<?xml version="1.0" standalone="yes"?>
<Paper uid="W99-0506">
  <Title>On Some Aspects of Le rtcal Standardtzauon On Some Aspects of Lexical Standardization</Title>
  <Section position="5" start_page="43" end_page="44" type="concl">
    <SectionTitle>
4 Conclusion
</SectionTitle>
    <Paragraph position="0"> Standardizing lexicons represent an interesting intellectual and practical endeavor Past experience at CRL m developing, processing and using many large lexicons for several tasks, including machme-translauon systems, machine-aided translation tools, and mformatmn processmg systems shows that a first set of dlfhculttes hes m the lack of a standard format that Is flexible enough to cover many dlffeient languages and applications, but sutficlently l lgid to enable the use of a single lexlcal toolset shared across all these languages and apphcatlons This problems have been addressed by developing a generic dictionary software architecture that is now use to manage several large d,cuonartes designed for machine-aided translation as well as for machine translauon The second set of problems is almost as acute as the first It is very difficult to start des,gnmg a sound lexlcal architecture from scratch, hst the all the features that must be present for a variety of NLP apphcauons, predict the interaction between the various sub-structures, and predict the needs of the various NLP tools that would be access,ng the dlcuonary This has been done many times at CRL and thin knowledge is m part incorporated in the generic standard lexical structure briefly presented ,n Secuon 3 When developing a new dictionary, the linguist must use a pie-defined dtcuonary entry structure and follow a set of gmdelmes for defining the language-specific features This guarantees that the dlctlonary can be developed and maintained us,rig a standard dictionary management toolset, and that the reformat,on contained in the dictionary can actually be used for a variety of NLP at~phcauons which requllements are not always obv,ous for a non-expert The construction o! such a standard lextcal structure ~s stdl howevel an open task some areas are defined with more precision than others We started from a tairly unconstrained structuie and cross- null On Some A~pects of Le.ucal Standm dtzatton language work brought out commonalmes that have been progress,rely incorporated m the standard structure Although the standard structure presented m th~s paper has been now stable for over a year, further research and expenmentauon couldyteld new constraints that could be ,ncorporated m the archltecture Finally, the problem of hngumUc standards per se Is addressed only parually by the definmons of gmdehnes and the use of a standard lexlcal entry structure In particular, a standard entry structure imposes a spectfic orgamzatton of the hngu~st~c mformatmn encoded m an entry It defines the kind of hngmst,c reformation to be encoded and how to structure this mformauon In a multdmgual setting, it xs probably possible to define multdmgual types, such as a standard hst of part-of-speech However, our exper,ence on more than 6 different languages show that trying to estabhsh a set of multlhngual types Is not worth the effort the use of a standard lex~cal structure allows the lmgu~st to narrow down rapidly on the inventory of language-spec,fic types which can then be hsted with relatlve ease Standardlzauon ol lex~cal content ~s sull a very much open problem, and th~s research area related to the quest tol a umve~sal grammar In the Boas project (Nlrenburg &amp; Raskm 98), the hngmst defines language-specific propert,es using a knowledge ehcltauon system that contains knowledge about the set of possible hngmst~c parameters and values The hngutst ~s graded through a set of queries and answers, the result ot which Is a hngulstlc profile of a language From thls language profile, the goal is to generate automatically the set ot language-specific lexlcal propemes that define the schema ol a d~cuonary</Paragraph>
  </Section>
class="xml-element"></Paper>