Validating the ontolex-lemon lexicography module with K dictionaries'' multilingual data

Valida ing he On oLex-lemon Lexicog aphy Module wi h K Dic iona ies’ Mul ilingual Da a Julia Bosque-Gil1,3, Do ielle Lonke2, Jo ge G acia3, Ilan Ke ne man2 1 On ology Enginee ing G oup, Uni e sidad Poli écnica de Mad id, Spain 2 K Dic iona ies, Tel A i , Is ael 3 A agon Ins i u e o Enginee ing Resea ch (I3A), Uni e si y o Za agoza, Spain E-mail: jbosque@ uniza .es, do ielle@kdic iona ies.com, jog acia@uniza .es, ilan@kdic iona ies.com Abs ac The On oLex-lemon model has g adually acqui ed he s a us o de- ac o s anda d o he ep esen a ion o lexical in o ma ion acco ding o he p inciples o Linked Da a (LD). Exposing he con en o lexicog aphic esou ces as LD b ings bo h bene i s o hei easie sha ing, disco e y, eusabili y and en ichmen a a Web scale, as well as o hei in e nal linking and be e euse o hei componen s. Howe e , wi h lemon being o iginally de ised o he lexicaliza ion o on ologies, a 1:1 mapping be ween i s elemen s and hose o a lexicog aphic esou ce is no always a ainable. In his pape we epo ou expe ience o alida ing he new lexicog module o On oLex-lemon, which aims a pa ing he way o b idge hose gaps. To ha end, we ha e applied he module o ep esen lexicog aphic da a coming om he Global mul ilingual se ies o K Dic iona ies (KD) as a eal use case scena io o his module. A en ion is d awn o he s uc u es and anno a ions ha lead o modelling challenges, he ways he lexicog module ackles hem, and whe e his modelling phase s ands as ega ds he con e sion p ocess and design decisions o KD’s Global se ies. Keywo ds: Linguis ic Linked Da a; RDF; mul ilingual; On oLex-lemon; K Dic iona ies 1. In oduc ion Linked da a (LD) echnologies a e inc easingly adop ed in lexicog aphy, whe he in academic esea ch and de elopmen , he indus y, o combining bo h (see o ins ance Klimek and B ümme (2015), Decle ck e al. (2015), Ab omei e al. (2016), Pa izi e al. (2016), Bosque-Gil e al. (2016a) and Kal enböck & Ke ne man (2017)). LD e e s o a se o bes p ac ices o exposing, sha ing and connec ing da a on he Web (Bize e al., 2009). The adop ion o LD in lexicog aphy enhances he endency o s anda dize he ways o ep esen a ion and que y o lexical con en a a Web scale. Connec ions can also be es ablished o o he LD esou ces, so ha lexicog aphic da a can be en iched wi h di e en ypes o complemen a y in o ma ion, such as addi ional ansla ions, de ini ions, examples o usage, e c. 726 P oceedings o eLex 2019 The de- ac o s anda d o ep esen ing on ology lexica, he lemon model (McC ae e al., 2012) and i s mo e ecen e sion, On oLex-lemon1 (McC ae e al., 2017), ha e been he p e e ed choice by de elope s o con e lexicog aphic esou ces in o LD. Ea ly expe iences in using lemon show ha he model is highly e ec i e as ega ds he accoun ing o he co e lexical in o ma ion in lexicog aphic esou ces (Klimek & B ümme , 2015; Decle ck e al., 2015; Ab omei e al., 2016; McC ae e al., 2019). Howe e , he e a e a ious si ua ions in which no pe ec ma ch is a ailable be ween he elemen s o he model and hose ound in lexicog aphic en ies, o in which he model alls sho o cap u ing ce ain peculia i ies o lexicog aphic wo ks, e.g. he o de o senses in an en y, de ails on he mo phological ea u es o wo d- o ms when used o a speci ic sense, e c. In his con ex , he W3C On oLex communi y g oup2 has analysed he main issues ega ding he ep esen a ion o lexicog aphic in o ma ion as LD and is eleasing his yea an upda ed module o ep esen lexicog aphic da a ha ex ends he lemon co e model – he lexicog module.3 In his pape we analyse he applica ion o he lexicog module o LD-based ep esen a ion o he Global se ies o K Dic iona ies (KD).4 The main con ibu ion o his wo k is wo old: 1. This pionee ing expe ience se es o alida e his new module wi h an ac ual use case as well as o in oduce some ecommenda ions o u u e applica ions. 2. By ocusing on KD’s da a, we examine how he limi a ions o he On oLex- lemon model al eady epo ed in he li e a u e (Klimek & B ümme , 2015; Bosque-Gil e al., 2016b) a e success ully add essed by he module. The es o his pape is s uc u ed as ollows: Sec ion 2 p o ides an o e iew o KD’s Global se ies and elabo a es on he mo i a ion o i s con e sion o LD, as well as a summa y o p e ious con e sions o hese da a o LD and he challenges encoun e ed in his p ocess. In Sec ion 3 he lexicog module is in oduced. Sec ion 4 b ie ly p esen s he di e en s ages o LD gene a ion, and whe e he modelling wi h lexicog s ands wi h espec o he whole con e sion o KD’s da a o he Resou ce Desc ip ion F amewo k (RDF), along wi h he echnologies and he design decisions we adop ed. Sec ion 5 add esses some o he limi a ions p e iously de ec ed in he li e a u e on he con e sion o KD’s da a and p o ides a modelling solu ion in e ms o lexicog. Concluding ema ks and u u e lines o wo k a e p esen ed in Sec ion 6. 