Representation of Word Meaning in the Intermediate Projection Layer of a Neural Language Model

Steven Derby; Paul Miller; Brian Murphy; Barry Devereux

doi:10.18653/v1/w18-5449

Representation of Word Meaning in the Intermediate Projection Layer of a Neural Language Model

Steven Derby, Paul Miller, Brian Murphy, Barry Devereux

School of Electronics, Electrical Engineering and Computer Science

Research output: Contribution to conference › Paper › peer-review

167 Downloads (Pure)

Abstract

Performance in language modelling has been significantly improved by training recurrent neural networks on large corpora. This progress has come at the cost of interpretability and an understanding of how these architectures function, making principled development of better language models more difficult. We look inside a state-of-the-art neural language model to analyse how this model represents high-level lexico-semantic information. In particular, we investigate how the model represents words by extracting activation patterns where they occur in the text, and compare these representations directly to human semantic knowledge.

Original language	English
Pages	362-364
Number of pages	3
DOIs	https://doi.org/10.18653/v1/w18-5449
Publication status	Published - 01 Nov 2018
Event	BlackBoxNLP 2018: Workshop on analyzing and interpreting neural networks for NLP - EMNLP 2018, Brussels, Belgium Duration: 31 Oct 2018 → 01 Nov 2018 https://blackboxnlp.github.io/

Workshop

Workshop	BlackBoxNLP 2018: Workshop on analyzing and interpreting neural networks for NLP
Abbreviated title	BlackBoxNLP 2018
Country/Territory	Belgium
City	Brussels
Period	31/10/2018 → 01/11/2018
Internet address	https://blackboxnlp.github.io/

Access to Document

10.18653/v1/w18-5449Licence: CC BY

Representation of Word Meaning in the Intermediate Projection Layer of a Neural Language Model
Copyright 2018 the authors. This is an open access article published under a Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution and reproduction in any medium, provided the author and source are cited.
Final published version, 166 KBLicence: CC BY

Interpretable semantic representations from neural language models and computer vision
Derby, S. (Author), Murphy, B. (Supervisor), Miller, P. (Supervisor) & Devereux, B. (Supervisor), Jul 2022
Student thesis: Doctoral Thesis › Doctor of Philosophy

File

Cite this

@conference{16f8ca96ed404b23b843f7471e13567a,

title = "Representation of Word Meaning in the Intermediate Projection Layer of a Neural Language Model",

abstract = "Performance in language modelling has been significantly improved by training recurrent neural networks on large corpora. This progress has come at the cost of interpretability and an understanding of how these architectures function, making principled development of better language models more difficult. We look inside a state-of-the-art neural language model to analyse how this model represents high-level lexico-semantic information. In particular, we investigate how the model represents words by extracting activation patterns where they occur in the text, and compare these representations directly to human semantic knowledge.",

author = "Steven Derby and Paul Miller and Brian Murphy and Barry Devereux",

year = "2018",

month = nov,

day = "1",

doi = "10.18653/v1/w18-5449",

language = "English",

pages = "362--364",

note = "BlackBoxNLP 2018: Workshop on analyzing and interpreting neural networks for NLP, BlackBoxNLP 2018 ; Conference date: 31-10-2018 Through 01-11-2018",

url = "https://blackboxnlp.github.io/",

}

TY - CONF

T1 - Representation of Word Meaning in the Intermediate Projection Layer of a Neural Language Model

AU - Derby, Steven

AU - Miller, Paul

AU - Murphy, Brian

AU - Devereux, Barry

PY - 2018/11/1

Y1 - 2018/11/1

N2 - Performance in language modelling has been significantly improved by training recurrent neural networks on large corpora. This progress has come at the cost of interpretability and an understanding of how these architectures function, making principled development of better language models more difficult. We look inside a state-of-the-art neural language model to analyse how this model represents high-level lexico-semantic information. In particular, we investigate how the model represents words by extracting activation patterns where they occur in the text, and compare these representations directly to human semantic knowledge.

AB - Performance in language modelling has been significantly improved by training recurrent neural networks on large corpora. This progress has come at the cost of interpretability and an understanding of how these architectures function, making principled development of better language models more difficult. We look inside a state-of-the-art neural language model to analyse how this model represents high-level lexico-semantic information. In particular, we investigate how the model represents words by extracting activation patterns where they occur in the text, and compare these representations directly to human semantic knowledge.

U2 - 10.18653/v1/w18-5449

DO - 10.18653/v1/w18-5449

M3 - Paper

SP - 362

EP - 364

T2 - BlackBoxNLP 2018: Workshop on analyzing and interpreting neural networks for NLP

Y2 - 31 October 2018 through 1 November 2018

ER -

Representation of Word Meaning in the Intermediate Projection Layer of a Neural Language Model

Abstract

Workshop

Access to Document

Fingerprint

Student theses

Interpretable semantic representations from neural language models and computer vision

Cite this