Improving Compatibility of Terminological Collections with a Bridging Classification of Data Categories

Improving compatibility of terminological collections with a bridging classification of data categories IGOR KUDASHEV University of Helsinki INTRODUCTION In spite of the existence of a standardized set of data categories (ISO 12620:1999) and standardized methods of terminology work (ISO 704:2000), it is still hard to find two terminological databases created in two different organizations which could be easily merged without loss of data or distortion of the original principles of compilation of each of the databases. The most obvious reasons for that are as follows: - national languages and traditions differ; - LSP domains differ; - backgrounds and approaches of the compilers differ; - technical solutions differ. At the same time, one of the trends in the management of terminological resources and other kinds of data has been towards the aggregation of scattered resources into bigger portals or services, usually without merging them physically. One of the examples is the EuroTermBank portal (http:// www.eurotermbank.com) which provides search in multiple internal and external terminological databases and can make a compilation of the results. As users’ demands and volumes of the data grow, it becomes necessary to provide advanced search which covers all fields of the entry and not just headwords and to tailor entries according to users’ preferences. In this article, we propose a classification of data categories which can serve as a bridge between terminological collections. The classification addresses the following problems related to the management of terminology collections with different sets of data categories: - aggregation and merging of terminological resources; - organization of search which covers all fields of the entries; - tailoring the entries from multiple collections in accordance with users’ preferences. 36 Igor Kudashev | Improving Compatibility of Terminological. Terminological database may contain different types of data in addition to terminological data proper, such as information about sources, users and terminology management transactions. In this article we focus on the classification of data categories related to the description of LSP expressions. WHAT IS A DATA CATEGORY? Ditferent types of data are placed in different fields in terminological databases. Data category is the result of specification of a given data field (ISO 1087-2:2000: 13). Terminologists use different metaphors in order to explain the concept of a data category. One metaphor is that of a wardrobe where different drawers are used to store different types of clothes. Another metaphor is a shopping basket where each kind of goods is wrapped in its own package so that they do not get mixed. TYPICAL MISMATCHES BETWEEN DATA CATEGORIES Data categories are the result of data classification. Data can be classified in many different ways depending on the views of the classifier and the needs of the users. Using the wardrobe metaphor one can say that different wardrobes come with different number of drawers which may be of different shape. size, etc. At least six types of mismatching are possible between data categories in terminological databases: - mismatch ot the names of data categories; - mismatch of the “sizes” of data categories, i.e. different “granulation”; - mismatch of the “places” of data categories, i.e. their location in the classification scheme; - mismatch of the classification principles (overlapping); - mismatch of the contents of data categories; - mixed cases. Below are a few examples which illustrate some common mismatches. Case 1: data categories are named differently. Example: data category is called note in one database, comment in another, remark in the third one and NB in the fourth one. Terminologija | 2009 | 16 37 Case 2: the name of a data category is used in different senses. . s“ - ss Example: data category synonym may correspond to “full synonym”, “near-synonym” or “full or near-synonym”. Case 3: names of data categories are “false friends” Example: English abbreviation and acronym as defined in the ISO standard (ISO 12620:1999: 6-7) and Russian ab66pesuamypa and akpornum correspond to each other crosswise (Kudashev 4 Hajutin 2003: 104-105). Case 4: granulation of data categories is different. Example: data category abbreviated form of term is split into five subclasses (abbreviation, short form of the term, initialism, acronym and clipped term) in ISO 12620:1999, but there is no such division in ISO 12616:2002 (“Translation-oriented terminographyTM). Case 5: overlapping of data categories. Example: data categories example and context overlap. Some examples are contexts and some contexts may serve as examples, but the two categories are not identical. Case 6: the same data category is put