%K adjectives bag bow compositional distributional matrices matrix nouns of semantics vectors words
%P 1183--1193
%T Nouns Are Vectors, Adjectives Are Matrices: Representing Adjective-noun Constructions in Semantic Space
%U http://dl.acm.org/citation.cfm?id=1870658.1870773
%X We propose an approach to adjective-noun composition (AN) for corpus-based distributional semantics that, building on insights from theoretical linguistics, represents nouns as vectors and adjectives as data-induced (linear) functions (encoded as matrices) over nominal vectors. Our model significantly outperforms the rivals on the task of reconstructing AN vectors not seen in training. A small post-hoc analysis further suggests that, when the model-generated AN vector is not similar to the corpus-observed AN vector, this is due to anomalies in the latter. We show moreover that our approach provides two novel ways to represent adjective meanings, alternative to its representation via corpus-based co-occurrence vectors, both outperforming the latter in an adjective clustering task.
@inproceedings{baroni2010nouns,
abstract = {We propose an approach to adjective-noun composition (AN) for corpus-based distributional semantics that, building on insights from theoretical linguistics, represents nouns as vectors and adjectives as data-induced (linear) functions (encoded as matrices) over nominal vectors. Our model significantly outperforms the rivals on the task of reconstructing AN vectors not seen in training. A small post-hoc analysis further suggests that, when the model-generated AN vector is not similar to the corpus-observed AN vector, this is due to anomalies in the latter. We show moreover that our approach provides two novel ways to represent adjective meanings, alternative to its representation via corpus-based co-occurrence vectors, both outperforming the latter in an adjective clustering task.},
acmid = {1870773},
added-at = {2014-08-26T17:39:49.000+0200},
address = {Stroudsburg, PA, USA},
author = {Baroni, Marco and Zamparelli, Roberto},
biburl = {https://www.bibsonomy.org/bibtex/2bf6de79ebf28455054d2718708456944/jil},
booktitle = {Proceedings of the 2010 Conference on Empirical Methods in Natural Language Processing},
description = {Nouns are vectors, adjectives are matrices},
interhash = {272c5513b7d14e2529a1b9bb06eb9e8c},
intrahash = {bf6de79ebf28455054d2718708456944},
keywords = {adjectives bag bow compositional distributional matrices matrix nouns of semantics vectors words},
location = {Cambridge, Massachusetts},
numpages = {11},
pages = {1183--1193},
publisher = {Association for Computational Linguistics},
series = {EMNLP '10},
timestamp = {2014-08-26T17:39:49.000+0200},
title = {Nouns Are Vectors, Adjectives Are Matrices: Representing Adjective-noun Constructions in Semantic Space},
url = {http://dl.acm.org/citation.cfm?id=1870658.1870773},
year = 2010
}