Exploiting word embeddings for modeling bilexical relations

There has been an exponential surge of text data in the recent years. As a consequence, unsupervised methods that make use of this data have been steadily growing in the field of natural language processing (NLP). Word embeddings are low-dimensional vectors obtained using unsupervised techniques on...

Descripción completa

Detalles Bibliográficos
Autor: Madhyastha, Pranava Swaroop
Tipo de recurso: tesis doctoral
Fecha de publicación:2017
País:España
Institución:Universitat Politècnica de Catalunya (UPC)
Repositorio:UPCommons. Portal del coneixement obert de la UPC
Idioma:inglés
OAI Identifier:oai:upcommons.upc.edu:2117/112211
Acceso en línea:https://hdl.handle.net/2117/112211
https://dx.doi.org/10.5821/dissertation-2117-112211
Access Level:acceso abierto
Palabra clave:Àrees temàtiques de la UPC::Informàtica
id ES_5ef6bd5e6fbca24bb68c032ea0f951c7
oai_identifier_str oai:upcommons.upc.edu:2117/112211
network_acronym_str ES
network_name_str España
repository_id_str
dc.title.none.fl_str_mv Exploiting word embeddings for modeling bilexical relations
title Exploiting word embeddings for modeling bilexical relations
spellingShingle Exploiting word embeddings for modeling bilexical relations
Madhyastha, Pranava Swaroop
Àrees temàtiques de la UPC::Informàtica
title_short Exploiting word embeddings for modeling bilexical relations
title_full Exploiting word embeddings for modeling bilexical relations
title_fullStr Exploiting word embeddings for modeling bilexical relations
title_full_unstemmed Exploiting word embeddings for modeling bilexical relations
title_sort Exploiting word embeddings for modeling bilexical relations
dc.creator.none.fl_str_mv Madhyastha, Pranava Swaroop
author Madhyastha, Pranava Swaroop
author_facet Madhyastha, Pranava Swaroop
author_role author
dc.contributor.none.fl_str_mv Quattoni, Ariadna Julieta
Carreras Pérez, Xavier
dc.subject.none.fl_str_mv Àrees temàtiques de la UPC::Informàtica
topic Àrees temàtiques de la UPC::Informàtica
description There has been an exponential surge of text data in the recent years. As a consequence, unsupervised methods that make use of this data have been steadily growing in the field of natural language processing (NLP). Word embeddings are low-dimensional vectors obtained using unsupervised techniques on the large unlabelled corpora, where words from the vocabulary are mapped to vectors of real numbers. Word embeddings aim to capture syntactic and semantic properties of words. In NLP, many tasks involve computing the compatibility between lexical items under some linguistic relation. We call this type of relation a bilexical relation. Our thesis defines statistical models for bilexical relations that centrally make use of word embeddings. Our principle aim is that the word embeddings will favor generalization to words not seen during the training of the model. The thesis is structured in four parts. In the first part of this thesis, we present a bilinear model over word embeddings that leverages a small supervised dataset for a binary linguistic relation. Our learning algorithm exploits low-rank bilinear forms and induces a low-dimensional embedding tailored for a target linguistic relation. This results in compressed task-specific embeddings. In the second part of our thesis, we extend our bilinear model to a ternary setting and propose a framework for resolving prepositional phrase attachment ambiguity using word embeddings. Our models perform competitively with state-of-the-art models. In addition, our method obtains significant improvements on out-of-domain tests by simply using word-embeddings induced from source and target domains. In the third part of this thesis, we further extend the bilinear models for expanding vocabulary in the context of statistical phrase-based machine translation. Our model obtains a probabilistic list of possible translations of target language words, given a word in the source language. We do this by projecting pre-trained embeddings into a common subspace using a log-bilinear model. We empirically notice a significant improvement on an out-of-domain test set. In the final part of our thesis, we propose a non-linear model that maps initial word embeddings to task-tuned word embeddings, in the context of a neural network dependency parser. We demonstrate its use for improved dependency parsing, especially for sentences with unseen words. We also show downstream improvements on a sentiment analysis task.
publishDate 2017
dc.date.none.fl_str_mv 2017
2017-07-25
2017
2017-12-12
dc.type.none.fl_str_mv doctoral thesis
http://purl.org/coar/resource_type/c_db06
VoR
http://purl.org/coar/version/c_970fb48d4fbd8a85
dc.type.openaire.fl_str_mv info:eu-repo/semantics/doctoralThesis
format doctoralThesis
dc.identifier.none.fl_str_mv https://hdl.handle.net/2117/112211
https://dx.doi.org/10.5821/dissertation-2117-112211
url https://hdl.handle.net/2117/112211
https://dx.doi.org/10.5821/dissertation-2117-112211
dc.language.none.fl_str_mv Inglés
eng
language_invalid_str_mv Inglés
language eng
dc.rights.none.fl_str_mv open access
http://purl.org/coar/access_right/c_abf2

