On Advancing Natural Language Interfaces: Data Collection, Model Development, and User Interaction.
저자
발행사항
Ann Arbor : ProQuest Dissertations & Theses, 2021
학위수여대학
The Ohio State University Computer Science and Engineering
수여연도
2021
작성언어
영어
주제어
학위
Ph.D.
페이지수
277 p.
지도교수/심사위원
Advisor: Sun, Huan.
Natural language provides a universal and efficient way for humans to express their intent and perceive the world. This inspires a surge of natural language interface (NLI) systems, which enable humans to acquire knowledge and solve problems using solely natural language. These include question answering systems such as the early BASEBALL system and IBM Watson, as well as virtual assistants such as Amazon Alexa, Apple Siri, Google Home, and Microsoft Cortana.Despite the remarkable progress, building NLIs that can reliably serve users in the long term has never been easy. In this dissertation, we characterize and study the three stages of the NLI life cycle: (1) Data Collection, where system developers collect training data to bootstrap the NLI system; (2) Model Development, where system developers design and implement the backend machine learning model, improving its capacity until its performance reaches the commercial grade. Note that both Data Collection and Model Development are before the system deployment. (3) User Interaction, where the NLI system is expected to interact with users and serve them reliably after its deployment. In this dissertation, we will first summarize the history and the status quo of the NLI study, as well as the challenges in each of the three stages. Following them, we will present our research achievements towards advancing NLIs in each stage.Specifically, in Part II, we will discuss solutions to improving the first two stages of the NLI construction (i.e., before deployment). We focus on constructing NLIs to code snippets, with applications in software engineering. Collecting training data for such specialized domains is typically expensive since domain expertise is needed from annotators. To address the problem, we explore training a machine learning model to automatically extract data from domain-specific online forums (e.g., Stack Overflow). Bootstrapped with a small amount of annotations, our model is trained and applied to collect a large-scale question-code dataset from Stack Overflow. The dataset has been widely adopted to facilitate follow-up research. Based on the dataset, we further study improving the NLI capacity via a novel "model collaborative framework". Specifically, we consider machine learning models for two tasks: code retrieval, which aims to retrieve code solutions from a code base to address the given question, and code annotation, whose goal is to generate a natural language sentence to describe a given code snippet. Instead of training them separately, we let the code retrieval model utilize the generated annotations as features to improve the retrieval performance, and leverage a retrieval-based signal to drive the code annotation model towards producing more useful annotations.In Part III, we will then show our research on building NLIs that can proactively interact with and learn from users (after deployment). This idea is instantiated in the task of semantic parsing, such as converting user utterances to Application Programming Interface (API) commands for automating smart-home services, or to Structured Query Language (SQL) queries for querying databases. Our preliminary study shows that user utterances in such applications could be very ambiguous. This inspires us to construct an interactive semantic parsing agent that can request user clarification whenever it finds uncertain about the user intent. We further propose MISP, a general-purpose interactive semantic parsing framework that can apply to various semantic parser architectures and parsing schemas. Our experimental results on the challenging text-to-SQL semantic parsing task show that, when a parser is allowed to interact with users, the SQL parsing accuracy can be significantly improved without additional system training. More importantly, as users now have a way to explore and open the blackbox of the system, user trust can be increased. Such interaction can also be leveraged as training signals for the semantic parsing agent to continually improve itself in the long-term deployment. Another intriguing problem derived from interactive semantic parsing is, when the parsing agent receives user feedback (which typically corresponds to the user intent of revising the previously generated semantic parse), how should it make the revision? In the last chapter of this part, we introduce our proposed model for incrementally editing tree-structured data (e.g., an Abstract Syntax Tree of a semantic parse) based on the given user feedback. By infusing interactivity, the research stimulates NLIs towards being more reliable and responsible to their decisions, and being adaptive and controllable to boost human productivity.In the final chapter, we will outline future research and conclude the dissertation. In particular, we will first summarize the key contributions in our existing work, and then present possible research directions along our defined three stages of the NLI life cycle. In summary, we seek to further relieve the burden of NLI construction and enable NLIs that can reliably serve and interact with users in practical applications.
분석정보
서지정보 내보내기(Export)
닫기소장기관 정보
닫기권호소장정보
닫기오류접수
닫기오류 접수 확인
닫기음성서비스 신청
닫기음성서비스 신청 확인
닫기이용약관
닫기학술연구정보서비스 이용약관 (2017년 1월 1일 ~ 현재 적용)
학술연구정보서비스(이하 RISS)는 정보주체의 자유와 권리 보호를 위해 「개인정보 보호법」 및 관계 법령이 정한 바를 준수하여, 적법하게 개인정보를 처리하고 안전하게 관리하고 있습니다. 이에 「개인정보 보호법」 제30조에 따라 정보주체에게 개인정보 처리에 관한 절차 및 기준을 안내하고, 이와 관련한 고충을 신속하고 원활하게 처리할 수 있도록 하기 위하여 다음과 같이 개인정보 처리방침을 수립·공개합니다.
주요 개인정보 처리 표시(라벨링)
목 차
3년
또는 회원탈퇴시까지5년
(「전자상거래 등에서의 소비자보호에 관한3년
(「전자상거래 등에서의 소비자보호에 관한2년
이상(개인정보보호위원회 : 개인정보의 안전성 확보조치 기준)개인정보파일의 명칭 | 운영근거 / 처리목적 | 개인정보파일에 기록되는 개인정보의 항목 | 보유기간 | |
---|---|---|---|---|
학술연구정보서비스 이용자 가입정보 파일 | 한국교육학술정보원법 | 필수 | ID, 비밀번호, 성명, 생년월일, 신분(직업구분), 이메일, 소속분야, 웹진메일 수신동의 여부 | 3년 또는 탈퇴시 |
선택 | 소속기관명, 소속도서관명, 학과/부서명, 학번/직원번호, 휴대전화, 주소 |
구분 | 담당자 | 연락처 |
---|---|---|
KERIS 개인정보 보호책임자 | 정보보호본부 김태우 | - 이메일 : lsy@keris.or.kr - 전화번호 : 053-714-0439 - 팩스번호 : 053-714-0195 |
KERIS 개인정보 보호담당자 | 개인정보보호부 이상엽 | |
RISS 개인정보 보호책임자 | 대학학술본부 장금연 | - 이메일 : giltizen@keris.or.kr - 전화번호 : 053-714-0149 - 팩스번호 : 053-714-0194 |
RISS 개인정보 보호담당자 | 학술진흥부 길원진 |
자동로그아웃 안내
닫기인증오류 안내
닫기귀하께서는 휴면계정 전환 후 1년동안 회원정보 수집 및 이용에 대한
재동의를 하지 않으신 관계로 개인정보가 삭제되었습니다.
(참조 : RISS 이용약관 및 개인정보처리방침)
신규회원으로 가입하여 이용 부탁 드리며, 추가 문의는 고객센터로 연락 바랍니다.
- 기존 아이디 재사용 불가
휴면계정 안내
RISS는 [표준개인정보 보호지침]에 따라 2년을 주기로 개인정보 수집·이용에 관하여 (재)동의를 받고 있으며, (재)동의를 하지 않을 경우, 휴면계정으로 전환됩니다.
(※ 휴면계정은 원문이용 및 복사/대출 서비스를 이용할 수 없습니다.)
휴면계정으로 전환된 후 1년간 회원정보 수집·이용에 대한 재동의를 하지 않을 경우, RISS에서 자동탈퇴 및 개인정보가 삭제처리 됩니다.
고객센터 1599-3122
ARS번호+1번(회원가입 및 정보수정)