1 h ps://www.w3.o g/2016/05/on olex/ 2 h ps://www.w3.o g/communi y/on olex/ 3 The lexicog module and epo a e a ailable a a h p://www.w3.o g/ns/lemon/lexicog# and h p://www.w3.o g/2019/09/lexicog/ espec i ely. 4 h p://www.lexicala.com/. 727 P oceedings o eLex 2019 2. K Dic iona ies’ Global se ies In his sec ion we b ie ly desc ibe he dic iona y da a ha we used o alida e he lexicog module, which s ems om he Global se ies o KD. This se ies is based on he monolingual lexicog aphic co es o 25 di e en languages and hei bilingual and mul ilingual e sions, and includes nea ly 100 language pai s and nume ous mul ilingual a ia ions. We discuss he mo i a ion o con e ing i in o LD and desc ibe p elimina y con e sions ha we e pe o med in he pas . 2.1 Con e ing KD’s da a o he RDF: mo i a ion and o e iew The Global se ies o KD (Ke ne man, 2009, 2011, 2015) has been concei ed as a c oss- lingual, mul i-laye mosaic o lexicog aphic esou ces ha e ol e wi hin a single sys emic amewo k, sha ing a common echnical mac os uc u e and a common en y mic os uc u e ha is able o accommoda e and adap o pa icula cha ac e is ics o di e en languages. All he language se s sha e he same XML schema (DTD), whe ein ce ain languages can ea u e addi ional o hog aphic sc ip s (e.g., ha e Kanji, Hi agana, Ka akana and Romaji o Japanese, o encompass di e se in lec ed e b o ms, o example, pe ec i e/impe ec i e o Polish and Russian). Each language esou ce is c ea ed on i s own, based on deep co pus analysis om which s em i s edi o ial s yle guide, headwo d lis , lexical deciphe ing and mapping, and di e se seman ic and syn ac ic a ibu es. The esul is a de ailed monolingual co e ha migh con ain o e lapping elemen s, such as de ini ions alongside sense disambigua ion elemen s, synonyms o an onyms, e c., which can hen be used selec i ely o cus omize ha da a o he needs o pa icula a ge audiences and usages. This co e is eady o be complemen ed by ansla ion equi alen s (o he senses, examples o usage, and mul iwo d uni s) o de eloping bilingual e sions, which a e jux aposed and o m a mul ilingual ne wo k e ol ing a ound he ini ial monolingual se . E en ually, he ansla ions (and o he componen s) o each language ne wo k can also be in e linked o each o he and exponen ially mul iply he c oss-lingual connec ions. Since i s incep ion in 2005, 25 language co es we e c ea ed, and al oge he nea ly 100 language pai s a e a ailable so a , besides nume ous mul ilingual combina ions. Ra he han aim o compile any speci ic dic iona y p oduc , he idea was o de elop mul i unc ional da a se s ha can be applied in di e en o ms and media, ei he independen ly o in conjunc ion wi h o he da a, whe he in ended o publish a p in dic iona y, de elop an online o a mobile dic iona y, o e lexical se ices, o be inco po a ed in NLP applica ions. The ad en in ecen yea s o Linguis ic LD and Seman ic Web echnologies has opened new ho izons o enhance his s a egic app oach o c ea ing well-s uc u ed, de ailed and ex ensi e lexicog aphic da a a he han single dic iona y p oduc s, by ein o cing and u he expanding exis ing da a, and imp o ing in e ope abili y be ween con en om he Global se ies and o he mul ilingual da a on he Web, a aining ecip ocal en ichmen o he Global se ies by ex e nal esou ces (on 728 P oceedings o eLex 2019 he one hand), and enhanced inco po a ion o da a om he Global esou ces in o ex e nal ones (on he o he hand). To pu his no ion in o p ac ice i became necessa y o i s ans o m KD’s Global da a om i s o iginal XML (hie a chical) o ma o an RDF s uc u e (knowledge g aph), o smoo he linking o ex e nal esou ces. Thus, KD decided o apply he bes -known LD s anda d model o ep esen ing lexicog aphic con en , i s in he o m o lemon, hen con o ming o On oLex-lemon, and mos ecen ly in line wi h i s up- o-da e lexicog module. The mo i a ion o KD o ocus and in es in his en u e can hus be explained by he in aluable upg ade his should ca y o i s esou ces, h ough acili a ing hei in e ope abili y and enhancing