http://creativecommons.org/licenses/by-nc-sa/4.0/
dc.rights.openaire.fl_str_mv info:eu-repo/semantics/openAccess
rights_invalid_str_mv open access
http://purl.org/coar/access_right/c_abf2

http://creativecommons.org/licenses/by-nc-sa/4.0/
eu_rights_str_mv openAccess
dc.format.none.fl_str_mv application/pdf
dc.publisher.none.fl_str_mv Universitat Politècnica de Catalunya
publisher.none.fl_str_mv Universitat Politècnica de Catalunya
dc.source.none.fl_str_mv reponame:UPCommons. Portal del coneixement obert de la UPC
instname:Universitat Politècnica de Catalunya (UPC)
instname_str Universitat Politècnica de Catalunya (UPC)
reponame_str UPCommons. Portal del coneixement obert de la UPC
collection UPCommons. Portal del coneixement obert de la UPC
repository.name.fl_str_mv
repository.mail.fl_str_mv
_version_ 1869409169221615616
spelling Exploiting word embeddings for modeling bilexical relationsMadhyastha, Pranava SwaroopÀrees temàtiques de la UPC::InformàticaThere has been an exponential surge of text data in the recent years. As a consequence, unsupervised methods that make use of this data have been steadily growing in the field of natural language processing (NLP). Word embeddings are low-dimensional vectors obtained using unsupervised techniques on the large unlabelled corpora, where words from the vocabulary are mapped to vectors of real numbers. Word embeddings aim to capture syntactic and semantic properties of words. In NLP, many tasks involve computing the compatibility between lexical items under some linguistic relation. We call this type of relation a bilexical relation. Our thesis defines statistical models for bilexical relations that centrally make use of word embeddings. Our principle aim is that the word embeddings will favor generalization to words not seen during the training of the model. The thesis is structured in four parts. In the first part of this thesis, we present a bilinear model over word embeddings that leverages a small supervised dataset for a binary linguistic relation. Our learning algorithm exploits low-rank bilinear forms and induces a low-dimensional embedding tailored for a target linguistic relation. This results in compressed task-specific embeddings. In the second part of our thesis, we extend our bilinear model to a ternary setting and propose a framework for resolving prepositional phrase attachment ambiguity using word embeddings. Our models perform competitively with state-of-the-art models. In addition, our method obtains significant improvements on out-of-domain tests by simply using word-embeddings induced from source and target domains. In the third part of this thesis, we further extend the bilinear models for expanding vocabulary in the context of statistical phrase-based machine translation. Our model obtains a probabilistic list of possible translations of target language words, given a word in the source language. We do this by projecting pre-trained embeddings into a common subspace using a log-bilinear model. We empirically notice a significant improvement on an out-of-domain test set. In the final part of our thesis, we propose a non-linear model that maps initial word embeddings to task-tuned word embeddings, in the context of a neural network dependency parser. We demonstrate its use for improved dependency parsing, especially for sentences with unseen words. We also show downstream improvements on a sentiment analysis task.En els darrers anys hi ha hagut un sorgiment notable de dades en format textual. Conseqüentment, en el camp del Processament del Llenguatge