Publication: Türkçe Sözlük’te tarifler (uzman görüşüne ve yapay zekâya göre)
Abstract
Türkçe Sözlük’te tarifler (uzman görüşüne ve yapay zekâya göre) Sözlükler, tarih boyunca insanların en temel başvuru kaynakları olmuştur. Dilin devamı, insanların anlaşması, ortak bir kelime hazinesinin oluşması sözlüklere bağlıdır. İnsanlar hayatlarında tarif edebildikleri şeyleri kabul ederler. Dolayısıyla bir şey, tarif edilemiyorsa o şeyin insan zihninde tam bir karşılığının olmadığı söylenebilir. Bu sebeple sözlükler, kelimelerin kimlikleridir; kelimeler sözlüklerle var olurlar. Kelime öğretimi denilince Türkçe dersleri ön plana çıkmaktadır. Maarif zerine inşa edilen Türkçe ders kitaplarındaki etkinliklerde Türkçe Sözlük’ün kullanılması teşvik edilmektedir. Dolayısıyla Türkçe Sözlük’ün niteliği, öğrencilerin dil gelişiminde ve fikir dünyalarının oluşumunda oldukça etkilidir. Dilin oluşması, düşüncenin de oluşmasını sağladığı için sözlüklerdeki tarifler; insanların doğru düşünmesini, benzerlikleri ve farklılıkları analiz edebilmesini, yeni fikirler bulmasını ve bu şekilde de nitelikli üretim yapabilmesini sağlar. Bu bakımdan sözlüklerdeki tariflerin anlaşılırlığı ve tutarlılığı oldukça mühimdir. Bu çalışmada da tarif yapılırken uyulması gereken ilkeler doğrultusunda TDK Türkçe Sözlük’teki (TS) tarif sorunlarının tespiti, analizi ve çözüm tekliflerinin getirilmesi aynı zamanda üretken yapay zekâ ve uzman görüşü destekli kategori bazlı tarif çerçevelerinin oluşturulması amaçlanmaktadır. Literatür tarandığında TS üzerinde farklı araştırmacıların çeşitli sorunları tespit ettiği saptanmıştır ancak bu sorunların çözümüne yönelik tekliflerin teorik olduğu, uygulamaya dönük çalışmaların pek fazla bulunmadığı görülmüştür. Bu sebeple TS’de belirlenen kategorilerin her birinde problemli olan onar madde başı için problemlerin tespitinde ve tarif çerçevelerinin oluşturulmasında uzman görüşü ile beraber üretken yapay zekâ sohbet robotlarına başvurulmuştur. Aynı zamanda problemli olan maddeler oluşturulan çerçeveler esas alınarak üretken yapay zekâ sohbet robotlarına tarif ettirilmiştir. Üretken yapay zekâ sohbet robotlarından yararlanılması alanda ilklerden biri olması bakımından oldukça önemlidir. Nitel araştırma yöntemlerinden durum araştırması deseninde yapılan çalışmada veriler, doküman analizi ve görüşme tekniği ile toplanmıştır. Çalışmada doküman incelemesi ile toplanan veriler yönlendirilmiş içerik analizi yapılarak, yarı yapılandırılmış görüşmeler yoluyla toplanan veriler içerik analizi yapılarak değerlendirilmiştir. Çalışmanın sonucunda ilk olarak tarifte uyulması gereken ilkeler doğrultusunda TS’deki tarif sorunları 19 problem durumu altında tasnif edilerek örnek maddelerle analiz edilmiştir. İkinci olarak TS’de belirlenen 8 kategorinin her birinde problemli olan onar madde başı, üretken yapay zekâ ve uzman görüşü neticesinde oluşturulan çerçevelere göre yeniden tarif edilmiştir. Madde başlarının tarifinde ön plana çıkan bazı problemler şunlardır: Aynı kategoride olan üyelerin tariflerinde içeriğe konu olan özellikler bakımından tutarsızlıklar olması; tariflerin eksik veya yüzeysel olması; tariflerin belirsizlikler içermesi, açık ve anlaşılır olmaması; örneklerin yetersiz, anlama ve üretim becerilerine katkı sağlayacak nitelikte olmaması vb. ChatGPT ve DeepSeek sohbet robotlarının problem tespitinde ve aynı kategorideki kelimelerin tarifinde kullanacakları çerçeveleri oluşturmada başarılı oldukları uzman görüşleri neticesinde teyit edilmiştir. ChatGPT’nin açıklamalarının -uzman görüşüne paralel- daha genel ve kapsayıcı olduğu; DeepSeek’in cevaplarının ise daha detaylı bilgiler içerdiği tespit edilmiştir. İki modelin de tariflerde, örneklere göre daha başarılı oldukları ancak yanlış, eksik bilgiler verebildikleri için sözlük bilimcinin kontrolünde kullanılması gerektiği sonucuna varılmıştır. Çalışmada oluşturulan çerçeveler, daha fazla uzman görüşü alınarak geliştirilebilir; farklı kategoriler için tarif çerçeveleri oluşturulabilir. Başka üretken yapay zekâ modelleri üzerinde çalışılarak tarif ve örnek üretmede en etkili dil modelinin ne olabileceği üzerinde disiplinler arası çalışmalar yapılabilir.