under different broader categories. Example: data category context is considered concept-related data in ISO 12620:1999, apparently because contexts may provide additional information on the concept. However, a more common function of contexts is to provide information about term usage and collocations, so “many databases classify context as a term-related category” (ISO 12620:1999: 25). Case 7: language, sign system or notation of the data differ. Example: a part of speech may be coded as noun in one database, n. in another, subst. in the third one and as a graphical symbol in the forth one. Case 8: identical or similar values of data categories are used in different meanings and different connections. Example: value neologism in the ISO 12620:1999 term provenance data category sounds like a chronological label while in fact it refers to the methodology employed in creating the term. 38 Igor Kudashev | Improving Compatibility of Terminological... Case 9: values of semi-closed classes differ due to different classification of linguistic phenomena in different languages and different traditions. Example: in ISO 12620:1999 labels for describing LSP expressions belonging to the “lower style” include slang register and vulgar register. They roughly correspond to labels npogdeccuonajronbiū xeapzonusm (professional slang) and npogeccuonarvnoe npocmopeuue (professional colloquialism) in Russian. In Finnish, however, there is only one category, ammattislangi (professional slang; see Sanastotyon kūsikirja 1989: 12). It should also be noted that some data may be expressed implicitly in terminological collections. Example I: the presence of the label obsolete in some cases and its absence in others in a terminological collection implies that terms which are not marked with the label, belong to the active LSP stock. Example 2: it two or more terms are put in the same entry, this usually implies that they are synonyms or equivalents. Data exchange or aggregation of terminological resources may require that such implicit data is made explicit. For example, if data is stored in the form of ontology rather than static entries, then implicit information about synonyms and equivalents mentioned above has to be made explicit and saved during the “dissembling” of the entries. ISO INVENTORIES AND CLASSIFICATIONS OF DATA CATEGORIES ISO 12620:1999 specities a set of data categories for recording terminological information. It does not prescribe what data categories should be used but rather serves as an inventory. This set of data categories is being broadened in the ISOcat project (www.isocat.org) which documents widely accepted linguistic data categories. While the ISOcat project is undoubtedly useful for linguists who get a chance to better categorize and define linguistic concepts and for designers of term banks who can pick ready data categories from a vast inventory, it is unlikely to bring relief to the problem of data exchange, aggregation and full-entry search in multiple databases. Firstly, it can't prevent compilers of terminological products from Terminologija | 2009 | 16 39 classifying data in a different way and from using their own data categories. Secondly, the more data categories are used in terminological collections, the more diverse they become, making data exchange even more challenging. The list of data categories is infinite in theory and quite vast in practice. For example, the list of data categories related to terminology in the ISOcat inventory already exceeds 500 items. One of the practices used in the ISOcat project can increase this number many times. We refer to presenting values of closed and semi-closed classes as individual data categories. For example, values of the data category register, such as vulgarRegister. slangRegister, etc. are now presented as data categories in their own right. Likewise, data category reliabilityCode is split into reliabilityCodel, reliabilityCode2, etc. There can be different opinions whether this is reasonable from the practical point of view, but data can be organized in many different ways, so there are no formal restrictions for splitting data categories all the way to the primitive classes which can only accept values yes and no. It is clear, however, that this realm of data categories needs proper structuring and classification in order to remain manageable and well-organized. Meanwhile, classification of data categories proposed in ISO 12620:1999 is problematic in several respects. To start with, there are some inconsistencies concerning the principal division of the data. According to section 6.2 (Typology of data categories). data categories are divided into three major groups: term and term-related information, descriptive data and administrative data. However, in Annex D (Systematic listing of data categories) the second group is called Data categories related to