dep h, p ecision, and c oss-linguali y. Such imp o ed ea u es a e needed o deal wi h he eme ging mul ilingual single digi al ma ke , p ima ily in Eu ope and e en ually all o e he wo ld, which calls o mul iple adap a ions o con en and echnology, in e na ional s anda ds, mul i-disciplina i y, e c. LD me hods a e a he o e on o he cu en gene a ion o powe ul language echnology solu ions, a he hea o human-machine in e ac ion. P o iding quali y c oss-lingual lexical da a, wi h he LD-d i en op ion o linking o o he sou ces, g ea ly inc eases he appeal and uniqueness o he KD esou ces and places KD in a le e aged posi ion o o he compe ing dic iona y APIs. The new API o KD, enamed Lexicala API, p o ides access o he Global (and o he KD) da a in JSON, wi h he i s ouches o JSON-LD. I cons i u es a i al s ep in an inno a i e end o u ning passi e dic iona y p oduc s in o ac i e lexical da a se ices ha in e ope a e wi h eal-wo ld compu a ional linguis ics applica ions. Two ongoing H2020 p ojec s employ Lexicala API as pa o hei solu ions: Lynx5 will in eg a e KD (as well as e minological and o he ) esou ces wi h da a om he legal domain in he hea o i s Legal Knowledge G aph pla o m o mul ilingual compliance se ices; and Elexis6 will make use o he API o ecei e KD con en o i s u u e Eu opean lexicog aphic in as uc u e. Making KD esou ces a ailable in s a e-o - he- a RDF con o ming o wo ld-class s anda ds will bo h help o enhance he ope a ion o Lynx and Elexis pla o ms, and hose o a mul i ude o u u e applica ions, and o ein o ce and expand KD con en h ough in e ac ion wi h mo e LD esou ces. 2.2 P e ious ep esen a ions o KD’s da a as RDF The cu en con e sion o KD’s mul ilingual Global se ies is no he i s e o o con e his da a o RDF. In 2014 KD became in ol ed in he i s a emp o con e Global da a om XML o ma o RDF, adhe ing o he lemon model and ocusing on he Ge man monolingual da ase (Klimek & B ümme , 2015). The nex massi e s ep was aken in he wo-yea p ojec ca ied ou in 2015-2017 as pa o a EUREKA 5 h p://lynx-p ojec .eu/ 6 h ps://elex.is/ 729 P oceedings o eLex 2019 bila e al amewo k be ween KD and Seman ic Web Company (SWC), called Linked Da a Lexicog aphy o High-End Language Technology Applica ion (LDL4HELTA).7 As pa o he LDL4HELTA p ojec , he Global da a o h ee languages (English, Ge man and Spanish) was con e ed o RDF in line wi h he On oLex-lemon model (Bosque-Gil e al., 2016b). In he i s wo k (Klimek & B ümme , 2015), he au ho s iden i ied some gaps in he lemon model wi h ega d o ep esen ing KD’s da a, o ins ance, he way o link a compound ph ase de ined inside o a sense g oup o ha same sense. The lack o an on ology o p o ide on ological e e ences o lemon:LexicalSenses was also highligh ed. This poin is s ongly connec ed o he o iginal aim o he lemon model o se e o lexicalize on ologies, no o ep esen lexicog aphic esou ces in he Web o Da a. In addi ion, he au ho s iden i ied some gaps in he LexIn o8 ca alogue o g amma ca ego ies ( ypically used in conjunc ion wi h lemon) and c ea ed hei own cus om ocabula y o cap u e he alues o KD’s DTD a ibu es. In he la e con e sion o he se ies o On oLex-lemon (Bosque-Gil e al., 2016b), some p oblems ha we e iden i ied in he p e ious con e sion we e no longe ele an , as bo h he model and i s modules had e ol ed o co e mo e cases (e.g. now he a ans module allows o ep esen lexical ela ions). I is wo h no ing ha , whe eas in he i s wo a emp s he con e sion was ca ied ou unde he s ic p inciple o ound- ipping, i.e. aiming o ob ain ull and comple e 1:1 da a ans o ma ion om XML o RDF and om RDF back o XML – so he RDF s uc u e had o con ey each and e e y de ail o he complex ea u es o he o iginal XML s uc u e – he cu en wo k was eleased om his obliga ion. The easons o applying such a demand in he i s place we e, on he one hand, o se e as alida ion o pe ec ans e om XML o RDF while, on he o he hand, o be able o bene i om he po en ial en ichmen o he da a in RDF when linking o o he da a esou ces and impo ing such new da a back o he exis ing esou ce in XML. Remo ing his es ic ion has helped o libe a e and enhance he da a low om one o ma o ano he , and emphasized he au onomous s a us o each model and he ac ha e e y o ma should beha e eely and e lec i s au onomous cha ac e is ics ha a e di e en om he o he . Howe e , On oLex-lemon p o ed o be no exhaus i e enough o co e he ep esen a ion equi emen s o he o iginal esou ce. Fou kinds o challenging si ua ions we e