We introduce a new type of deep contextualized word representation that
models both (1) complex characteristics of word use (e.g., syntax and
semantics), and (2) how these uses vary across linguistic contexts (i.e., to
model polysemy). Our word vectors are learned functions of the internal states
of a deep bidirectional language model (biLM), which is pre-trained on a large
text corpus. We show that these representations can be easily added to existing
models and significantly improve the state of the art across six challenging
NLP problems, including question answering, textual entailment and sentiment
analysis. We also present an analysis showing that exposing the deep internals
of the pre-trained network is crucial, allowing downstream models to mix
different types of semi-supervision signals.
%0 Generic
%1 peters2018contextualized
%A Peters, Matthew E.
%A Neumann, Mark
%A Iyyer, Mohit
%A Gardner, Matt
%A Clark, Christopher
%A Lee, Kenton
%A Zettlemoyer, Luke
%D 2018
%K deep elmo embedding learning
%T Deep contextualized word representations
%U http://arxiv.org/abs/1802.05365
%X We introduce a new type of deep contextualized word representation that
models both (1) complex characteristics of word use (e.g., syntax and
semantics), and (2) how these uses vary across linguistic contexts (i.e., to
model polysemy). Our word vectors are learned functions of the internal states
of a deep bidirectional language model (biLM), which is pre-trained on a large
text corpus. We show that these representations can be easily added to existing
models and significantly improve the state of the art across six challenging
NLP problems, including question answering, textual entailment and sentiment
analysis. We also present an analysis showing that exposing the deep internals
of the pre-trained network is crucial, allowing downstream models to mix
different types of semi-supervision signals.
@misc{peters2018contextualized,
abstract = {We introduce a new type of deep contextualized word representation that
models both (1) complex characteristics of word use (e.g., syntax and
semantics), and (2) how these uses vary across linguistic contexts (i.e., to
model polysemy). Our word vectors are learned functions of the internal states
of a deep bidirectional language model (biLM), which is pre-trained on a large
text corpus. We show that these representations can be easily added to existing
models and significantly improve the state of the art across six challenging
NLP problems, including question answering, textual entailment and sentiment
analysis. We also present an analysis showing that exposing the deep internals
of the pre-trained network is crucial, allowing downstream models to mix
different types of semi-supervision signals.},
added-at = {2018-08-14T18:03:57.000+0200},
author = {Peters, Matthew E. and Neumann, Mark and Iyyer, Mohit and Gardner, Matt and Clark, Christopher and Lee, Kenton and Zettlemoyer, Luke},
biburl = {https://www.bibsonomy.org/bibtex/23562677774cce5dcf84a5cde13f0867d/hotho},
description = {Deep contextualized word representations},
interhash = {9c972d0afbdb8696a1e2b4ac4e18f88f},
intrahash = {3562677774cce5dcf84a5cde13f0867d},
keywords = {deep elmo embedding learning},
note = {cite arxiv:1802.05365Comment: NAACL 2018. Originally posted to openreview 27 Oct 2017. v2 updated for NAACL camera ready},
timestamp = {2018-08-14T18:03:57.000+0200},
title = {Deep contextualized word representations},
url = {http://arxiv.org/abs/1802.05365},
year = 2018
}