| { |
| "paper_id": "C00-1022", |
| "header": { |
| "generated_with": "S2ORC 1.0.0", |
| "date_generated": "2023-01-19T13:30:50.241327Z" |
| }, |
| "title": "Exogeneous and Endogeneous Approaches to Semantic Categorization of Unknown Technical Terms", |
| "authors": [ |
| { |
| "first": "Farid", |
| "middle": [], |
| "last": "Cerbah", |
| "suffix": "", |
| "affiliation": { |
| "laboratory": "Dassault Aviation -DPR DESA -78", |
| "institution": "", |
| "location": { |
| "addrLine": "quai Marcel Dassault", |
| "postCode": "92552", |
| "settlement": "Saint-Cloud cedex 3 0 0 -F rance" |
| } |
| }, |
| "email": "farid.cerbah@dassault-aviation.fr" |
| } |
| ], |
| "year": "", |
| "venue": null, |
| "identifiers": {}, |
| "abstract": "Acquiring and updating terminological resources are di cult and tedious tasks, especially when semantic information should be provided. This paper deals with Term Semantic Categorization. The goal of this process is to assign semantic categories to unknown technical terms. We propose two approaches to the problem that rely on di erent knowledge sources. The exogeneous approach exploits contextual information extracted from corpora. The endogeneous approach relies on a lexical analysis of the technical terms. After describing the two implemented methods, we present the experiments that we conducted on signi cant test sets. The results demonstrate that term categorization can provide a reliable help in the terminology acquisition processes.", |
| "pdf_parse": { |
| "paper_id": "C00-1022", |
| "_pdf_hash": "", |
| "abstract": [ |
| { |
| "text": "Acquiring and updating terminological resources are di cult and tedious tasks, especially when semantic information should be provided. This paper deals with Term Semantic Categorization. The goal of this process is to assign semantic categories to unknown technical terms. We propose two approaches to the problem that rely on di erent knowledge sources. The exogeneous approach exploits contextual information extracted from corpora. The endogeneous approach relies on a lexical analysis of the technical terms. After describing the two implemented methods, we present the experiments that we conducted on signi cant test sets. The results demonstrate that term categorization can provide a reliable help in the terminology acquisition processes.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Abstract", |
| "sec_num": null |
| } |
| ], |
| "body_text": [ |
| { |
| "text": "Terminological resources have been found useful in many NLP applications, including Computer-Aided Translation and Information Retrieval. However, to have a signi cant impact on applications, terminological knowledge should not be limited to at nomenclatures of single-word and multi-word terms. They should include semantic knowledge, such as semantic categories and various kinds of semantic relations hyperonymy hyponymy, synonymy. This paper focuses on the assignment of semantic categories to technical terms. Semantic categories may play a crucial role in many applications, and particularly when disambiguation processes are required. For example, in our applicative framework, semantic categories ensure coarse sense division of polysemous terms and are actually used to improve the French-to-English translation of the technical documentation. Assigning semantic categories to techni-cal terms is a di cult and time-consuming task. Highly specialized skills are required and, even though the major concepts represented in these terminological resources pertain to the aeronautic domain, various related knowledge areas are concerned, such as Data Processing, Mechanics, and Manufacturing Processes. Our goal is to elaborate a tool that helps terminologists to assign semantic categories when updating the reference terminology. We think that signicant support can be provided to the terminologists by taking advantage of existing categorized terms and their usages in documents. Such a tool can be integrated within a terminology acquisition environment as a complement to a term extraction component. We distinguish two kinds of approaches to semantic categorization. In a way similar to corpus-based methods for Word Sense Disambiguation WSDYarowski, 1992; Ide and V eronis, 1998 , an exogeneous approach exploits contextual information extracted from corpora in order to determine the most plausible categories. By contrast, an endogeneous approach relies solely on a lexical analysis of multi-word terms which are very frequent in terminological databases. This approach is based on the assumption that lexical units used in the composition of technical terms are relevant indicators of semantic domains. The rest of the paper is organized as follows. Section 2 describes the terminological resources used in this study. Section 3 compares term categorization with related issues, such as thesaurus extension, WSD and term clustering. Sections 4 and 5 describe the two proposed methods. Results and evaluation are given and discussed in section 6. Directions for further research are pointed out in last section.", |