de ec ed in he modelling o KD’s mul ilingual da a: 1. Cases in which solely applying he On oLex-lemon model would lead o a loss o s uc u al in o ma ion e lec ing lexical dis inc ions. Fo example, en ies no o iginally concei ed as dic iona y en ies in KD da a a e ea ed equally as 7 h ps://www.eu ekane wo k.o g/p ojec /id/9898 8 h ps://lexin o.ne /on ology/2.0/lexin o.owl 730 P oceedings o eLex 2019 o iginal en ies in he RDF ep esen a ion (i.e. as on olex:LexicalEn y). This highligh ed a lack o elemen s o ep esen ing he componen s o a lexicog aphic en y in cases in which he e is no 1:1 mapping wi h On oLex-lemon classes and p ope ies. In KD da a, we encoun e se e al examples o his ype o si ua ion: compounds, synonyms, an onyms, and ansla ions. Compounds a e de ined inside he dic iona y en y as one o i s componen s and do no occu as lemmas (i.e. in hei own dic iona y en y). Synonyms and an onyms, e en hough hey a e usually independen lemmas in ha same esou ce, a e embedded in dic iona y en ies and hey do no necessa ily ha e hei co esponding dic iona y en y in ha esou ce (bu could occu as dic iona y en ies in ano he KD dic iona y). In addi ion, a ansla ion o a headwo d in o ano he language is ea ed as an on olex:LexicalEn y (Bosque-Gil e al., 2016b), oo, bu jus as a synonym, and he sou ce da a in i s cu en s a e does no gua an ee o he wo d o be a lemma in he dic iona y o he a ge language. This ac called o a dis inc ion be ween an o iginal dic iona y en y and he on olex:LexicalEn y newly c ea ed in he p ocess, hus eco ding he ou come o he headwo d selec ion s ep in he compila ion o he dic iona y. In lexicog aphic esou ces o he han KD, he same gap would su ace in hose cases in which a dic iona y en y needs o be spli in o mo e han one on olex:LexicalEn y, each wi h a di e en pa o speech, in o de o be On oLex-complian . 2. Cases in which On oLex-lemon o LexIn o alls sho o co e ing he ep esen a ion needs ha KD’s dic iona y en ies gi e ise o. This conce ned he ep esen a ion o examples and ansla ions o examples, which a e ai ly common elemen s in o he dic iona ies as well (Bosque-Gil e al., 2017). 3. Cases in which On oLex-lemon does con ain elemen s o co e a pa icula ype o in o ma ion, bu he e a e no speci ic guidelines on how o use hem in he p ocess o con e sion o lexicog aphic da a o RDF (wi hou in ol ing on ology lexicaliza ion). Fo example, he ep esen a ion o lexicog aphic de ini ions wi h he On oLex co e (e.g. wi h skos:Concep o on olex:LexicalConcep ), he encoding o geog aphical usage es ic ions on senses, o he modelling o selec ional es ic ions o p edica e a gumen s. 4. Misma ches be ween LexIn o elemen s and KD’s DTD ags and alues. Since si ua ions o ypes (1) and (2) we e also gene alizable o o he lexicog aphic esou ces, lexicog was p oposed as an ex ension o On oLex-lemon (Bosque-Gil e al., 2017). Fo cases o ype (3), he On oLex Communi y, in i s bi-weekly elcos on lexicog aphy, discussed a se ies o p ac ices o he use o On oLex-lemon elemen s in he con e sion o lexicog aphic da a o RDF.9 These p ac ices eme ged as solu ions o 9 h ps://www.w3.o g/communi y/on olex/wiki/Lexicog aphy 731 P oceedings o eLex 2019 a lis o issues de ec ed in he li e a u e. A se ies o guidelines, wi h mo e examples and ecommenda ions, a e also planned as u u e s eps in he On oLex communi y. Cases o ype (4) we e add essed in 2016 by c ea ing a cus om on ology o KD, which is cu en ly unde e ision and upda e. 3. The lemon lexicog aphy module: lexicog The lemon model has been ex ensi ely used o ep esen ing lexicog aphic da a. Howe e , some limi a ions we e de ec ed in se e al p elimina y expe iences, as epo ed in Sec ion 1. Such issues we e collec ed and analysed by he W3C On oLex communi y g oup wi h he aim o eaching some ag eemen ha allows o a be e and mo e in e ope able mig a ion o exis ing dic iona ies in o LD. As a esul o his communi y e o , he On olex-lemon lexicog aphy module (lexicog) was de ined as an ex ension o he On oLex-lemon model.10 The module is a ge ed a he ep esen a ion o dic iona ies and any o he linguis ic esou ce con aining lexicog aphic da a, and add esses s uc u es and anno a ions commonly ound in lexicog aphy. The lexicog module o e comes some limi a ions o On oLex-lemon when modelling lexicog aphic in o ma ion as LD. I aims a cap u ing he unde lying o iginal s uc u e and anno a ions o he lexicog aphic en y in a way ha keeps he pu ely lexical con en sepa a e om he