Abstract:

The capture and identification of timeless expressions is a natural language processing task of critical importance for situations that require meaningful output. Today, the leading language processing technologies in the field of artificial intelligence require time expressions to be resolved in many research and application stages. This article introduces the first temporary expression capture and identification system in the literature for Turkish. Again, for the first time in literature, a test data set that can be used in the studies on this subject has been created and presented to the service of researchers. The open-source system is based on the HeidelTime architecture that is commonly used for other languages and has achieved an average of about 90 percent of success for four different types (date, time, time, remarkable time expressions) on the data set created. The system is evaluated by different assessment criteria in different literature and compared to the main Turkish asset name recognition and automatically created HeidelTime language sources. It is expected that the developed system will complete an important structure that is missing in Turkish natural language research and will benefit from further studies.

Keywords:

Abstract:

Temporal expression recognition and disambiguation is a critical stage for natural language processing tasks that require semantic inference. Nowadays, language processing technologies (a subfield of artificial intelligence) require the analysis of temporal expressions in many phases. In this article, a temporal expression recognition and disambiguation system for Turkish is presented for the first time in the literature. Additionally, again for the first time, a Turkish temporal expression test data set has been created and made publicly available for researchers. The introduced system was developed on HeidelTime architecture which is frequently used for other languages. On average, 90 percent performance was achieved for four different types of temporal expressions (date, time, duration, and set) on the introduced data set. The system performance is evaluated under different evaluation criteria and compared with a baseline named entity recognizer and HeidelTime automatically created language resources. It is anticipated that the system fills an important gap in the Turkish natural language studies so far and will benefit future studies.