Definitions in the Türkçe sözlük (according to expert opinions and artificial intelligence) Dictionaries have been the most fundamental reference sources for people throughout history. The continuity of language, human understanding, and the formation of a common vocabulary depend on dictionaries. People accept things they can define in their lives. Therefore, if something cannot be defined, it can be said that it has no complete counterpart in the human mind. For this reason, dictionaries are the identities of words; words exist through dictionaries.When it comes to teaching vocabulary, Turkish lessons come to the fore. The use of the Türkçe Sözlük is encouraged in the activities in Turkish textbooks based on the Maarif Programme. Therefore, the quality of the Türkçe Sözlük has a significant impact on students’ language development and the formation of their intellectual world. Since the formation of language enables the formation of thought, the definitions in dictionaries enable people to think correctly, analyse similarities and differences, find new ideas, and thus produce quality work. In this respect, the comprehensibility and consistency of the definitions in dictionaries are very important. This study aims to identify and analyse definition problems in the TDK Türkçe Sözlük (TS) in line with the principles that should be followed when making definitions, to propose solutions, and to create category-based definition frames supported by generative artificial intelligence and expert opinion. A review of the literature reveals that various researchers have identified different problems in the TS, but it has been observed that the proposed solutions to these problems are theoretical, and there are not many application-oriented studies. For this reason, generative artificial intelligence chatbots were consulted, along with expert opinion, to identify problems and create definition frames for ten problematic entries in each of the categories identified in the TS. At the same time, the problematic entries were defined by generative artificial intelligence chatbots based on the frames created. The use of generative artificial intelligence chatbots is quite important as it is one of the first in the field. In the study, which was conducted using the case study design from qualitative research methods, data were collected through document analysis and interview techniques. In the study, the data collected through document review were evaluated by conducting a guided content analysis, and the data collected through semi-structured interviews were evaluated by conducting a content analysis. As a result of the study, first, the definition problems in TS were classified under 19 problem situations in line with the principles to be followed in the definition and analysed with sample entries. Second, the ten problematic headword in each of the eight categories identified in TS were redefined according to the frames created as a result of generative artificial intelligence and expert opinion. Some of the problems that came to the fore in the definition of headwords are as follows: Inconsistencies in the definition of members in the same category in terms of the characteristics subject to the content; definitions being incomplete or superficial; definitions containing ambiguities, not being clear and understandable; examples being insufficient and not contributing to comprehension and production skills, etc. Expert opinions have confirmed that ChatGPT and DeepSeek chatbots are successful in identifying problems and creating frames to be used in defining words in the same category. It was found that ChatGPT’s explanations were more general and inclusive, in line with expert opinion, while DeepSeek’s answers contained more detailed information. It was concluded that both models were more successful in definitions than in examples but could provide incorrect or incomplete information and should therefore be used under the supervision of a lexicographer. The frames created in the study can be developed by obtaining more expert opinions; definition frames can be created for different categories. Interdisciplinary studies can be conducted on other generative artificial intelligence models to determine the most effective language model for generating definitions and examples.
Definitions in the Türkçe sözlük (according to expert opinions and artificial intelligence) Dictionaries have been the most fundamental reference sources for people throughout history. The continuity of language, human understanding, and the formation of a common vocabulary depend on dictionaries. People accept things they can define in their lives. Therefore, if something cannot be defined, it can be said that it has no complete counterpart in the human mind. For this reason, dictionaries are the identities of words; words exist through dictionaries.When it comes to teaching vocabulary, Turkish lessons come to the fore. The use of the Türkçe Sözlük is encouraged in the activities in Turkish textbooks based on the Maarif Programme. Therefore, the quality of the Türkçe Sözlük has a significant impact on students’ language development and the formation of their intellectual world. Since the formation of language enables the formation of thought, the definitions in dictionaries enable people to think correctly, analyse similarities and differences, find new ideas, and thus produce quality work. In this respect, the comprehensibility and consistency of the definitions in dictionaries are very important. This study aims to identify and analyse definition problems in the TDK Türkçe Sözlük (TS) in line with the principles that should be followed when making definitions, to propose solutions, and to create category-based definition frames supported by generative artificial intelligence and expert opinion. A review of the literature reveals that various researchers have identified different problems in the TS, but it has been observed that the proposed solutions to these problems are theoretical, and there are not many application-oriented studies. For this reason, generative artificial intelligence chatbots were consulted, along with expert opinion, to identify problems and create definition frames for ten problematic entries in each of the categories identified in the TS. At the same time, the problematic entries were defined by generative artificial intelligence chatbots based on the frames created. The use of generative artificial intelligence chatbots is quite important as it is one of the first in the field. In the study, which was conducted using the case study design from qualitative research methods, data were collected through document analysis and interview techniques. In the study, the data collected through document review were evaluated by conducting a guided content analysis, and the data collected through semi-structured interviews were evaluated by conducting a content analysis. As a result of the study, first, the definition problems in TS were classified under 19 problem situations in line with the principles to be followed in the definition and analysed with sample entries. Second, the ten problematic headword in each of the eight categories identified in TS were redefined according to the frames created as a result of generative artificial intelligence and expert opinion. Some of the problems that came to the fore in the definition of headwords are as follows: Inconsistencies in the definition of members in the same category in terms of the characteristics subject to the content; definitions being incomplete or superficial; definitions containing ambiguities, not being clear and understandable; examples being insufficient and not contributing to comprehension and production skills, etc. Expert opinions have confirmed that ChatGPT and DeepSeek chatbots are successful in identifying problems and creating frames to be used in defining words in the same category. It was found that ChatGPT’s explanations were more general and inclusive, in line with expert opinion, while DeepSeek’s answers contained more detailed information. It was concluded that both models were more successful in definitions than in examples but could provide incorrect or incomplete information and should therefore be used under the supervision of a lexicographer. The frames created in the study can be developed by obtaining more expert opinions; definition frames can be created for different categories. Interdisciplinary studies can be conducted on other generative artificial intelligence models to determine the most effective language model for generating definitions and examples.