| "cite_spans": [ |
| { |
| "start": 1768, |
| "end": 1790, |
| "text": "Ide and V eronis, 1998", |
| "ref_id": "BIBREF4" |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Introduction", |
| "sec_num": "1" |
| }, |
| { |
| "text": "Categories POS anti-d erapage a long terme long-action antislip NAV N assiette de consigne au d ecollage required takeo attitude CKI N barre de remorquage tow bar TOO N d erive n STR N d erive wander PRO, FLP N d eriver to ", |
| "cite_spans": [], |
| "ref_spans": [ |
| { |
| "start": 15, |
| "end": 242, |
| "text": "anti-d erapage a long terme long-action antislip NAV N assiette de consigne au d ecollage required takeo attitude CKI N barre de remorquage tow bar TOO N d erive n STR N d erive wander PRO, FLP N d eriver to", |
| "ref_id": null |
| } |
| ], |
| "eq_spans": [], |
| "section": "English", |
| "sec_num": null |
| }, |
| { |
| "text": "We use in this study a French English bilingual terminology of the aeronautic domain. This hand-crafted database results from a multidisciplinary e ort involving technical writers, translators, terminologists and engineers. In its current state, the database contains 12,267 French English term couples, structured in 70 semantic categories. As already observed in several terminological databases, multi-word terms cover the larger part of the database nearly 80. The described terms are mostly nouns but the database also contains about 200 verbs and verbal phrases. Table 1 gives some examples of terms and a short description of the associated categories. These linguistic resources are integrated in a computer-aided translation environment used by technical writers. Semantic categories have originally been introduced in order to distinguish the various senses of polysemous terms. Each term couple is annotated with one or more categories specifying the contexts in which the translation is recommended. An entry is associated to a term for each identi ed meaning. For example, the french term d epassement has at least two possible meanings, corresponding to two di erent translations: over ow in the Data Processing category DPR and out-of-ushness in the Aircraft Structure category STR. As shown in the examples of table 1, the assignment of semantic categories has been extended to monosemous terms.", |
| "cite_spans": [], |
| "ref_spans": [ |
| { |
| "start": 569, |
| "end": 576, |
| "text": "Table 1", |
| "ref_id": "TABREF1" |
| } |
| ], |
| "eq_spans": [], |
| "section": "The Terminological Resources", |
| "sec_num": "2" |
| }, |
| { |
| "text": "In our experiments of term categorization, only the french terms have been used.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "The Terminological Resources", |
| "sec_num": "2" |
| }, |
| { |
| "text": "Term Semantic Categorization is on several aspects similar to Thesaurus Extension Uramoto, 1996; Tokunaga et al., 1997 . Our methods are close to those used for positioning unknown words in thesauri. However, the two issues can be di erentiated with respect to the manipulated data. A thesaurus is intended to cover a large set of conceptual domains while a terminological database is focused accurately on a speci c topic and its related domains. For example, in Tokunaga et al., 1997 , the thesaurus to be extended contains more than 500 categories. This tends to make the problem harder, but, since many categories are strictly independent, it is easier to nd distinctive features between categories. By contrast, our terminological database contains only 70 categories. But, in this restricted set, we nd categories corresponding to close or even overlapping knowledge areas. It is more di cult to di erentiate them.", |
| "cite_spans": [ |
| { |
| "start": 82, |
| "end": 96, |
| "text": "Uramoto, 1996;", |
| "ref_id": "BIBREF8" |
| }, |
| { |
| "start": 97, |
| "end": 118, |
| "text": "Tokunaga et al., 1997", |
| "ref_id": "BIBREF7" |
| }, |
| { |
| "start": 464, |
| "end": 485, |
| "text": "Tokunaga et al., 1997", |
| "ref_id": "BIBREF7" |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Related Work", |
| "sec_num": "3" |
| }, |
| { |
| "text": "Furthermore, the endogeneous approach, which exploits the multi-word nature of terminological units, cannot be applied to thesaurus extension because of the large amount of single-word thesaurus entries 1 . It is useful to compare exogeneous term categorization with corpus-based WSD methods.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Related Work", |
| "sec_num": "3" |
| }, |
| { |
