Publication: Omsieve - Çözüm Kümesi Programlama Tabanlı Muhakeme Edilen Esgönderge Sieve Çözümlenmesi
Loading...
Date
2017-01-01
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
OmSieve Proje?sinde Çözüm Kümesi Programlama yöntemini kullanarakEsgönderge Çözümlenmesini farklı açılar ile inceledik. Çözüm Kümesi Programlama(ÇKP), tekdüze olmayan nedensellestirme islemlerini, bilgi gösterimini, ve karmabilgi tabanlı nedensellestirmeyi destekleyen genel amaçlı bir mantık programlamayapısıdır. Bir ÇKP mantık programında, mantıksal kavramlar arasındaki iliskiyive cözümlerdeki kısıtları içeren olası çözüm kümelerini tanımlarız. Bu sekildegösterilen bir ÇKP çözümcü (bir yazılım aracı) belirlenen iliskilere ve kısıtlarabaglı çözümleri hesaplamaktadır. Esgönderge Çözümlenmesi dogal dilde metinedilen sözcük öbeklerinin dünyadaki aynı varlıga denk geldigini bulmayı saglayanBilgisayar Dilbilimi konusudur. Örnegin, ?O, insanlara: ?Yardımınıza ihtiyacımvar? dedi? ana cümledeki özne olan kisi ?O?, ve ?ihtiyacım? kelimesini söyleyenkonusmacıyı tanımlarken, ?insanlar? ve ?yardımınız? kelimesindeki ?siz? kelimesidinleyicileri tanımlamaktadır. OmSieve projesinde, asagıda belirtilen arastırmakatkıları tarafımızdan gerçeklestirilmistir: (I) ?Sieve? esgönderge çözümlenmesiyaklasımının kurallarını biçimsel bir sekilde analiz ettik, (II) ÇKP için basamaklıicra çerçevesi gelistirdik, (III) orjinal Sieve?i bu çerçeve içerisinde olusturduk,(IV) ilerleyen kısıtlı basamaklar yerine tahminde bulunabilen karma Sieve?iolusturduk, (V) Sieve?i degerlendirip Stanford Dcoref sistemiyle karsılastırdık,(VI) ODTÜ-Sabancı Türkçe Agaç Yapılı Derlemi için bir esgönderge tanımlamaprojesi organize ettik, (VII) birçok etiketlemelerden otomatik olarak bir altınstandard kurmaya yönelik yeni bir metodu gelistirdik, (VIII) Marmara TürkçeEsgönderge Derlemi olusturduk, ve (IX) Türkçe esgönderge çözümlenmesi içinyeni bir referans sistemi yaratıp bunu kendi derlemimizi kullanarak teyid ettik.Bu proje süresince yapılan isler ile 2 adet tam metin konferans makalesi, bir kitapbölümü ve bir dergi makalesi (SCI-E) yayınlanmıstır. Diger dergi makalesi isedergiye gönderilme asamasındadır. Ayrıca, Marmara Türkçe Esgönderge Derlemionline ve herkese açık olarak bir Git veri havuzunda yayınlanmıstır.
In the OmSieve Project we work on several aspects of Coreference Resolution using methods of AnswerSet Programming. Answer Set Programming (ASP) is a general purpose logic programming formalism,that supports nonmonotonic reasoning, comfortable representation of knowledge, and reasoning withhybrid knowledge bases. In an ASP logic program we describe a set of potential solutions, relationshipsbetween concepts in the solution, and constraints on solutions. Given such a representation anASP solver (a software tool) computes those solutions that adhere to the specified relationships and constraints.Coreference resolution is the Computer Linguistics task of finding out which phrases of a naturallanguage discourse refer to the same entity in the world. For example in the sentence “He said to thepeople: ’I need your help’” the task is to find out that “he” and “I” refers to the same entity (the speaker),furthermore “the people” and “your” refers to the same entity (the listeners). In the OmSieve project wemade the following research contributions: we (I) analyzed the rule-based ‘Sieve’ approach to coreferenceresolution, (II) developed a stepwise execution framework for ASP (III) realized the original Sievein that framework, (IV) realized a hybrid Sieve that performs guesses instead of following strict steps,(V) evaluated the Sieve and compared it to the Stanford Dcoref system, (VI) organized a coreferenceannotation project for the METU-Sabanci Turkish Treebank, (VII) created a new approach for automaticallyadjudicating coreference annotations, (VIII) assembled the Marmara Turkish Coreference Corpusbased on that project, and (IX) created a baseline for Turkish Coreference Resolution and verified it onour new corpus. The work done in this project has been published in 2 full-length conference papers,one book chapter, and one journal (SCI-E). A further journal is under review (SCI-E). Furthermore theMarmara Turkish Coreference Corpus has been published online as a public Git repository.
In the OmSieve Project we work on several aspects of Coreference Resolution using methods of AnswerSet Programming. Answer Set Programming (ASP) is a general purpose logic programming formalism,that supports nonmonotonic reasoning, comfortable representation of knowledge, and reasoning withhybrid knowledge bases. In an ASP logic program we describe a set of potential solutions, relationshipsbetween concepts in the solution, and constraints on solutions. Given such a representation anASP solver (a software tool) computes those solutions that adhere to the specified relationships and constraints.Coreference resolution is the Computer Linguistics task of finding out which phrases of a naturallanguage discourse refer to the same entity in the world. For example in the sentence “He said to thepeople: ’I need your help’” the task is to find out that “he” and “I” refers to the same entity (the speaker),furthermore “the people” and “your” refers to the same entity (the listeners). In the OmSieve project wemade the following research contributions: we (I) analyzed the rule-based ‘Sieve’ approach to coreferenceresolution, (II) developed a stepwise execution framework for ASP (III) realized the original Sievein that framework, (IV) realized a hybrid Sieve that performs guesses instead of following strict steps,(V) evaluated the Sieve and compared it to the Stanford Dcoref system, (VI) organized a coreferenceannotation project for the METU-Sabanci Turkish Treebank, (VII) created a new approach for automaticallyadjudicating coreference annotations, (VIII) assembled the Marmara Turkish Coreference Corpusbased on that project, and (IX) created a baseline for Turkish Coreference Resolution and verified it onour new corpus. The work done in this project has been published in 2 full-length conference papers,one book chapter, and one journal (SCI-E). A further journal is under review (SCI-E). Furthermore theMarmara Turkish Coreference Corpus has been published online as a public Git repository.
Description
Keywords
Bilgisayar Bilimleri, Donanım ve Mimari, Bilgisayar Bilimleri, Teori ve Metotlar