ChatGPT i prepoznavanje metafora u hrvatskom jeziku

master's thesis

• Title (english): ChatGPT’s ability to recognize metaphors in croatian
• Author: Tomislav Veble
• Mentor: Daniela Katunar (mentor)
• Mentor: Božo Bekavac (komentor)
• Committee member: Daniela Katunar (predsjednik povjerenstva)
• Committee member: Božo Bekavac (član povjerenstva)
• Committee member: Ivana Simeon (član povjerenstva)
• Granter: University of Zagreb; Faculty of Humanities and Social Sciences; (Department of Linguistics); Zagreb
• Defense date and country: 2024-09-25, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing; Artificial Intelligence
• Scientific / art field, discipline and subdiscipline: HUMANISTIC SCIENCES; Philology; General Linguistics

Abstract

ChatGPT je jezični model koji koristi sposobnosti napredne umjetne inteligencije kako bi odgovarao na zadanu ulaznu informaciju ili input generiranjem odgovora koji imitiraju prirodni jezik. Zbog toga što ChatGPT imitira prirodan ljudski jezik, potrebno je provoditi istraživanja na njemu kojima se preispituju njegove sposobnosti prepoznavanja, razumijevanja i korištenja jezika. Cilj ovog istraživanja bio je testirati koliko dobro ChatGPT prepoznaje metafore na hrvatskom jeziku. Putem dva različita testa, od kojih jedan koristi ulomke prirodnog jezika pronađene u hrWac korpusu, a drugi ciljano osmišljene rečenice, testirali smo ChatGPT-jeve sposobnosti prepoznavanja 45 metaforičkih izraza na hrvatskom jeziku, koji pripadaju domenama tri različite konceptualne metafore: SPORT JE SILA, SPORT JE RAT i ŽIVOT JE HRANA. Dodatnim testom smo pokazali da ChatGPT odlično prepoznaje metafore u hrvatskom jeziku, ali da određenim postupcima utječemo na vjerojatnost da će neki jezični element označiti metaforičnim, kada taj element to ustvari nije.

Abstract (english)

ChatGPT is a language model using advanced-AI capabilities to answer a specific input, by generating an answer that imitates natural, human language. Precisely because ChatGPT imitates natural language, it is necessary to continue researching its language recognition and identification capabilities, as well as usage. The goal of this study was to examine how well does ChatGPT recognise metaphors in croatian. By using two different tests, comprising of paragraphs of natural language retrieved from the hrWac corpus, as well as carefully thought-out sentences, we tested out its capabilities to recognise 45 different croatian metaphorical expressions. These expressions were sourced from two different domains of discourse, and are manifestations of three distinct conceptual metaphors: SPORT IS FORCE, SPORT IS WAR and LIFE IS FOOD. Additional testing has shown that ChatGPT identifies metaphorical expressions with a high degree of accuracy, but we have also shown that certain practices actively impact the possibility that the model will tag something as a metaphor even when it is not.

• Keywords: ChatGPT; prepoznavanje; konceptualna metafora; veliki jezični modeli
• Keywords (english): ChatGPT; identification; conceptual metaphor; large language models
• Language: croatian
• URN:NBN: urn:nbn:hr:131:657177
• Study programme: Title: Linguistics (double major); specializations in: General Linguistics, Comparative Linguistics, Computational Linguistics, Aplied Linguistics and Cognitive Linguistics, Cognitive; Course: Computational Linguistics; Study programme type: university; Study level: graduate; Academic / professional title: magistar/magistra lingvistike (magistar/magistra lingvistike)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Open access
• Created on: 2024-09-27 08:05:33