| "text": "In both cases, contextual information extracted from corpora are used in order to assign the most plausible semantic tags to words. In WSD, the contextual cues that co-occur with the target word constitute the main training source whereas, in term categorization, the contextual information occurring with the term to be categorized should not be included in training data since the term is supposed to be unknown. The only relevant training sources are the contextual cues surrounding the already categorized terms. This is a basic di erence that explains why WSD tasks usually achieve better performance than term categorization and thesaurus extension.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Related Work", |
| "sec_num": "3" |
| }, |
| { |
| "text": "In a terminology acquisition framework, Habert et al. 1998 propose an exogeneous categorization method of unknown simple words. They use local context of simple words provided by a term extraction system. Endogeneous term categorization can also be compared with some approaches to term clustering Bourigault and Jacquemin, 1999; Assadi, 1997 . These approaches take advantage of the lexical and syntactic structures of technical terms in order to build semantic clusters.", |
| "cite_spans": [ |
| { |
| "start": 40, |
| "end": 58, |
| "text": "Habert et al. 1998", |
| "ref_id": "BIBREF3" |
| }, |
| { |
| "start": 298, |
| "end": 329, |
| "text": "Bourigault and Jacquemin, 1999;", |
| "ref_id": "BIBREF1" |
| }, |
| { |
| "start": 330, |
| "end": 342, |
| "text": "Assadi, 1997", |
| "ref_id": "BIBREF0" |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Related Work", |
| "sec_num": "3" |
| }, |
| { |
| "text": "We tested several classi cation models. Our rst experiments were carried out with Example-based classi ers. We used our own implementation of K-nearest neighbors algorithm kNN, and then the TiMBL learner Daelemans et al., 1999 , which provides several extensions to kNN, well-suited for NLP problems. Nevertheless, in the current state of our work, better results were obtained with a probabilistic classi er similar to the one used by Tokunaga et al., 1997 for thesaurus extension. Due to lack of space, only this method will be described in this paper.", |
| "cite_spans": [ |
| { |
| "start": 204, |
| "end": 226, |
| "text": "Daelemans et al., 1999", |
| "ref_id": "BIBREF2" |
| }, |
| { |
| "start": 436, |
| "end": 457, |
| "text": "Tokunaga et al., 1997", |
| "ref_id": "BIBREF7" |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Exogeneous Categorization", |
| "sec_num": "4" |
| }, |
| { |
| "text": "We use as contextual cues the open-class words nouns, verbs, adjectives, adverbs that cooccur in the corpus with the technical terms. More precisely, the cues are open-class words surrounding the occurrences of the term in some window of prede ned size. Each new term to be categorized is represented by the overall set of contextual cues that have been extracted from a part of the corpus test corpus.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Exogeneous Categorization", |
| "sec_num": "4" |
| }, |
| { |
| "text": "Let us consider a term T for which the contextual cues fw i g n i=1 have been collected in the test corpus. The categorization of this term amounts to nd the category C that maximizes probability PCjT:", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Probability Model", |
| "sec_num": "4.1" |
| }, |
| { |
| "text": "C = arg max C PCjT 1", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Probability Model", |
| "sec_num": "4.1" |
| }, |
| { |
| "text": "According to the exogeneous approach, the probability that a term T belongs to category C depends on the contextual cues of T:", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Probability Model", |
| "sec_num": "4.1" |
| }, |
| { |
| "text": "C = arg max C n X i=1 PCjw i P w i jT 2", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Probability Model", |
| "sec_num": "4.1" |
| }, |
| { |
| "text": "After applying Bayes'rule:", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Probability Model", |
| "sec_num": "4.1" |
| }, |
| { |
| "text": "C = arg max C PC n X i=1 Pw i jCPw i jT Pw i 3", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Probability Model", |
| "sec_num": "4.1" |
| }, |
| { |
| "text": "The probabilities of the equation 3 are estimated from training data:", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Probability Model", |
| "sec_num": "4.1" |
| }, |
| { |
| "text": "Pw i jC is the probability that a word w i co-occurs with a term belonging to category C. It is estimated in the following way:", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Probability Model", |
| "sec_num": "4.1" |
| }, |
| { |