lexicog aphic one, minimizing in o ma ion loss and allowing que ies es ic ed o he lexical laye . By being able o keep eco d o he o iginal dic iona y a angemen as RDF, he module does no impose a ce ain iew on he lexicon and hus becomes agnos ic o he s andpoin o he lexicog aphe . Fo ha pu pose, new on ology elemen s ha e been added ha e lec he dic iona y s uc u e (e.g., sense o de ing, en y hie a chies, e c.) and complemen he On oLex-lemon lexicon. Figu e 1 depic s he main classes and ela ions de ined in he lexicog module. We e e o he speci ica ion documen o mo e de ails, bu we gi e he e an o e iew o i s main modelling ing edien s:  Lexicog aphicResou ce, which ep esen s a collec ion o lexicog aphic en ies in acco d wi h he lexicog aphic c i e ia ollowed in he de elopmen o ha esou ce.  En y, a s uc u al elemen ha ep esen s a lexicog aphic a icle o eco d as i is a anged in a sou ce lexicog aphic esou ce.  Lexicog aphicComponen , which is a s uc u al elemen aimed a ep esen ing he (sub)s uc u es o lexicog aphic a icles p o iding in o ma ion abou en ies, 10 A eco d o he discussed issues and in e media e design decisions can be ound a h ps://www.w3.o g/communi y/on olex/wiki/Lexicog aphy. 732 P oceedings o eLex 2019 senses o suben ies. Lexicog aphic componen s can be a anged in a speci ic o de and/o hie a chy. Figu e 1: Scheme o he lexicog aphy module ( aken om he “On oLex-lemon Lexicog aphy Module” W3C communi y g oup inal epo ). The h ee abo e elemen s accoun o he basic s uc u e o he Lexicog aphicResou ce. To ha end, a p ope y en y ela es a Lexicog aphicResou ce o each o i s en ies. An En y in u n can g oup se e al Lexicog aphicComponen s. We can indica e ha he componen s belong o an en y by simply using he RDF na i e mechanisms o con aine s.11 In pa icula , he d s:membe p ope y can be used i he o de o he componen s is no ele an , and d s:Con aine Membe shipP ope y ( d :_1, d :_2, ...) when he o de o he componen s needs o be ep esen ed. No ice ha an En y is a pa icula subclass o Lexicog aphicComponen used o ep esen he main “en y poin ” in he dic iona y, i.e., he headwo d o he oo o he lexicog aphic eco d. The lexicog aphic componen s only e lec he s uc u e o he dic iona y and do no encode any lexical con en hemsel es. To associa e hem o hei co esponding lexical in o ma ion (e.g. lexical en ies o lexical senses), he p ope y “desc ibes” is used. Such 11 h ps://www.w3.o g/TR/ d -schema/#ch_con aine ocab 733 P oceedings o eLex 2019 elemen s belonging o a lexicon a e aken om On oLex, in pa icula :  on olex:LexicalEn y, which consis s o a se o o ms ha a e g amma ically ela ed and a se o base meanings ha a e associa ed wi h all o hese o ms.  on olex:LexicalSense, which ep esen s he lexical meaning o a lexical en y when in e p e ed as e e ing o he co esponding on ology elemen .  lime:Lexicon, o a collec ion o lexical en ies o a pa icula language o domain. These classes can be u he connec ed wi h many o he elemen s ha desc ibe he lexicon and ha can be ound in he On oLex-lemon speci ica ion. Pa icula ly, he on olex:Fo m class, o accoun o he g amma ical ealiza ion o a lexical en y ( ypically by means o i s w i en ep esen a ion) and he on olex:LexicalConcep class, ha can be used o s o e de ini ions h ough he p ope y skos:de ini ion. Finally, we men ion he UsageExample class o he lexicog module, which is in ended o ep esen a ex ual example o he usage o a sense in a gi en lexicog aphic eco d. 4. Me hodology 4.1 Inc emen al app oach and s eps aken The p ocess o con e ing KD da a in o LD was ca ied ou wi h an inc emen al app oach, s a ing wi h he e y basics o a single en y (headwo d, senses, pa o speech, de ini ions) and p oceeding wi h mo e complica ed elemen s (synonyms, ansla ions, examples o use, compounds, e c.), alida ing he esul s o he con e sion a e each i e a ion. This app oach allowed o cons an alida ion and e o elimina ion, and acili a ed he echnical con e sion p ocess. P io o con e ing ac ual da a, some g oundwo k was necessa y. Fo his pu pose, he DTD o KD’s XML da a was examined, and each XML pa h in KD da a was manually de ined as a co esponding On oLex, lexicog o LexIn o elemen . Nex , a URI naming s a egy was es ablished, ollowing he p e ious con e sion o he Global se ies (Bosque-Gil e al., 2016b). In addi ion, he DTD was e ised and edi ed whe e possible, adhe ing o he s anda ds se by LexIn o and On oLex and p io i izing smoo h con e sion and adap able esul s. A e