Natural (NLP, de l'anglès "Natural Language Processing") s'han desenvolupat mètodes no supervistats que fan ús d'aquestes dades. Els anomenats "word embeddings", o embeddings de paraules, són vectors de dimensionalitat baixa que s'obtenen mitjançant tècniques no supervisades aplicades a corpus textuals de grans volums. Com a resultat, cada paraula del diccionari es correspon amb un vector de nombres reals, el propòsit del qual és capturar propietats sintàctiques i semàntiques de la paraula corresponent. Moltes tasques de NLP involucren calcular la compatibilitat entre elements lèxics en l'àmbit d'una relació lingüística. D'aquest tipus de relació en diem relació bilèxica. Aquesta tesi proposa models estadístics per a relacions bilèxiques que fan ús central d'embeddings de paraules, amb l'objectiu de millorar la generalització del model lingüístic a paraules no vistes durant l'entrenament. La tesi s'estructura en quatre parts. A la primera part presentem un model bilineal sobre embeddings de paraules que explota un conjunt petit de dades anotades sobre una relaxió bilèxica. L'algorisme d'aprenentatge treballa amb formes bilineals de poc rang, i indueix embeddings de poca dimensionalitat que estan especialitzats per la relació bilèxica per la qual s'han entrenat. Com a resultat, obtenim embeddings de paraules que corresponen a compressions d'embeddings per a una relació determinada. A la segona part de la tesi proposem una extensió del model bilineal a trilineal, i amb això proposem un nou model per a resoldre ambigüitats de sintagmes preposicionals que usa només embeddings de paraules. En una sèrie d'avaluacións, els nostres models funcionen de manera similar a l'estat de l'art. A més, el nostre mètode obté millores significatives en avaluacions en textos de dominis diferents al d'entrenament, simplement usant embeddings induïts amb textos dels dominis d'entrenament i d'avaluació. A la tercera part d'aquesta tesi proposem una altra extensió dels models bilineals per ampliar la cobertura lèxica en el context de models estadístics de traducció automàtica. El nostre model probabilístic obté, donada una paraula en la llengua d'origen, una llista de possibles traduccions en la llengua de destí. Fem això mitjançant una projecció d'embeddings pre-entrenats a un sub-espai comú, usant un model log-bilineal. Empíricament, observem una millora significativa en avaluacions en dominis diferents al d'entrenament. Finalment, a la quarta part de la tesi proposem un model no lineal que indueix una correspondència entre embeddings inicials i embeddings especialitzats, en el context de tasques d'anàlisi sintàctica de dependències amb models neuronals. Mostrem que aquest mètode millora l'analisi de dependències, especialment en oracions amb paraules no vistes durant l'entrenament. També mostrem millores en un tasca d'anàlisi de sentiment.Universitat Politècnica de CatalunyaQuattoni, Ariadna JulietaCarreras Pérez, Xavier20172017-07-2520172017-12-12doctoral thesishttp://purl.org/coar/resource_type/c_db06VoRhttp://purl.org/coar/version/c_970fb48d4fbd8a85info:eu-repo/semantics/doctoralThesisapplication/pdfhttps://hdl.handle.net/2117/112211https://dx.doi.org/10.5821/dissertation-2117-112211reponame:UPCommons. Portal del coneixement obert de la UPCinstname:Universitat Politècnica de Catalunya (UPC)Inglésengopen accesshttp://purl.org/coar/access_right/c_abf2http://creativecommons.org/licenses/by-nc-sa/4.0/info:eu-repo/semantics/openAccessoai:upcommons.upc.edu:2117/1122112026-05-27T15:37:01Z
score 15.301603

Similares en LA Referencia