| Title: | Generative AI in pragmatics : assessing the accuracy of automated speech act classification in Pinter’s The birthday party |
|---|
| Authors: | ID Todorović, Tadej (Author)
ID Flogie, Andrej (Author)
ID Hari, Daniel (Author) |
|---|
| Files: |  RAZ_Todorovic_Tadej_2025.pdf (364,39 KB)
MD5: B293E0E2C8067C3C38D6E90BBCF39672
 https://doi.org/10.4312/elope.22.1.19-34
|
|---|
| Language: | English |
|---|
| Work type: | Scientific work |
|---|
| Typology: | 1.01 - Original Scientific Article |
|---|
| Organization: | FF - Faculty of Arts
|
|---|
| Abstract: | This study explores the feasibility of using generative AI (ChatGPT, Gemini, and DeepSeek) to automate speech act annotation in Harold Pinter’s play The Birthday Party. Three chatbots – ChatGPT, Gemini, and DeepSeek – were tested under three scenarios varying in the amount of theoretical material provided. Each chatbot’s output was compared to a manually annotated reference via a Python script measuring classification accuracy. Scenario 2 produced the highest accuracy overall (75–82%), while Scenario 1 underperformed, owing to incorrect reliance on external typologies, and Scenario 3 showed signs of overfitting. ChatGPT o1 emerged as the most accurate model, achieving 82% accuracy in Scenario 2. The findings suggest that GenAI chatbots can serve as valuable preliminary annotators when good prompt-engineering and well-curated theoretical material are provided. Future research could extend this methodology to more context-dependent texts, further refining prompt-engineering strategies and exploring larger linguistic corpora.Študija raziskuje smiselnost rabe generativne umetne inteligence (ChatGPT, Gemini in DeepSeek) za avtomatizacijo anotacije govornih dejanj v Pinterjevi drami Zabava za rojstni dan. Trije klepetalni roboti – ChatGPT, Gemini in DeepSeek – so bili testirani v treh scenarijih, ki so se razlikovali glede na obseg predloženega teoretičnega gradiva. Rezultati vsakega klepetalnega robota so bili primerjani z ročno anotirano različico s pomočjo Python skripte, ki je izmerila natančnost klasifikacije. Scenarij 2 je na splošno dosegel najvišjo natančnost (75–82 %), medtem ko je bil scenarij 1 zaradi neustreznega zanašanja na tuje tipologije preslab, scenarij 3 pa je kazal znake preprileganja (angl. overfitting). ChatGPT o1 se je izkazal za najnatančnejši model, saj je v scenariju 2 dosegel 82-odstotno zanesljivost. Ugotovitve kažejo, da lahko klepetalni roboti GEN-UI služijo kot koristni predhodni anotatorji, če so na voljo dobro zasnovani pozivi in dobro pripravljeno teoretično gradivo. Prihodnje raziskave bi lahko to metodologijo razširile na besedila, ki so bolj odvisna od konteksta, nadalje izpopolnile strategije inženiringa pozivov in raziskale večje jezikovne korpuse. |
|---|
| Keywords: | pragmatics, speech act analyses, ChatGPT, DeepSeek, Gemini |
|---|
| Publication date: | 07.07.2025 |
|---|
| Year of publishing: | 2025 |
|---|
| Number of pages: | str. 19-34 |
|---|
| Numbering: | Letn. 22, št. 1 |
|---|
| PID: | 20.500.12556/DKUM-93650  |
|---|
| UDC: | 81’33:821.111.09-2]:004.89 |
|---|
| ISSN on article: | 1581-8918 |
|---|
| COBISS.SI-ID: | 241882115 |
|---|
| DOI: | 10.4312/elope.22.1.19-34 |
|---|
| Publication date in DKUM: | 22.07.2025 |
|---|
| Views: | 124 |
|---|
| Downloads: | 11 |
|---|
| Metadata: |    |
|---|
| Categories: | Misc.
|
|---|
|
:
|
Copy citation |
|---|
| | | | Average score: | (0 votes) |
|---|
| Your score: | Voting is allowed only for logged in users. |
|---|
| Share: |  |
|---|
Hover the mouse pointer over a document title to show the abstract or click
on the title to get all document metadata. |
