<?xml version="1.0" standalone="yes"?>
<Paper uid="C92-1010">
  <Title>PARSING AGGLUTINATIVE WORD STRUCTURES AND ITS APPLICATION TO SPELLING CHECKING FOR TURKISH</Title>
  <Section position="8" start_page="0" end_page="0" type="concl">
    <SectionTitle>
5. Conclusions
</SectionTitle>
    <Paragraph position="0"> lit tlfi~ paper, we have presenled a lnorl)hological parser for all agghlthi;ttive lallguagc. Turkisli. all(\] its  application to spelling checking of this language. Parsing agglutinative word structures necessitates rather nontrivial phonological and morphological analyses which present special difficulties in the development of parsers for such languages, not usually encountered in parsers for other languages. As a resub, the number of parsers developed for agghitinafive languages, and particularly for Turkish, is quite limit.ed, and they have certain shortcomings. We have solved most of the problelns encountered in the previous parsers by lnaking a detailed and careful research on Turkish word formation rules and their exceptions \[20\]. These results may hopefully be helpful for future researchers on Turkish linguistics. We should note t.hat ewm though it is claimed that word formarion rules in Turkish are well-defined and Turkish is a very regular language, as used today it shows many irregularities that cause the ln'oblem of parsing I.his language to become a very hard and interesting problem Many grammar books haw~ been referred to collect, Turkish word format.ion rules, hi those books, after each rule is defined, usually it is reminded that tliel'e lllay occur sonte exceptions to that, rule ill SOllle condit,ions, hut mostly those conditions can not be &amp;quot;well&amp;quot; defined, For example, in all Turkish grammar books, it is said that &amp;quot;When a Turkish word ending wit.h one of the consonants P, ~', T, K receives a suffix beginning with a consonant, that final consonant is soft,cued, bul t, here are some such words whose final consonant does not change.&amp;quot; \]lowever, none of the books says what the common propert.y of those words which do not obey t,o that rule is, because most AcrEs DE COLING-92, NA~n'ES. 23-28 Ao\[rr 1992 4 4 PRec. OF COLING-92, NAI~rrES, AUG. 23-28, 1992 probably it is not known yet. Ill order t,o inchlde that rule correctly in the parser, all words having the indicated prol)erty have been examined, the list of t,he irregular ones have been obtained, and speeial checks have hi!ell dolle t,o catch those irregularit.ies. Ill order t.o obtain reliabh? resuhs front the spelling checker, all of the known rifles and theh' except, ions have been inlplellll!tlt,Od The spelling checker ,'4OllletillleS i'e\[)ol'LS correcI woFds aS illcorreet, ()lie reason ell, his is the absellce of SOllle words in ore' dictionary. Although the dict, ionary is reasonably complete, there still remains many technl cal terms and proper names which are not included.</Paragraph>
    <Paragraph position="1"> Adding more and nlore words will obviously increase tile flmelional performance of the checker. Another reason is that, most of the derivational sultixes are not mchtded rote'die rules. If( stem that is derived by such a suffix is not present ill the dictionary, it is reported as misspelh~d. Additionally. for th( deriw~lionel sullixes that. are included in our rules, the lis~ of the roots that they can be a\[lixed to may no( he full~ determined. This problem can also be solved by examining the dictionary As far as execution pertbrmance goes. our iml)hmtentation is very S;atisfaC/lory giving an ahnost. 1000 words/second word analysis throughput \[19\].</Paragraph>
    <Paragraph position="2"> Ac31~s DECOLING-92, NANTES, 23-28 Aour 1992 4 5 Plot. OF COLING-92, NANTES, AU6. 23-28, 1992</Paragraph>
  </Section>
class="xml-element"></Paper>