| "text": "Pw i jC = N w w i ; C P w j N w w j ; C 4 N w w i ; C is the numberof times that w i co-occurs with a term belonging to category C. This probability accounts for the weight o f cue w i in category C.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Probability Model", |
| "sec_num": "4.1" |
| }, |
| { |
| "text": "Pw i jT is the probability that w i cooccurs with T:", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Probability Model", |
| "sec_num": "4.1" |
| }, |
| { |
| "text": "Pw i jT = N w w i ; T P w j Nw j ; T 5 N w w i ;", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Probability Model", |
| "sec_num": "4.1" |
| }, |
| { |
| "text": "T is the numberof times that w i co-occurs with T. Pw i is the probability of cue w i 2 :", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Probability Model", |
| "sec_num": "4.1" |
| }, |
| { |
| "text": "Pw i = N w w i P w j N w w j 6", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Probability Model", |
| "sec_num": "4.1" |
| }, |
| { |
| "text": "PC is the prior probability that a term of the corpus belongs to the category C:", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Probability Model", |
| "sec_num": "4.1" |
| }, |
| { |
| "text": "PC = N t C P C i N t C i 7", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Probability Model", |
| "sec_num": "4.1" |
| }, |
| { |
| "text": "where N t C is the occurrence numberin training data of terms belonging to C. This probability accounts for the weight of category C in the corpus. Figure 1 gives some examples of exogeneous categorization 3 .", |
| "cite_spans": [], |
| "ref_spans": [ |
| { |
| "start": 148, |
| "end": 156, |
| "text": "Figure 1", |
| "ref_id": null |
| } |
| ], |
| "eq_spans": [], |
| "section": "Probability Model", |
| "sec_num": "4.1" |
| }, |
| { |
| "text": "Our approach to endogeneous categorization is simpler. It is exclusively based on a quantitative analysis of the lexical composition of technical terms. Henceforth, the open-class words used to compose technical terms will be called terminological components. The endogeneous approach relies on a much more restricted source of data than the exogeneous approach, since the component set of a terminological database is quantitatively limited compared with the set of contextual cues extracted from corpora. Nevertheless, we make the assumption that this quantitative limitation is partly compensated by the ", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Endogeneous Categorization", |
| "sec_num": "5" |
| }, |
| { |
| "text": "To estimate the dependency between components and categories, we experimented several association criteria. The choice of these criteria has been in uenced by the comparative study described in Yang and Perdersen, 1997 on feature selection criteria for text categorization. We tested several measures, including component frequency, information gain and mutual information. Our best results were achieved with mutual information which is estimated using:", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Association Score", |
| "sec_num": "5.1" |
| }, |
| { |
| "text": "Iw;C log 2 N w w;C N w N w C N w w 8 N w w;C is the frequency of component w in category C. N w is the total numberof component occurrences. N w C is the total number of component occurrences in category C. This factor reduces the e ect of the components weakly represented in category C, compared with the other components of C. N w w is the frequency of component w in the terminological database. This factor reduces the e ect of the components that denote basic concepts spread all over the database. For example, the components speed, altitude, pressure have high frequencies in category FLP Flight Parameters, but, as basic concepts, they also appear frequently in many categories. Table 2 gives for two categories the ten most representative components according to this score.", |
| "cite_spans": [], |
| "ref_spans": [ |
| { |
| "start": 688, |
| "end": 695, |
| "text": "Table 2", |
| "ref_id": "TABREF4" |
| } |
| ], |
| "eq_spans": [], |
| "section": "Association Score", |
| "sec_num": "5.1" |
| }, |
| { |
| "text": "The association score between a term T with components fw i g n i=1 and a category C is given according to the components of T:", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Association Score", |
| "sec_num": "5.1" |
| }, |
| { |
| "text": "A t T ; C = N t C N t n X i=1 Iw i ; C 9", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Association Score", |
| "sec_num": "5.1" |
| }, |
| { |
| "text": "N t C is the number of terms pertaining to category C and N t is the total number of terms. The factor NtC Nt favors larger categories. The categorization task determines the category C that maximizes the association score: C = arg max C A t T ; C ", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Association Score", |
| "sec_num": "5.1" |
| }, |
| { |