concept description. At the same time, this group contains subgroup Note which "stands alone because it can be associated with any one of the other categories and therefore cannot be subordinated to any other specific subgroup” (ISO 12620:1999: 4). If we assume that the intended division included four groups: term and term-related information, concept-related information, administrative data and Note, this classification still raises many questions. To name a few: - Why examples and contexts are concept-related data and not termrelated data? Cf. the description of the context data category: “A text or part of a text in which a term occurs” (ISO 12620:1999: 25). - Why synonym and equivalence are term-related data while otherwise everything related to the meaning is concept-related data? Cf. description of the data category degree of equivalence: “The extent to which the intensions of two or more concepts overlap” (ISO 12620:1990: 21). 40 Igor Kudashev | Improving Compatibiiity of Terminological... - Why antonym and homograph are administrative data and not termor concept-related data? - Why audio, video, etc. are (just) concept-related data? - What is the exact definition of administrative data? Why this class includes such heterogeneous categories? Division of terminological data into concept-related and term-related may be useful from the technical point of view because it supports the concept-oriented approach which reduces the number of relations between terms by linking synonymous terms to the same concept. However, such division in general and its implementation in ISO 12620:1999 in particular may present a challenge for common users of terminological databases — translators and domain experts. Indeed, it is not easy to comprehend why synonyms should be searched in term-related data, but antonyms in administrative data and examples in concept-related data. Common users work with terms — words and word combinations, so for them it is more natural to speak about the meaning of a term, synonyms of a term, examples of term usage, etc. This implies that classification of data categories aimed at providing full-entry search in multiple collections should be term-oriented, intuitively clear and based on common linguistic categories. BRIDGING MISMATCHES BETWEEN DATA CATEGORIES The problem of mismatches between data categories can be solved with the help of mapping. If mismatches are nominal, for example, if names of data categories differ or the same contents are presented differently, direct mapping between data categories can be used. Once the direct mapping is applied, the data in the aggregated resource can be searched with the same precision as in the original databases. More substantial differences between data categories require finding a common denominator through the mapping between classifications of data category sets. The use of a common denominator somewhat decreases the precision of the search, but this is the only way to provide common search for terminological resources with different structure. If the user is not ready to sacrifice the precision of the search, he has to perform separate search in each individual collection. As with data exchange in general, using an intermediate format, i.e. some bridging classification of data categories, is more effective in the long Terminologija | 2009 | 16 41 run than multiple mappings between different classifications. The general principles of the classification of data categories intended for serving as a common interface between other classifications are as follows: 1. Since classes of the classification are supposed to serve as common denominators, they have to be on a higher level of abstraction than most “primitive” data categories which are not subdivided further. At the same time the level of abstraction may not be too high because the users would not be interested in a classification which is too general. In practice a classification with two levels of abstraction is sufficient. 2. The classification should cover all types of LSP expressions which are typically described in terminological databases, including term elements, proper names, nomenclature, set phrases, etc. 3. The classification should contain only those data categories which are directly related to the description of LSP expressions. Description of technical and administrative aspects (e.g. medium, encodings, sources, reliability, etc.) is a different task. 4. The classification should be hierarchical, but several bases of division should be allowed on the same level of abstraction. 