東華大學圖書館 |

Deep Contextual Modeling for Natural Language Understanding, Generation, and Grounding.

紀錄類型:	書目-電子資源 : Monograph/item
正題名/作者:	Deep Contextual Modeling for Natural Language Understanding, Generation, and Grounding./
作者:	Zhang, Rui.
出版者:	Ann Arbor : ProQuest Dissertations & Theses, : 2020,
面頁冊數:	170 p.
附註:	Source: Dissertations Abstracts International, Volume: 82-12, Section: B.
Contained By:	Dissertations Abstracts International82-12B.
標題:	Computer science. -
電子資源:	http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=27739047
ISBN:	9798516931536

Deep Contextual Modeling for Natural Language Understanding, Generation, and Grounding.
Zhang, Rui.

Deep Contextual Modeling for Natural Language Understanding, Generation, and Grounding. - Ann Arbor : ProQuest Dissertations & Theses, 2020 - 170 p.

Source: Dissertations Abstracts International, Volume: 82-12, Section: B.

Thesis (Ph.D.)--Yale University, 2020.

This item must not be sold to any third party vendors.

Natural language is a fundamental form of information and communication.In both human-human and human-computer communication, people reason about the context of text and world state to understand language and produce language response.Natural Language Processing aims to build intelligent systems that can understand language, generate language, and ground language in other forms such as formal programs.This dissertation aims to present several deep neural network based systems that first understand the meaning of language grounded in various contexts where the language is used, and then generate effective language responses in different forms for information retrieval and human-computer communication.In particular, the following research objectives and challenges are addressed:(1) Deep Neural Modeling of Text Units. We present several end-to-end deep neural networks for (i) entity extraction and coreference resolution in documents, (ii) sentiment analysis and text classification for sentences and documents, and (iii) addressee and response selection for multi-turn multi-party conversations based on explicit representations for different discourse participants.(2) Text Summarization for Generating Email Subject Lines.We create the first dataset for this task and find that email subject line generation favors extremely abstractive summary, differentiating it from news headline generation or news single document summarization.For proper evaluation, we build a neural network to score the quality of an email subject given the email body.To summarize messages to short subjects with a high compression ratio, we combine the extractive and abstractive approaches and employ a multi-stage training strategy with supervised pretraining and reinforcement learning to optimize the email subject quality scores.(3) Neural Learning-to-rank for Low-Resource Cross-lingual Information Retrieval.We propose to combine evidence from different query and document translations by using cross-lingual word embeddings and deep relevance ranking models in a low-resource setting.By including the query likelihood retrieval score as an extra feature, the model effectively learns to rerank from only a few hundred relevance labels.In addition, by aligning word embedding spaces for multiple languages, the model can be directly applied under a zero-shot transfer setting when no training data is available for another language pair.(4) Multi-turn Text-to-SQL Generation by Language Grounding to Relational Databases.Generating SQL queries from user utterances is important to help people acquire information from databases.Furthermore, in real-world applications, users often access information in a multi-turn interaction with the system by asking a sequence of related questions.We propose an editing-based model for the cross-domain multi-turn text-to-SQL task.Observing that adjacent natural language questions are often linguistically dependent and their corresponding SQL queries tend to overlap, we utilize the interaction history by editing the previous predicted query to improve the generation quality.Moreover, to deal with complex table structures in different domains, we employ an utterance-table encoder and a table-aware decoder to incorporate the context of the user utterance and the table schema.

ISBN: 9798516931536Subjects--Topical Terms:

523869
Computer science.
Subjects--Index Terms:

Human-human communication

Deep Contextual Modeling for Natural Language Understanding, Generation, and Grounding.
LDR:04408nmm a2200337 4500 001 2343784
005 20220513114317.5
008 241004s2020 ||||||||||||||||| ||eng d
020 $a 9798516931536
035 $a (MiAaPQ)AAI27739047
035 $a AAI27739047
040 $a MiAaPQ $c MiAaPQ
100 1 $a Zhang, Rui. $3 1273551
245 1 0 $a Deep Contextual Modeling for Natural Language Understanding, Generation, and Grounding.
260 1 $a Ann Arbor : $b ProQuest Dissertations & Theses, $c 2020
300 $a 170 p.
500 $a Source: Dissertations Abstracts International, Volume: 82-12, Section: B.
500 $a Advisor: Radev, Dragomir R.
502 $a Thesis (Ph.D.)--Yale University, 2020.
506 $a This item must not be sold to any third party vendors.
520 $a Natural language is a fundamental form of information and communication.In both human-human and human-computer communication, people reason about the context of text and world state to understand language and produce language response.Natural Language Processing aims to build intelligent systems that can understand language, generate language, and ground language in other forms such as formal programs.This dissertation aims to present several deep neural network based systems that first understand the meaning of language grounded in various contexts where the language is used, and then generate effective language responses in different forms for information retrieval and human-computer communication.In particular, the following research objectives and challenges are addressed:(1) Deep Neural Modeling of Text Units. We present several end-to-end deep neural networks for (i) entity extraction and coreference resolution in documents, (ii) sentiment analysis and text classification for sentences and documents, and (iii) addressee and response selection for multi-turn multi-party conversations based on explicit representations for different discourse participants.(2) Text Summarization for Generating Email Subject Lines.We create the first dataset for this task and find that email subject line generation favors extremely abstractive summary, differentiating it from news headline generation or news single document summarization.For proper evaluation, we build a neural network to score the quality of an email subject given the email body.To summarize messages to short subjects with a high compression ratio, we combine the extractive and abstractive approaches and employ a multi-stage training strategy with supervised pretraining and reinforcement learning to optimize the email subject quality scores.(3) Neural Learning-to-rank for Low-Resource Cross-lingual Information Retrieval.We propose to combine evidence from different query and document translations by using cross-lingual word embeddings and deep relevance ranking models in a low-resource setting.By including the query likelihood retrieval score as an extra feature, the model effectively learns to rerank from only a few hundred relevance labels.In addition, by aligning word embedding spaces for multiple languages, the model can be directly applied under a zero-shot transfer setting when no training data is available for another language pair.(4) Multi-turn Text-to-SQL Generation by Language Grounding to Relational Databases.Generating SQL queries from user utterances is important to help people acquire information from databases.Furthermore, in real-world applications, users often access information in a multi-turn interaction with the system by asking a sequence of related questions.We propose an editing-based model for the cross-domain multi-turn text-to-SQL task.Observing that adjacent natural language questions are often linguistically dependent and their corresponding SQL queries tend to overlap, we utilize the interaction history by editing the previous predicted query to improve the generation quality.Moreover, to deal with complex table structures in different domains, we employ an utterance-table encoder and a table-aware decoder to incorporate the context of the user utterance and the table schema.
590 $a School code: 0265.
650 4 $a Computer science. $3 523869
650 4 $a Language. $3 643551
650 4 $a Information technology. $3 532993
653 $a Human-human communication
653 $a Human-computer communication
690 $a 0984
690 $a 0489
690 $a 0679
710 2 $a Yale University. $b Computer Science. $3 3682254
773 0 $t Dissertations Abstracts International $g 82-12B.
790 $a 0265
791 $a Ph.D.
792 $a 2020
793 $a English
856 4 0 $u http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=27739047