| "text": "To estimate the accuracy of the exogeneous method, we used a domain-speci c corpus of 541,964 words, composed of documents pertaining to various textual genres software speci cations, maintenance procedures, manufacturing notices. . . . This corpus covered 63 of the 70 categories. Each run starts with the selection of a document among the corpus documents. The known terms identi ed in this document are considered as test terms. We used relatively wide contexts. The cues were extracted in a window of 20 words around the term.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Experiments and Evaluation", |
| "sec_num": "6" |
| }, |
| { |
| "text": "Each run involved more than 70,000 contexts of term occurrences. To experiment the endogeneous approach, test sets of compound terms have been randomly extracted from the terminological database. We adopted an evaluation scheme similar to that de ned in Tokunaga et al., 1997 for thesaurus extension. The categorization is considered successful if the right category appears among the k rst categories assigned by the classi er. Within a semi-automatic acquisition framework, this evaluation scheme is more suitable than strict evaluation where only the rst category assigned by the classi er is considered as relevant evaluation restricted to k=1 4 .", |
| "cite_spans": [ |
| { |
| "start": 254, |
| "end": 275, |
| "text": "Tokunaga et al., 1997", |
| "ref_id": "BIBREF7" |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Experiments and Evaluation", |
| "sec_num": "6" |
| }, |
| { |
| "text": "From our application perspective, it is useful to provide to the terminologist a restricted set of less than 5 plausible categories instead of the complete set of 70 categories without prior ltering.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Experiments and Evaluation", |
| "sec_num": "6" |
| }, |
| { |
| "text": "In the experiments described in Tokunaga et al., 1997 , k takes the values 5, 10, 20 and 30 for averaged performance ranging from 26.4 to 55.9 the choice is made among 544 categories. Some of their precise experiments yielded an accuracy greater than 80 for k = 30. In our experiments, we measured accuracy for k=1 to 5. Some results are given in tables 3 and 4. The scores are higher than those achieved in thesaurus extension, especially with the exogeneous approach from 52.75 to 97.61. We should however keep in mind that we deal with a di erent kind of data see section 3.", |
| "cite_spans": [ |
| { |
| "start": 32, |
| "end": 53, |
| "text": "Tokunaga et al., 1997", |
| "ref_id": "BIBREF7" |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Experiments and Evaluation", |
| "sec_num": "6" |
| }, |
| { |
| "text": "We h a ve presented in this paper two approaches to term semantic categorization that have been fully implemented and experimented on signicant test sets. The results achieved in this work demonstrate that term categorization tasks could be integrated within a semi-automatic terminology acquisition process to provide an active support to terminologists.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Conclusion and Further Work", |
| "sec_num": "7" |
| }, |
| { |
| "text": "The solutions to this problem can be considerably improved and we have identi ed several promising directions for further research. Our experiments show that exogeneous categorization is noticeably the most e cient of both approaches. However, it requires much more knowledge sources and computational overhead. It is more exposed to data sparseness, since large amounts of contextual data are not always available, especially in technical domains. We should stress that this study bene ted from the availability of a highly relevant corpus. This means that, for sake of robustness, other methods even less e cient and relevant knowledge sources should not be neglected. The two proposed approaches are complementary in the sense that they take advantage of distinct knowledge sources. Further work will investigate the various ways to combine them in order to improve the overall performance. The use of relational information, and particularly syntactic relations, is another major direction for further research. Exogeneous categorization is based on a bag of words lemmas model since wide contexts of lemmatized words were used, without consideration for the positions of these cues and their potential syntactic relationships with the target terms. Syntactic information extracted from local context, as verb-object relations, is another major source for exogeneous categorization that has been exploited in thesaurus extension methods. The endogeneous approach can also be improved by exploiting the syntactic structure of the technical terms. In our approach, all components of technical terms are equally weighted, independently of their syntactic roles within the terms. More accurate association scores can beintroduced by taking advantage of head modi er relations. Finally, w e should note that the bilingual nature of our terminological resources has not been taken into account. Minor changes are required to make the two classi ers work for English. Further experiments will be conducted on the English resources. In this bilingual context, either the French or the English expression or both could be used to categorize a given term.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Conclusion and Further Work", |