se ing he ounda ions o con e sion, he ollowing s eps we e aken o each i e a ion:  Iden i ying a ew en i ies in lexicog o es , and manually c ea ing an example RDF en y wi h eal KD da a. Only a hand ul o componen s comp ising a comple e dic iona y en y we e selec ed o each i e a ion, o simpli y each s ep and go e n he esul s mo e easily. In o de o main ain ha he con e sion was ca ied ou exhaus i ely and accu a ely, logs we e kep , and he URI naming 734 P oceedings o eLex 2019 5.3 Usage Examples The da a om KD’s Global se ies p o ides, o each sense o a headwo d, a usage example in he sou ce language and he ansla ions o he headwo d in he a ge language. The examples, in u n, a e also ansla ed o he a ge language and se e as example o usage o he ansla ion. Example 1.5 shows ano he exce p o he en y a e in Spanish. Inside he SenseG p encapsula ing he in o ma ion o he i s sense, he e is an elemen T ansla ionClus e wi h Locale g oups ha include he headwo d ansla ions o a e in i s i s sense: kuns (Du ch) and kuns (No wegian). Below he ansla ions ollows an ExampleC n wi h he example o usage o a e in ha sense, La música, la danza y la pin u a son o mas de a e ‘Music, dance and pain ing a e a o ms’. This example is in u n ansla ed o Du ch and No wegian. <SenseG p iden i ie ="SE00007455" e sion="1"> [...] <T ansla ionClus e iden i ie ="TC00017354" ex ="mani es ación humana con in ención es é ica" ype="de "> <Locale lang="nl"> <T ansla ionBlock> <T ansla ionC n> <T ansla ion>kuns </T ansla ion> [...] </T ansla ionC n> </T ansla ionBlock> </Locale> <Locale lang="no"> <T ansla ionBlock> <T ansla ionC n> <T ansla ion>kuns </T ansla ion> [...] </T ansla ionC n> </T ansla ionBlock> </Locale> [...] </T ansla ionClus e > <ExampleC n ype="sid" e sion="1"> <Example>La música, la danza y la pin u a son o mas de a e.</Example> <T ansla ionClus e iden i ie ="TC00017355" [...]> <Locale lang="nl"> <T ansla ionBlock> <T ansla ionC n> <T ansla ion>Muziek, dans en schilde en zijn o men an kuns .</T ansla ion> </T ansla ionC n> </T ansla ionBlock> 741 P oceedings o eLex 2019 </Locale> <Locale lang="no"> <T ansla ionBlock> <T ansla ionC n> <T ansla ion>Musikk, dans og maling e kuns ype .</T ansla ion> </T ansla ionC n> </T ansla ionBlock> </Locale> [...] </T ansla ionClus e > </ExampleC n> </SenseG p> Example 1.5: An ex ac o he Spanish en y a e wi h ansla ions in o Du ch and No wegian examples and ansla ions o he examples. While he lemon model p o ided a class lemon:UsageExample and a p ope y lemon:example, used p e iously in he li e a u e o cap u e his in o ma ion (Klimek & B ümme , 2015), hese a e no longe included in he On oLex-lemon model. P e ious con e sions o KD’s da a (Bosque-Gil e al., 2016b) p oposed a cus om class in o de no o ins an ia e bo h lemon and On oLex-lemon in he same esou ce. I an example is o be linked o a sense, he p ope y skos:example would su ice o include he example as a s ing a he sense le el. Fo cases in which he example has addi ional in o ma ion, o has elemen s linkable o i , he lexicog modules o e s he class lexicog:UsageExample o link an on olex:LexicalSense o an elemen ep esen ing he example. A lexicog:UsageExaple and be u he linked o o he elemen s and desc ibed wi h da a- ype p ope ies. Example 1.6 shows he RDF Tu le ep esen a ion o example 1.5. As showed in example 1.2, he headwo d and he ansla ions belong o di e en lexica, one pe language. (Con inua ion) :lexiconES/a e-n a on olex:LexicalEn y ; on olex:sense :lexiconES/a e-n-SE00007455- sense . :lexiconNL/kuns -n a on olex:LexicalEn y; on olex:sense :lexiconNL/kuns -n-a e-n- SE00007455-sense . :lexiconNO/kuns -n a on olex:LexicalSense; on olex:sense :lexiconNO/kuns -n-a e-n- SE00007455-sense . :lexiconNO a lime:Lexicon; lime:language "no"; 742 P oceedings o eLex 2019 lime:en y :lexiconNO/ kuns -n . :lexiconES/a e-n-SE00007455-sense a on olex:LexicalSense ; lexicog:usageExample :lexiconES/a e-n-SE00007455-sense- TC00017355-ex . :lexiconNL/kuns -n-a e-n-SE00007455-sense a on olex:LexicalSense ; lexicog:usageExample :lexiconES/a e-n-SE00007455-sense- TC00017355-ex . :lexiconNO/kuns -n-a e-n-SE00007455-sense a on olex:LexicalSense ; lexicog:usageExample :lexiconES/a e-n-SE00007455-sense- TC00017355-ex . : anSe ES-NL/a e-n-SE00007455-sense-kuns -n-a e-n-SE00007455-sense-TC00017354- ans a a ans:T ansla ion ; a ans:sou ce :lexiconES/a e-n-SE00007455-sense; a ans: a ge :lexiconNL/kuns -n-a e-n-SE00007455- sense; dc:sou ce :mlds-ES3 . : anSe ES-NO/a e-n-SE00007455-sense-kuns -n-a e-n-SE00007455-sense-TC00017354- ans a a ans:T ansla ion ; a ans:sou ce :lexiconES/a e-n-SE00007455-sense; a ans: a ge :lexiconNO/kuns -n-a e-n-SE00007455- sense dc:sou ce :mlds-ES3 . :lexiconES/a e-n-SE00007455-sense-TC00017355-ex a lexicog:UsageExample ; d : alue "La música, la danza y la pin u a son o mas de a e."