Báo cáo khoa học: "Semantic Role Labeling: Past, Present and Future"

Chia sẻ: Hongphan_1 Hongphan_1 | Ngày: | Loại File: PDF | Số trang:1

Thêm vào BST

Báo xấu

55
lượt xem 2
download

Download Vui lòng tải xuống để xem tài liệu đầy đủ

Semantic Role Labeling (SRL) consists of, given a sentence, detecting basic event structures such as “who” did “what” to “whom”, “when” and “where”. From a linguistic point of view, a key component of the task corresponds to identifying the semantic arguments ﬁlling the roles of the sentence predicates. Typical predicate semantic arguments include Agent, Patient, and Instrument, but semantic roles may also be found as adjuncts (e.g., Locative, Temporal, Manner, and Cause). The identiﬁcation of such event frames holds potential for signiﬁcant impact in many NLP applications, such as Information Extraction, Question Answering, Summarization and Machine Translation. Recently, the...

Chủ đề:

Bình luận(0) Đăng nhập để gửi bình luận!

Lưu

Nội dung Text: Báo cáo khoa học: "Semantic Role Labeling: Past, Present and Future"

Semantic Role Labeling: Past, Present and Future Llu´s M` rquez ı a TALP Research Center Software Department Technical University of Catalonia lluism@lsi.upc.edu 1 Introduction 2 Content Overview and Outline This tutorial has two differentiated parts. In Semantic Role Labeling (SRL) consists of, given the ﬁrst one, the state-of-the-art on SRL will be a sentence, detecting basic event structures such overviewed, including: main techniques applied, as “who” did “what” to “whom”, “when” and existing systems, and lessons learned from the “where”. From a linguistic point of view, a key CoNLL and SemEval evaluation exercises. This component of the task corresponds to identifying part will include a critical review of current prob- the semantic arguments ﬁlling the roles of the sen- lems and the identiﬁcation of the main challenges tence predicates. Typical predicate semantic argu- for the future. The second part is devoted to the ments include Agent, Patient, and Instrument, but lines of research oriented to overcome current lim- semantic roles may also be found as adjuncts (e.g., itations. This part will include an analysis of Locative, Temporal, Manner, and Cause). The the relation between syntax and SRL, the devel- identiﬁcation of such event frames holds potential opment of joint systems for integrated syntactic- for signiﬁcant impact in many NLP applications, semantic analysis, generalization across corpora, such as Information Extraction, Question Answer- and engineering of truly semantic features. See ing, Summarization and Machine Translation. the outline below. Recently, the compilation and manual annota- 1. Introduction tion with semantic roles of several corpora has • Problem deﬁnition and properties enabled the development of supervised statistical • Importance of SRL approaches to SRL, which has become a well- • Main computational resources and systems avail- deﬁned task with a substantial body of work and able for SRL comparative evaluation. Signiﬁcant advances in 2. State-of-the-art SRL systems many directions have been reported over the last • Architecture several years, including but not limited to: ma- • Training of different components chine learning algorithms and architectures spe- • Feature engineering cialized for the task, feature engineering, inference 3. Empirical evaluation of SRL systems to force coherent solutions, and system combina- tions. • Evaluation exercises at SemEval and CoNLL conferences However, despite all the efforts and the con- • Main lessons learned siderable degree of maturity of the SRL technol- 4. Current problems and challenges ogy, the use of SRL systems in real-world ap- plications has so far been limited and, certainly, 5. Keys for future progress below the initial expectations. This fact has to • Relation to syntax: joint learning of syntactic and do with the weaknesses and limitations of current semantic dependencies systems, which have been highlighted by many • Generalization across domains and text genres • Use of semantic knowledge of the evaluation exercises and keep unresolved • SRL systems in applications for a few years (e.g., poor generalization across corpora, low scalability and efﬁciency, knowledge 6. Conclusions poor features, too high complexity, absolute per- formance below 90%, etc.). 3 Tutorial Abstracts of ACL-IJCNLP 2009, page 3, Suntec, Singapore, 2 August 2009. c 2009 ACL and AFNLP