| "sec_num": "7" |
| }, |
| { |
| "text": "For Japanese,Tokunaga et al., 1997 reports some promising experiments of endogeneous categorization to thesaurus extension. The approach relies on properties of Japanese word formation rules and, thus, it can hardly be adapted for other languages. Their experiments suggest that exogeneous and endogeneous approaches are complementary.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "Note that the categorization process could be simplied by eliminating P wi, since this quantity is constant for all categories.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "Some category labels are described in table 1.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "5.2 Training and TestOnly multi-word terms can be categorized with this method since our endogeneous approach i s by nature not relevant for simple words. A test set of compound terms is extracted from the terminological database. The remaining terms are used for training. The training terms are analyzed in order to assign to each category its terminological components. Then, component frequencies and association scores are computed.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "Some limitations of strict evaluation are also pointed out inResnik and Yarowski, 1999.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| } |
| ], |
| "back_matter": [], |
| "bib_entries": { |
| "BIBREF0": { |
| "ref_id": "b0", |
| "title": "Knowledge Acquisition from Texts : Using an Automatic Clustering Method Based on Noun-Modi er Relationships", |
| "authors": [ |
| { |
| "first": "H", |
| "middle": [], |
| "last": "Assadi", |
| "suffix": "" |
| } |
| ], |
| "year": 1997, |
| "venue": "35th Annual Meeting of the Association for Computational Linguistics", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "H. Assadi. 1997. Knowledge Acquisition from Texts : Using an Automatic Clustering Method Based on Noun-Modi er Relationships. In 35th Annual Meeting of the Association for Compu- tational Linguistics, Madrid.", |
| "links": null |
| }, |
| "BIBREF1": { |
| "ref_id": "b1", |
| "title": "Term Extraction + Term Clustering: An Integrated Platform for Computer-Aided Terminology", |
| "authors": [ |
| { |
| "first": "D", |
| "middle": [], |
| "last": "Bourigault", |
| "suffix": "" |
| }, |
| { |
| "first": "C", |
| "middle": [], |
| "last": "Jacquemin", |
| "suffix": "" |
| } |
| ], |
| "year": 1999, |
| "venue": "Proceedings of the European Chapter of the Association for Computational Linguistics EACL '99", |
| "volume": "", |
| "issue": "", |
| "pages": "15--22", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "D. Bourigault and C. Jacquemin. 1999. Term Extraction + Term Clustering: An Integrated Platform for Computer-Aided Terminology. In Proceedings of the European Chapter of the As- sociation for Computational Linguistics EACL '99, pages 15 22, Bergen.", |
| "links": null |
| }, |
| "BIBREF2": { |
| "ref_id": "b2", |
| "title": "TiMBL: Tilburg Memory Based Learner Version 2.0 Reference Guide", |
| "authors": [ |
| { |
| "first": "W", |
| "middle": [], |
| "last": "Daelemans", |
| "suffix": "" |
| }, |
| { |
| "first": "J", |
| "middle": [], |
| "last": "Zavrel", |
| "suffix": "" |
| }, |
| { |
| "first": "K", |
| "middle": [], |
| "last": "Van Del Sloot", |
| "suffix": "" |
| }, |
| { |
| "first": "Antal", |
| "middle": [], |
| "last": "Van Den", |
| "suffix": "" |
| }, |
| { |
| "first": "", |
| "middle": [], |
| "last": "Bosch", |
| "suffix": "" |
| } |
| ], |
| "year": 1999, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "W. Daelemans, J. Zavrel, K. van del Sloot, and Antal van den Bosch, 1999. TiMBL: Tilburg Memory Based Learner Version 2.0 Refer- ence Guide. Tilburg University.", |
| "links": null |
| }, |
| "BIBREF3": { |
| "ref_id": "b3", |
| "title": "Extending an Existing Specialized Semantic Lexicon", |
| "authors": [ |
| { |
| "first": "B", |
| "middle": [], |
| "last": "Habert", |
| "suffix": "" |
| }, |
| { |
| "first": "A", |
| "middle": [], |
| "last": "Nazarenko", |
| "suffix": "" |
| }, |
| { |
| "first": "P", |
| "middle": [], |
| "last": "Zweigenbaum", |
| "suffix": "" |
| }, |
| { |
| "first": "J", |
| "middle": [], |
| "last": "Bouaud", |