5. The classification should be extensible, i.e. it should contain category “other” on each level of the hierarchy. These principles have to be combined with the requirement of userfriendliness mentioned above. In our opinion classification based on linguistic functions of the data is a particularly strong candidate in the latter respect. In the bridging classification based on the linguistic functions the focus is shifted from the names of data categories to the function or functions of the data they contain. A comparison which can be made here is a library with its alphabetic and subject directories. Search by the names of data categories is similar to the search with the help of an alphabetical directory. The title gives some clue about the contents of the book, but sometimes it can be misleading. Besides, users cannot guess all the possible titles which cover a certain topic. Alphabetic directory is useful in cases when the users know for sure what item(s) they want to locate. Most library users, however, start their search with the subject directory because they do not know beforehand what items cover the topic they are interested in. Terminological portal which aggregates multiple terminological collec42 Igor Kudashev | Improving Compatibility of Terminological... tions with different sets of data categories poses the same problem to users as the realm of books on the shelves of a library. Users know what kind of information they are interested in (grammar, meaning, usage, etc.), but they do not necessarily know in what data categories this information can be found. In different terminological collections similar types of data may reside in different data fields. For example, information about the areal status of an LSP expression can be found in such fields as usage, regional label, language symbol, etc. If the search is based on the linguistic functions of the data, users don't need to care about the exact names of data categories. They just specify that they are looking for information about areal status, meaning or synonyms, and the terminology management system locates and displays full or abridged entries in different databases which contain the specified type of information. Search by the function may and should be complemented by search by the name of the fields for those users who know exactly in what fields they want to search. Data fields, like library items, may contain different kinds of linguistic information. For instance, example is a typical multifunctional field which may contain information about form, meaning and usage of a term, in different proportions. Each data category may be described as having one or several functions. It should also be possible to specify the degree to which a particular data category reflects certain functions. This characteristic may be verbal (e.g. primary and secondary function) or numerical (e.g. 0—-100%). The easiest way to do the mapping of a particular data category set to the intermediate classification is to consider the contents of data fields of a particular type uniform and make a simple table of correspondence. A more precise but also more complicated mapping would allow the compilers to specify deviations of individual data fields from the values used in the global table of correspondence. For example, contexts usually provide information about term usage and meaning. It is reasonable to include these types of information in the global table of correspondence as they pertain to every context field. However, some contexts may also contain encyclopedic information. This occasional use of encyclopedic information may be marked locally, on the field level. Terminologija | 2009 | 16 43 CLASSIFICATION OF DATA CATEGORIES BASED ON LINGUISTIC FUNCTIONS OF THE DATA Since LSP expressions are linguistic signs, information about them can be divided into information about their form, meaning, usage, relations to other units, origin and development. Below each of these classes is described in detail and some examples of the data belonging to them are provided. Please see Appendix 1 for the compact version of the classification. Data related to form LSP expressions have two forms — written and oral. Besides, data related to the form of an LSP expression may be subdivided into three classes: - data related to the canonical form; - data related to the formation of the unit; - data related to the inflection of the unit. DATA RELATED TO CANONICAL FORM Canonical form is the form in which the headword is given in the database. It serves as “representative” for other forms. For example, in most European languages the canonical form for nouns is nominative singular and for verbs infinitive presence. However, rules for choosing canonical form vary in different languages and lexicographical traditions. Here are a few examples of the data related to written canonical form: - type of expression by its written form (e.g. full form, abbreviated form, symbol, formula, etc.); - spelling of the form; - spelling variants of the form; - hyphenation, Examples of data related to oral canonical form: - type of expression by its oral form (e.g. initialism, acronym); - pronunciation; - pronunciation variants of the form; - syllabification. As one can see, some data categories may relate to both written and oral forms. For example, indicating that a particular expression is an initialism, i.e. an abbreviated form made of initial letters of the full term, in which these letters are pronounced individually (e.g. United Nations — UN), provides information about both written and oral forms. Besides, it may 44 [gor Kudashev | Improving Compatibility of Terminological... 