@es ; d : alue "Muziek, dans en schilde en zijn o men an kuns ."@nl ; d : alue "Musikk, dans og maling e kuns ype ."@no . Example 1.6: RDF Tu le ep esen a ion o an ex ac o he Spanish en y a e wi h ansla ions in o Du ch and No wegian, examples, and ansla ions o he examples. Each on olex:LexicalEn y has an on olex:LexicalSense, which is he b idge be ween he linguis ic desc ip ion and he seman ic laye , ollowing he no ion o seman ics by e e ence emb aced in lemon.12 The example is eco ded h ough an ins ance o lexicog:UsageExample linked o he senses ia lexicog:usageExample. No e ha his ins ance has di e en alues, each o he ealiza ion o ha example in a di e en language. 12 Due o he lack o on ology en i ies o ac as e e ences o on olex:LexicalSenses, he seman ics p o ided by de ini ions will be cap u ed h ough on olex:LexicalConcep s and he p ope y skos:de ini ion. Howe e , he ins an ia ion o he On oLex co e, beyond lexicog, is ou o he scope o his pape , and we e e he eade o he examples p o ided a he lexicog documen a ion page. 743 P oceedings o eLex 2019 6. Conclusions and u u e wo k In his pape we ha e p esen ed wo k on applying he new lexicog module o On oLex- lemon o KD’s mul ilingual da a as a eal use case scena io o he ex ension. We ha e shown ha lexicog add esses he gaps p e iously iden i ied in he li e a u e (Klimek & B ümme , 2015; Bosque-Gil e al., 2016b, 2017) as ega ds he loss o s uc u al and implici lexical in o ma ion in he o iginal esou ce, and p o ides elemen s o cap u e da a equen ly ound in lexicog aphic eco ds, such as usage examples, ansla ions, o anno a ions on mo phosyn a ic ea u es. In addi ion, and o se e as a basis o u u e ans o ma ions o lexicog aphic da a, we amed he modelling wi h lexicog in he whole con e sion p ocess o KD o LD. We ha e de ailed he inc emen al app oach ollowed in he con e sion p ocess and ou lined he di e en s eps pe o med as pa o he alida ion p ocess o he esul ing RDF. The nex s ep will be o p ocess he da a in a iple s o e, se ing bo h o u he alida e he lawless con e sion om XML o RDF and o p epa e he da a o linking o o he ex e nal LD esou ces. Then, he ac ual linking o such ex e nal da a esou ces can ake place. Fu u e wo k includes linking be ween di e en KD monolingual co es, c ea ing one in e connec ed, ully c oss-lingual g aph, as well as linking o ex e nal sou ces, hus enhancing he da a and p o iding e en mo e elabo a e and en iched da a o Lexicala API use s and o a ious esea ch and de elopmen pu poses. The ask o linking dic iona ies, by associa ing a ansla ion o a headwo d in he sou ce language dic iona y co e o i s co esponding en y in he a ge language dic iona y co e, is an ambi ious and elabo a e one. The main hind ance is au oma ing he p ocess, managing o link a ansla ion equi alen o he co ec senses ac oss languages, which is ul ima ely ela ed o wo d sense disambigua ion, and has been p e iously a emp ed wi h KD da a as pa o he LDL4HELTA p ojec and he T ansla ion In e ence Ac oss Dic iona ies sha ed asks and wo kshops (TIAD).13 The con e sion o KD monolingual co es o LD has laid he g oundwo k o his ype o g aph, and p o ided u he ideas o ca ying ou his goal in he u u e. In he mean ime, linking KD da a o o he sou ces should be signi ican ly acili a ed by he cu en con e sion. Linking KD da a wi h ex e nal, anno a ed o en iched esou ces, will g ea ly enhance bo h i s comme cial appeal and po en ial o u he esea ch, and can se e as a de ailed and e icien esou ce o language p ocessing and pa sing asks in he ealm o compu a ional linguis ics, hus expanding he ou each o LD-complian lexicog aphic da a ye u he . 13 h p://2019.ldk-con .o g/ iad-2019/ 744 P oceedings o eLex 2019 7. Acknowledgemen s This wo k has been suppo ed by he Spanish Minis y o Educa ion, Cul u e and Spo s h ough he FPU p og am, and by he Eu opean Union’s Ho izon 2020 esea ch and inno a ion p og amme h ough he p ojec s Lynx (g an ag eemen No 780602), Elexis (g an ag eemen No 731015) and P ê -à-LLOD (g an ag eemen No 825182). I has been also pa ially suppo ed by he Spanish Na ional p ojec s TIN2016-78011- C4-3-R (AEI/ FEDER, UE) and DGA/FEDER. 