| "suffix": "" |
| } |
| ], |
| "year": 1998, |
| "venue": "Proceedings of rst International Conference on Language Resources and Evaluation", |
| "volume": "", |
| "issue": "", |
| "pages": "663--668", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "B. Habert, A. Nazarenko, P. Zweigenbaum, and J. Bouaud. 1998. Extending an Exist- ing Specialized Semantic Lexicon. In Proceed- ings of rst International Conference on Lan- guage Resources and Evaluation, pages 663 668, Granada.", |
| "links": null |
| }, |
| "BIBREF4": { |
| "ref_id": "b4", |
| "title": "Introduction to the Special Issue on Word Sense Disambiguation: The State of the Art", |
| "authors": [ |
| { |
| "first": "N", |
| "middle": [], |
| "last": "Ide", |
| "suffix": "" |
| }, |
| { |
| "first": "J", |
| "middle": [], |
| "last": "", |
| "suffix": "" |
| } |
| ], |
| "year": 1998, |
| "venue": "Computational Linguistics", |
| "volume": "241", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "N. Ide and J. V eronis. 1998. Introduction to the Special Issue on Word Sense Disambigua- tion: The State of the Art. Computational Lin- guistics, 241:1 40.", |
| "links": null |
| }, |
| "BIBREF5": { |
| "ref_id": "b5", |
| "title": "MMORPH The Multext Morphology Program", |
| "authors": [ |
| { |
| "first": "D", |
| "middle": [], |
| "last": "Petitpierre", |
| "suffix": "" |
| }, |
| { |
| "first": "G", |
| "middle": [], |
| "last": "Russell", |
| "suffix": "" |
| } |
| ], |
| "year": 1995, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "D. Petitpierre and G. Russell. 1995. MMORPH The Multext Morphology Program. Technical report, Multext Deliverable 2.3.1.", |
| "links": null |
| }, |
| "BIBREF6": { |
| "ref_id": "b6", |
| "title": "Distinguishing systems and distinguishing senses: new evaluation methods for Word Sens Disambiguation", |
| "authors": [ |
| { |
| "first": "P", |
| "middle": [], |
| "last": "Resnik", |
| "suffix": "" |
| }, |
| { |
| "first": "D", |
| "middle": [], |
| "last": "Yarowski", |
| "suffix": "" |
| } |
| ], |
| "year": 1999, |
| "venue": "Natural Language Engineering", |
| "volume": "52", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "P. Resnik and D. Yarowski. 1999. Distinguish- ing systems and distinguishing senses: new eval- uation methods for Word Sens Disambiguation. Natural Language Engineering, 52:113 133.", |
| "links": null |
| }, |
| "BIBREF7": { |
| "ref_id": "b7", |
| "title": "Extending a thesaurus by classifying words", |
| "authors": [ |
| { |
| "first": "T", |
| "middle": [], |
| "last": "Tokunaga", |
| "suffix": "" |
| }, |
| { |
| "first": "A", |
| "middle": [], |
| "last": "Fujii", |
| "suffix": "" |
| }, |
| { |
| "first": "M", |
| "middle": [], |
| "last": "Iwayama", |
| "suffix": "" |
| } |
| ], |
| "year": 1997, |
| "venue": "Association for Computational Linguistics, editor, ACL Workshop on Automatic Information Extraction and Building of Lexical semantic Resources", |
| "volume": "", |
| "issue": "", |
| "pages": "16--21", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "T. Tokunaga, A. Fujii, and M. Iwayama. 1997. Extending a thesaurus by classifying words. In Association for Computational Linguistics, edi- tor, ACL Workshop on Automatic Information Extraction and Building of Lexical semantic Re- sources, pages 16 21.", |
| "links": null |
| }, |
| "BIBREF8": { |
| "ref_id": "b8", |
| "title": "Positioning unknown words in a thesaurus by using information extracted from a corpus", |
| "authors": [ |
| { |
| "first": "N", |
| "middle": [], |
| "last": "Uramoto", |
| "suffix": "" |
| } |
| ], |
| "year": 1996, |
| "venue": "Proceedings of Coling 96", |
| "volume": "956", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "N. Uramoto. 1996. Positioning unknown words in a thesaurus by using information extracted from a corpus. In Proceedings of Coling 96, pages 956 961.", |
| "links": null |
| }, |
| "BIBREF9": { |
| "ref_id": "b9", |
| "title": "A comparative study on feature selection in text categorization", |
| "authors": [ |
| { |
| "first": "Y", |
| "middle": [], |
| "last": "Yang", |
| "suffix": "" |
| }, |
| { |
| "first": "J", |
| "middle": [], |
| "last": "Perdersen", |
| "suffix": "" |
| } |
| ], |
| "year": 1997, |
| "venue": "Proceedings of the Fourteenth International Conference on Machine Learning ICML'97", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Y. Yang and J. Perdersen. 1997. A comparative study on feature selection in text categorization. In Proceedings of the Fourteenth International Conference on Machine Learning ICML'97,.", |
| "links": null |
| }, |
| "BIBREF10": { |
| "ref_id": "b10", |
| "title": "Word Sense Disambiguation Using Statistical Models of Roget's Categories Trained on Large Corpora", |
| "authors": [ |
| { |
| "first": "D", |
| "middle": [], |
| "last": "Yarowski", |
| "suffix": "" |