4.1.8. Data related to register restrictions. <termUsage: restrictions: register > 4.1.9. Data related to professional group restrictions. <termUsage: restrictions: professional Group > 4.1.10. Data related to combinatory restrictions. <termUsage: restrictions: combinatory> 4.1.11. Data related to compliance restrictions. <termUsage: restrictions: compliance> 4.2. Data related to frequency of use. <termUsage: frequency> 5. Data related to systematic relations. <termRelations> 5.1. Data related to synonymous relations. <termRelations: synonymousRelation> 5.2. Data related to antonymous relations. <termRelations: antonymousRelation> 5.3. Data related to homonymous relations. <termRelations: homonymousRelation> 5.4. Data related to paronymous relations. <termRelations: paronymousRelation> 5.5. Data related to generic relations. <termRelations: genericRelation> 5.6. Data related to partitive relations. <termRelations: partitiveRelation> 5.7. Data related to non-hierarchical ontological relations. <termRelations: non-hierarchicalOntologicalR elation> 5.8. Data related to equivalence relations. <termRelations: equivalenceRelation> 6. Data related to origin and development. <termOriginAndDevelopment > 7. Other types of data related to the description of an LSP expression <termOtherRelated Data > APPENDIX 2. EXAMPLE OF MAPPING OF SOME ISO 12620:1999 DATA CATEGORIES INTO THE FUNCTIONAL CLASSIFICATION OF TERMINOLOGICAL DATA See Appendix I for the decryption of the formal representation of the categories which are used for the sake of brevity. Question mark after a data category means that the presence of the specified kind of data is subject to the interpretation of the data category. If subcategories can be mapped in the exact same way as their parent category, only the parent category is described. Terminologija | 2009 | 16 51 Due to space limitations only the first subgroup of data categories from the ISO 12620:1999 standard is covered. A.1 term Note: term (headword) lies outside the scope of classification which covers data categories related to the description of LSP expressions. However, technically data category term is identical to the written form of the term. Functional classification: not applicable or <termForm: writtenForm> A.2.1 term type Note:in the ISO 12620:1999 standard this category includes various kinds of LSP expressions and may contain different types of data. See subcategories for more information. A.2.1.1 main entry term Note: information about the structure of the entry is not data related to the description of LSP expressions. However, this category indirectly reflects preference. Functional classification: not applicable or <termUsage: restrictions: compliance> A.2.1.2 synonym Note: when opposed to main entry term, this category reflects preference. Its primary function is, however, reflection of synonymous relations. Functional classification: <termRelations: synonymousRelation>; <termUsage: restrictions: compliance>? A.2.1.3 quasi-synonym Note: same as previous. Functional classification: <termRelations: synonymousRelation>; <termUsage: restrictions: compliance>? A.2.1.4 international scientific term Note: depending on the interpretation of this category and its contents, this data category may provide information about compliance, language restrictions. origination and development. Functional classification: <termUsage: restrictions: compliance>; <termUsage: restrictions: language>?; <termQOriginAndDevelopment>? A.2.1.5 common name Note: common name is a synonym of an international scientific term that is used in general discourse. This data category may provide information about restrictions related to register, compliance and professional group. 