330 likes | 482 Views
National Institute of Informatics Kiyoko Uchiyama. A Study for Introductory Terms in Logical Structure of Scientific Papers. 1. Outline. Purpose, background, motivation What’s “Introductory terms” Analysis of logical structure Analysis of structural role Apply to MIC theory Future works.
E N D
National Institute of Informatics Kiyoko Uchiyama A Study for Introductory Terms in Logical Structure of Scientific Papers 1
Outline Purpose, background, motivation What’s “Introductory terms” Analysis of logical structure Analysis of structural role Apply to MIC theory Future works 2
Result of author’s publications &similar papers and similar researchers
Purpose • Investigate the occurrence of introductory terms in logical structure of textbooks, research papers and encyclopedia • Categorize each sentence including introductory terms into structural roles • Analyze how to behave the introductory terms in Introduction section 9
Background • A lot of technical terms exit in specific domain • Difficult to identify the most important terms in the target field for novices • Novices should learn the basic and necessary terms in the field in the first priority 10
Our motivation • Apply to a method for advanced search • Assume that introductory terms .. • play a important role for describing domain knowledge • help novices to understand the content of academic papers 11
What is “introductory terms”? • Essential & basic terms for a target field • The terms that should make it a first priority to learn in a target field • Difficult to understand more difficult terms in the target field without the introductory terms 12
High →→ introductory degree →→ low Hidden Markov Model • Morphological analysis▼ • Syntactic analysis • Semantic analysis • Chasen • MeCab • JUMAN • KAKASI PaperA Conditional Random field Maximum entropy model Tutorial paper novice PaperB PaperC
Automatic definition • Define the introductory terms which are selected in common by a lot of experts • Experts of specific field wrote/edited the following resources • Textbooks • Encyclopedia • Research papers 14
Priority ( Frequency ) • Authors arrange the contents of their textbooks in an easy-to-understand order • Authors include important keywords in title, author-assigned keywords in academic papers • The table of contents of Encyclopedia is edited by a lot of experts 15
Logical Structure • Distribution patterns in IMRD structure (Introduction, Method, Result and Discussion) of the text might be informative for identifying the introductory terms • Assume that introductory terms are frequently used in introduction section 17
Data set • Target field: NLP, Target language: Japanese • Textbooks: 39 textbooks whose titles included “natural language processing” • Natural Language Processing Encyclopedia written in Japanese • Academic papers: 1421 papers of NLP research group in Information Processing Society of Japan from 1993 to 2007 18
Data collection • Morphological analysis by MeCab for Japanese • Extract sequential noun strings as the term candidate in • the textbooks(694 types) • table of contents of Encyclopedia(463 types) • title, abstract and author-assigned keywords in papers ( 13493 types) • 90 terms appeared in all of three resources 19
Analysis of Logical Structure • Use full text of research papers in NLP field • Target papers which describe experiments and results • Extract 100 papers which include words such as “experiment”, “evaluation”, “precision” and “%” and so on • Divide full texts into 6 sections 20
Analysis of structural role • Extract sentences including introductory terms in Introduction • “Introduction” section has several kinds of sentences outlining the research • Categorize each sentences into structural role by manual • Analyze the sentence from the viewpoint of various features 22
Structural role • Hypothesis • Motivation Problem • Background • Goal • Object • Method( new-old ) • Experiment • Model • Observation • Result • Conclusion Base on the the CoreSC Annotation scheme ( Soldatova & Liakata, 2007) 23
Features in structural role • Tense, aspect, modality • Verbs • Syntactic features • Lexical features 24
Tense, Aspect • Background • Recently, morphological analysis has been transitioning from the method based on heuristic knowledge to the method using probabilistic model. ( 近年、 〜しつつある。) • Related Works • The authors isproposing/proposed a method for morphological analysis using rule-based paraphrasing (提案している => 提案した) 25
Modality, Verbs • Modality • The high level of language processing would be needed for assigning semantic features to words • 必要かもしれない • Verbs • Specific verbs in present sense tend to be used in Object Ex. Propose、intend, design, tackle 26
Syntactic features • Temporal expression (Background) • Recently 近年、so far これまで、 • Several researches have been done …. • 研究が行われてきた • Fixed expression (Motivation, Related-works) • It is inevitable/necessary 〜必要である • The research has not be done … 〜の研究は行われていない • [Authors] is proposing … 提案している 28
Lexical features • Keywords related to structural role • Problem • One of the main problems is that unknown word and new terms have been increasing day by day. • it costs a lot of time … • Experiment • We conducted/ proceeded the experiment • In order to evaluate our proposed method, • Result • We show the result of the experiment … • We could obtain better precision … 29
Discussion • Introductory terms are frequently used in sentences to position the proposed method in a target field • Introductory terms and the structural role introduced the basic domain knowledge which is necessary for understanding the main purpose of papers • Possible to classify each sentence into specific structural role automatically 30
Future works • Categorize sentences including introductory terms into each structural role automatically • Analyze the collocation words with introductory terms • Syntactic information ( subject, object, modifier, and so on ) • Semantic relation between the introductory terms and other terms ( objective, method, target ) 31
Apply to MIC theory • Logical structure consists of structural roles • The authors consider the discourse of their paper based on their proposed model/method • MIC theory could be applied to sentence level and discourse level • The order strategy of structural roles might relate to meta-information 33
Analysis of Hierarchy • Sentence level • There are no researches for [METHOD] Basic expression → informative expression Discourse level • Background: Recently, [METHOD]has been used in… • Motivation: We need to consider [METHOD] for morphological analysis • Objective-New: We propose [METHOD] ←Focus 34
Conclusion • Might be interested in analysis of introductory terms and their surrounding syntactic and semantic information from the view point of MIC ( I’m not sure…) • The result of the analysis would hope to contribute the understanding of academic papers 35