| } |
| ], |
| "year": 1992, |
| "venue": "Proceedings of Coling 92", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "D. Yarowski. 1992. Word Sense Disambigua- tion Using Statistical Models of Roget's Cate- gories Trained on Large Corpora. In Proceed- ings of Coling 92, Nantes.", |
| "links": null |
| } |
| }, |
| "ref_entries": { |
| "TABREF1": { |
| "type_str": "table", |
| "html": null, |
| "num": null, |
| "text": "A sample of the terminological database.", |
| "content": "<table/>" |
| }, |
| "TABREF2": { |
| "type_str": "table", |
| "html": null, |
| "num": null, |
| "text": "The exogeneous classi er starts with the selection of test documents in the corpus. Technical terms found in these documents will form the test set. The remaining documents represent the training corpus. Training and test stages are the following:POS tagging. The test and training corpora are tagged with MultAna, a tagger designed as an extension of the Multex morphological analyzerPetitpierre and Russell, 1995. Occurrences of the technical terms are identi ed during this stage and the terms to be categorized are those which are identi ed in the test corpus.", |
| "content": "<table><tr><td>Incr ementation Increment DPR 0.2106, CKI 0.2027, RDR 0.1967 D ecr ementation Decrement DPR 0.2111, CKI 0.1843, RDR 0.1654 Renseigner Inform CKI 0.24, VOR 0.21, DPR 0.1276 Entr ee d'air air intake ENG 0.1895, FUE 0.1214, DOC 0.1192 Moteur Engine ENG 0.1494, CDV 0.1285, DOC 0.1095 E acement de donn ees Data clearing DPR 0.2059, DOC 0.1357, RTL 0.129 T el ephone de piste Ground telephone RTL 0.1251, ELE 0.1131, EQX 0.1011</td><td>DPR DPR NAV ENG ENG RTL RTL</td></tr><tr><td colspan=\"2\">Figure 1: Some results of the exogeneous cate-gorization.</td></tr><tr><td colspan=\"2\">Frequency calculation and probabil-ity estimation. Training data are ex-plored to compute the frequencies occur-rences and co-occurrences of cues, terms and categories. As mentioned earlier sec-tion 3, the cue occurrences which have been collected around the test terms are ignored during this step. The probabili-ties required for the categorization opera-tion are then computed.</td></tr><tr><td colspan=\"2\">Categorization of the test terms. The most probable categories are assigned to each test term see section 4.1.</td></tr><tr><td>Extraction of contextual cues. For each term occurrence in training and test data, the contextual cues are collected. Only the lemmas of open-class words are used and cues may correspond to multi-word terms. Each test term is then repre-sented by the set of cues which have been collected in test data.</td><td/></tr></table>" |
| }, |
| "TABREF4": { |
| "type_str": "table", |
| "html": null, |
| "num": null, |
| "text": "The top ten most signi cant terminological components of categories ARR Arrangement and MAP Manufacturing Processes.", |
| "content": "<table><tr><td>strong discrimination power of terminological components. The training phase assigns to each category a set of representative components with respect to some association score. The categorization phase determines the most plausible categories of a term according to its components.</td></tr></table>" |
| }, |
| "TABREF5": { |
| "type_str": "table", |
| "html": null, |
| "num": null, |
| "text": "T k=1 k=2 k=3 k=4 k=5 312 45.07 68.22 79.15 88.10 91.40 89 42.25 73.14 86.95 91.57 98.19 98 50.65 84.94 91.11 94.93 96.96 120 48.11 78.95 91.26 96.83 97.", |
| "content": "<table><tr><td>92 253 59.17 81.98 92.35 96.75 98.65 125 73.16 89.30 96.85 98.68 99.59 234 68.19 87.63 94.14 96.99 98.31 205 49.05 83.86 95.12 98.61 99.47 203 47.25 72.18 85.66 94.40 98.48 105 44.55 69.61 87.28 93.97 97.11 Tot. 1744 Avg. 52.75 78.98 89.99 95.08 97.61</td></tr></table>" |
| }, |
| "TABREF6": { |
| "type_str": "table", |
| "html": null, |
| "num": null, |
| "text": "Results for the exogeneous approach.", |
| "content": "<table><tr><td>Ten experiments have been run for a total test set of 1744 terms.</td></tr><tr><td>T k=1 k=2 k=3 k=4 k=5 229 47.96 62.05 71.23 75.77 78.50 232 42.60 56.38 62.28 69.73 74.23 228 41.5 57.89 63.03 72.76 76.27 227 45.54 61.05 68.57 76.74 80.18 239 46.38 61.01 67.93 73.41 75.94 229 43.41 57.54 65.42 71.16 75.34 231 43.75 60.74 70.04 73.97 77.62 228 42.85 60.18 66.97 70.50 73.85 237 45.68 58.62 66.37 72.10 74.67 240 42.30 58.82 68.48 70.29 74.05 Tot. 2320 Avg. 44.19 57.77 67.03 72.64 76.06</td></tr></table>" |
| }, |
| "TABREF7": { |
| "type_str": "table", |
| "html": null, |
| "num": null, |
| "text": "Results for the endogeneous approach.", |
| "content": "<table><tr><td>Ten experiments have been run for a total test set of 2320 terms.</td></tr><tr><td>During the test phase, each test term is anno-tated with the most plausible categories accord-ing to its components.</td></tr></table>" |
| } |
| } |
| } |
| } |