52 Igor Kudashev | Improving Compatibility of Terminological... Functional classification: <termUsage: restrictions: register>; <termUsage: restrictions: compliance>; <termUsage: restrictions: professional Group>? A.2.1.6 internationalism Note: depending on the contents this data category may provide information related to language restrictions, term formation, origination and development. Functional classification: <termUsage: restrictions: language>; <termForm: termFormation>7; <termQriginAndDevelopment>? A.2.1.7 full form Note: this data category provides information about the type of term by its form. Functional classification: <termForm: term Type> A.2.1.8 abbreviated form of term Note: depending on the contents, this data category and all its subcategories (A.2.1.8.1 abbreviation. A.2.1.8.2 short form of term, A.2.1.8.3 initialism, A.2.1.8.4 acronym and A.2.1.8.5 clipped term) may provide information related to type of term by its form, oral form: term formation, origin and development. Functional classification: <termForm: termType>; <termForm: oralForm>7, <termForm: termFormation>?: <termOriginAndDevelopment> A.2.1.9 variant Note: depending on the contents, this data category may provide information about preference, term formation, origin and development. Functional classification: <termUsage: restrictions: compliance>; <termForm: termFormation>?; <termQriginAndDevelopment>? A.2.1.10-A.2.1.14 (transliterated form, transcribed form, romanized form, symbol and formula) Note: these data categories provide information about the type of term by its form. Functional classification: <termForm: termType> A.2.1.15-A.2.1.17 (equation, logical expression, materials management categories) Note: these units can not be headwords in terminological databases. Functional classification: not applicable. A.2.1.18 phraseological unit Note: in ISO 12620:1999 this category is split into three subcategories: collocation, set phrase and synonymous phrase. Collocation can not be counted Terminologija | 2009 | 16 53 as a phraseological unit as it does not correspond to the definition of a phraseological unit provided in the standard. Synonymous phrase seems to be an unnecessary category. The same information can be expressed with two other data categories: set phrase and synonym. Functional classification: not applicable. A.2.1.18.1 collocation Note: this data category provides information about combinatory restrictions. Functional classification: <termUsage: restrictions: combinatory> A.2.1.18.2 set phrase Note: this data category provides information about the type of term by correspondence of its meaning to its form. Functional classification: <termMeaning: termType> A.2.18.3 synonymous phrase Note: as was stated above, this category is probably unnecessary. If used, it provides information about the type of term by correspondence of its meaning to its form and also about synonymous relations. Functional classification: <termMeaning: termType>; <termRelations: synonymousRelation> A.2.1.19 standard text Note: standard texts can not be headwords in terminological databases. Functional classification: not applicable. TERMINIJOS RINKINIY SUDERINAMUMO GERINIMAS TAIKANT DUOMENŲ KATEGORIJŲ KLASIFIKACIJĄ Viena iš informacijos ir ypač terminologijos išteklių plėtros tendencijų yra atskirų duomenų bazių jungimas į didelius portalus ir sąsajų tarp tokių portalų kūrimas. To pavyzdys gali būti EuroTermBankas (http://www.curotermbank.com). Didėjant informacijos apimčiai ir vartotojų tokiems produktams keliamiems reikalavimams, auga poreikis užtikrinti išlpėstinę paiešką, apimančią visus žodynų straipsnių laukus, ir galimybę pateikti tik tas informacijos kategorijas, kurios tuo metu domina vartotoją. Šiems uždaviniams spręsti reikalinga duomenų kategorijų klasifikacija, kurią būtų galima taikyti kaip terminologinių bazių, turinčių skirtingą struktūrą, „bendrą vardiklį“ Svarbus tokios klasifikacijos reikalavimas — jos paprastumas ir aiškumas paprastiems terminologijos produktų vartotojams. Straipsnyje aptariama lingvistinėmis duomenų funkcijomis paremta klasitikacija galėtų tapti tokiu „bendru vardikliu® >4 Igor Kudashev | Improving Compatibility of Terminological... KCNONb30BAKHHE METAKNACCOB JAHHBIX KAK CPENCTBO YJIYVUIEHMA COBMECTHMOCTU TEPMUHOJIOTHUECKHX KOJINEKŲMA Ojnnoli H3 TEKJEKHHOKŪ pa3BHTHAS HHPOPMALUMOHHBIX PECYPCOB B HE/IOM H TEDMHHOHAOTHUUECKHX B YACTHOCTH CTO 00be/TMHCHNEe pa3po3HeHHbIX Das fAHHBIX B KPYIIHBIE mOpTA/IbI H CO3]faHHe obmux HHTEpĖOečCcoR K Hum. IlpuMepoM MOXET CAvxHTbh EspoTepmbank (http://www.eurotermbank.com). [To mepe pocra 065eMos nudopmarnuu H Tpe6osBaHHH M10Ab30BaATENeH K HOJ00HBIM MPOTYKTaM paCTeT TOTPeOHOCTH B OGucCIIeUEHHH PacIIMPEHHOrO HOHCKA, OXBATbIBAIOIIHETO BCE IIO/IH CIOBAPIBIX ClaTel, a Takxe BO3MOXKHOCTH OTOOPAXKEeHHS IHL TeX KaTETODHĖ HHOOpMAaLHH, KOTOPBIE B TAHHbIH MOMEHT UHTEPECYIOT MOMb30BaTeNs. ITH 3aaun TPpeObyIoT namrans Kiaccudukanmnn HHOOpMAHOHONHbIX KaTETODHH, KOTOPAs MOI1a OBI CIIYXKATH 4OŠIO2HM 3HAMEHATCIE M» JA TEPMHUHOMOTHYECKHX Bas ¢ pasandHO CTpyKTypon. HeManorax usm rpebonannem K ITo00HON K]aCCHĖOKKALHH SBAAETCs Ce IpOCTOTa H MOHATHOCTh IIS IIPOCTHIX IHOJIb30BAaTe/1EK TEDMKHOIOTHUECKHX NpoOjyKTOB. B cTaThe OTMCEIBaCTCSA O/IMH H3 BO3MOXHBIX KaHJUIATOB Ha PO/b «ODUIEro 3HAMEHATe Is» — KIACCHPUKALUsI, OCHOBAHHAS Ha JIMHTBUCTHYECKUX HYHKITHAX JAHHBIX. Gauta 2009-10-14 Igor Kudashev University of Helsinki Palmenia Centre for Continuing Education PO. Box 239 (Paraatikentti 6) FIN-45100 Kouvola, Finland E-mail: igor. [email protected] Terminologija | 2009 | 16 55