8. Re e ences Ab omei , F., Chia cos, C., Fä h, C. & Iono , M. (2016). Linking he Towe o Babel: Modelling a Massi e Se o E ymological Dic iona ies as RDF. In P oceedings o he 5 h Wo kshop on Linked Da a in Linguis ics: Managing, Building and Using Linked Language Resou ces (LDL-2016). pp. 11–19. Bize , C., Hea h, T. & Be ne s-Lee, T. (2009). Linked Da a - The S o y So Fa . In e na ional Jou nal on Seman ic Web and In o ma ion Sys ems (IJSWIS), 5(3), pp. 1–22. Bosque-Gil, J., G acia, J. & Gómez-Pé ez, A. (2016a). Linked da a in lexicog aphy. Ke ne man Dic iona y News, (26), pp. 19–24. h ps://www.kdic iona ies.com/kdn/kdn24_2016.pd . Bosque-Gil, J., G acia, J. & Mon iel-Ponsoda, E. (2017). Towa ds a Module o Lexicog aphy in On oLex. In P oc. o he LDK wo kshops: On oLex, TIAD and Challenges o Wo dne s a 1s Language Da a and Knowledge con e ence (LDK 2017), Galway, I eland, olume 1899. Galway (I eland): CEUR-WS, pp. 74–84. h p://ceu -ws.o g/Vol-1899/On oLex{_}2017{_}pape {_}5.pd . Bosque-Gil, J., G acia, J., Mon iel-Ponsoda, E. & Aguado-de Cea, G. (2016b). Modelling Mul ilingual Lexicog aphic Resou ces o he Web o Da a: he K Dic iona ies case. In P oc. o GLOBALEX’16 wo kshop a LREC’16, Po o oz, Slo enia. Decle ck, T., Wand-Vog , E. & Mö h, K. (2015). Towa ds a Pan Eu opean Lexicog aphy by Means o Linked (Open) Da a. In I. Kosem, M. Jakubíček, J. Kallas & S. K ek (eds.) P oceedings o eLex 2015. Biennial Con e ence on Elec onic Lexicog aphy (eLex2015), elec onic lexicog aphy in he 21s cen u y: Linking lexical da a in he digi al age. Ljubljana/B igh on: T ojina, Ins i u e o Applied Slo ene S udies, Ljubljana. G acia, J., Villegas, M., Gómez-Pé ez, A. & Bel, N. (2018). The ape ium bilingual dic iona ies on he web o da a. Seman ic Web, 9(2), pp. 231–240. Kal enböck, M. & Ke ne man, I. (2017). In oducing LDL4HELTA: Linked da a lexicog aphy o high-end language echnology applica ion. Ke ne man Dic iona y News, (25), pp. 2–3. h ps://www.kdic iona ies.com/kdn/kdn25_2017.pd . Ke ne man, I. (2009). KD’s BLDS: a b ie in oduc ion. Ke ne man Dic iona y News, (17), pp. 1–2. h p://www.kdic iona ies.com/kdn/kdn17_2009.pd . 745 P oceedings o eLex 2019 Ke ne man, I. (2011). F om dic iona y o da abase: C ea ing a global mul i-language se ies. In I. Kosem & K. Kosem (eds.) Elec onic Lexicog aphy in he 21s Cen u y. New Applica ions o New Use s. P oceedings o eLex, pp. 113–121. h p://elex2011. ojina.si/Vsebine/p oceedings/eLex2011-14.pd . Ke ne man, I. (2015). A mul ilingual ilogy: De eloping h ee mul i-language lexicog aphic da ase s. Dic iona ies: Jou nal o he Dic iona y Socie y o No h Ame ica, 36(1), pp. 136–149. h p://elex.link/elex2015/p oceedings/ eLex_2015_24_Ke ne man.pd . Klimek, B. & B ümme , M. (2015). Enhancing lexicog aphy wi h seman ic language da abases. Ke ne man DICTIONARY News, 23, pp. 5–10. h ps://www.kdic iona ies.com/kdn/kdn23_2015.pd . K ek, S., Kosem, I., McC ae, J. P., Na igli, R., Pede sen, B. S., Tibe ius, C. & Wissik, T. (2018). Eu opean Lexicog aphic In as uc u e (ELEXIS). In The XVIII EURALEX In e na ional Cong ess. p. 159. McC ae, J., Aguado-de Cea, G., Bui elaa , P., Cimiano, P., Decle ck, T., Gómez-Pé ez, A., G acia, J., Hollink, L., Mon iel-Ponsoda, E., Spoh , D. & Wunne , T. (2012). In e changing lexical esou ces on he Seman ic Web. Language Resou ces and E alua ion, 46, pp. 701–719. McC ae, J., Bosque-Gil, J., G acia, J., Bui elaa , P. & Cimiano, P. (2017). The On oLexLemon Model: De elopmen and Applica ions. In I. Kosem e al. (eds.) Elec onic lexicog aphy in he 21s cen u y. P oc. o eLex 2017 con e ence, in Leiden, Ne he lands. Lexical Compu ing CZ s. .o., pp. 587–597. h ps://elex.link/elex2017/wp-con en /uploads/2017/09/pape 36.pd . McC ae, J., Tibe ius, C., Khan, F. A., Ke ne man, I., Decle ck, T., K ek, S. Monachini, M. & Ahmadi, S. (2019). The ELEXIS In e ace o In e ope able Lexical Resou ces. Deli e able, ELEXIS-Eu opean Lexicog aphic In as uc u e. Pa izi, A., Kohl, M., González, M. & Sau ’, R. (2016). Towa ds a Linguis ic On ology wi h an Emphasis on Reasoning and Knowledge Reuse. In P oceedings o he Ten h In e na ional Con e ence on Language Resou ces and E alua ion LREC 2016, Po o ož, Slo enia, May 23-28, 2016. This wo k is licensed unde he C ea i e Commons A ibu ion Sha eAlike 4.0 In e na ional License. h p://c ea i ecommons.o g/licenses/by-sa/4.0/ 746 P oceedings o eLex 2019