Chat GPT (Generative Pre-trained Transformer) is a groundbreaking artificial intelligence-powered language model that can create texts as if they were written by a human. Developed by OpenAI, a US-based research organization founded by Elon Musk, Sam Altman and other visionaries, Chat GPT uses advanced GPT-4 technology to process natural language. OpenAI’s mission is to develop artificial intelligence in a way that is accessible and safe for all of humanity. Chat GPT works by analyzing as well as processing huge amounts of text, from articles as well as books to websites, thereby gaining extensive knowledge on a wide variety of topics. The key to its success is a mechanism called a transformer, which enables it to effectively analyze context and generate answers that are not only consistent, but also sensible and relevant, even for complex questions.
The study by Dr. Rafal Mazur of Jagiellonian University analyzed the linguistic correctness of texts generated by ChatGPT, one of the state-of-the-art generative AI models. The goal was to explore the capabilities and limitations of this technology, especially in the context of the Polish language. Mazur used a free version of ChatGPT based on the GPT-3.5 language model to generate texts based on Polish language matriculation instructions. The analysis included an assessment of the linguistic correctness of the generated texts, using Andrzej Markowski’s typology of errors, which includes grammatical, syntactic, lexical, dictionary, phraseological, word-forming, phonetic, stylistic, spelling and punctuation errors.
Syntactic Correctness
The survey found that ChatGPT had the most problems with correct use of Polish syntax, accounting for 35% of all errors. There were frequent errors in the syntactic connectives of compounds of agreement and government, and in the forms of judgments and subjects. Examples of errors include:
- „Jednym z tych utworów, które doskonale ilustrują tę problematykę, jest dramat „Antygona” Sofoklesa oraz poemat „Konrad Wallenrod” Adama Mickiewicza”.
- „Natomiast poezja Adama Mickiewicza, zwłaszcza utwory „Dziady cz III” oraz „Konrad Wallenrod”, przedstawiają inną perspektywę na sprzeczności
człowieka” - „Wielu pisarzy i poetów, poprzez swoje utwory literackie, starali się zgłębić tę skomplikowaną naturę człowiek”.
Lexical Errors
Lexical errors accounted for 24% of all errors. ChatGPT often creates unusual word combinations and confuses the meanings of words. Examples include:
- „Makbet staje się namiętnie ambitny i gotów popełnić morderstwo, aby osiągnąć swoje cele”.
- „Fragmenty Księgi Rodzaju i Apokalipsy św. Jana przedstawiają dylematy moralne, wybory i konsekwencje, które dotykają człowieka”.
- „Rozważając człowieka jako istotę pełną sprzeczności, nie sposób nie odnieść się do postaci Makbeta, głównego bohatera tragedii Szekspira”.
Punctuation and Orthography
Punctuation errors accounted for 22% of all errors. The unnecessary use of commas and inappropriate separation of words or phrases by commas were common. Examples include:
- „W kontekście tych utworów literackich, warto odwołać się także do kontekstu filozoficznego i egzystencjalnego”.
- „Podsumowując, zarówno w „Makbecie” Szekspira, jak i w „Konradzie Wallenrodzie” Mickiewicza, autorzy ukazują człowieka jako istotę pełną sprzeczności”.
Despite the advancement of technology, the ChatGPT model still has difficulties with the correct use of the Polish language. The future of using ChatGPT is full of promise and possibilities. As the technology develops, we can expect to see more and more of its use in various areas of life. ChatGPT is already finding applications in creating marketing content, supporting customer service, generating ideas for movie scripts, and even writing scientific articles.
Syntactic and lexical errors result from a lack of deep understanding of context and grammatical rules. Punctuation problems can be caused by differences between languages. ChatGPT does well in creating stylistically neutral statements, but repetitive phrases can make texts perceived as unnatural. Despite its limitations, artificial intelligence is a useful tool that can speed up the text creation process. However, it always requires careful editing and proofreading.
Sources:
Rafał Mazur, O POPRAWNOŚCI JĘZYKOWEJ TEKSTÓW GENEROWANYCH PRZEZ SI NA PRZYKŁADZIE CHATUGPT, https://journals.akademicka.pl/lv/article/view/5756/5360
