In this paper, we introduce a general linguistic creativity test for humans and Large Language Models (LLMs). The test consists of various tasks aimed at assessing their ability to generate new original words and phrases based on word formation processes (derivation and compounding) and on metaphorical language use. We administered the test to 24 humans and to an equal number of LLMs, and we automatically evaluated their answers using OCSAI tool for three criteria: Originality, Elaboration, and Flexibility. The results show that LLMs not only outperformed humans in all the assessed criteria, but did better in six out of the eight test tasks. We then computed the uniqueness of the individual answers, which showed some minor differences between humans and LLMs. Finally, we performed a short manual analysis of the dataset, which revealed that humans are more inclined towards E(extending)-creativity, while LLMs favor F(ixed)-creativity.

In recent years, Large Language Models proved themselves worthy competitors of humans in various creative domains. Studies have shown them to be on a par with humans or even better, when tested for several types of creativity. However, the nature of their creative process is fundamentally different from humans, which leads to differences in their creative output, which are not yet fully understood. In this study, we investigate the degree and the nature of the differences between Large Language Models and human answers from a dataset composed of responses to tan ideational creativity test previously proposed in in the literature. We automatically analyzed the dataset as follows. We first extracted socio-linguistic features and computed the Linguistic Style Matching score from the dataset, using the LIWC tool. Then we performed style embeddings clustering to see how the human and LLM answers to the creativity test group together. Moreover, we computed and compared several types of readability indexes. Finally, we performed sentiment analysis and topic modeling on the dataset. The results showed a variety of dissimilarities between human and LLM answers to the creativity test, from significative differences in stop words, part of speech, and tense usage, to tone, emotions, and readability. Moreover, LLMs tend to generate more general and impersonal creative responses than humans, who rely more on their own experience and expectations.

Until fairly recently, creativity was a human-specific characteristic. Computational creativity or artificial creativity was established as a domain in the late 90’s, with different fields such as verbal, musical, or graphical creativity. With the latest technological advances and the appearance of Large Language Models (LLMs), creativity as a feature of machines gained more and more interest in the scientific community. The scope of this study is twofold: to design a comprehensive benchmark for verbal creativity assessment of LLMs and then to run the same creativity tests on different LLMs as well as on humans, for a direct comparison. We aimed to raise the replicability and extensibility of the creativity assessment of LLMs. Hence, we adapted different types of creativity tests and different criteria from psychology to fit the LLMs profile. We also employed computer-assisted evaluation methods, by using the Open Creativity Scoring with Artificial Intelligence (OCSAI), as we wanted to focus exclusively on automated approaches to assessing creativity. We quantitatively and qualitatively analyzed the data set of both human and machine-generated answers and interpreted the results. Finally, we provide both the original verbal creativity test that we have designed, and the curated data comprising all the collected answers, from the LLMs and from the humans that participated in this research.

The identification of cognates is a fundamental process in historical linguistics, on which any further research is based.Even though there are several cognate databases for Romance languages, they are rather scattered, incomplete, noisy, contain unreliable information, or have uncertain availability.In this paper we introduce a comprehensive database of Romance cognates and borrowings based on the etymological information provided by the dictionaries (the largest known database of this kind, in our best knowledge).We extract pairs of cognates between any two Romance languages by parsing electronic dictionaries of Romanian, Italian, Spanish, Portuguese and French.Based on this resource, we propose a strong benchmark for the automatic detection of cognates, by applying machine learning and deep learning based methods on any two pairs of Romance languages.We find that automatic identification of cognates is possible with accuracy averaging around 94% for the more difficult task formulations.

We train and evaluate two models for Romanian stress prediction: a baseline model which employs the consonant-vowel structure of the words and a cascaded model with averaged perceptron training consisting of two sequential models ‐ one for predicting syllable boundaries and another one for predicting stress placement. We show in this paper that Romanian stress is predictable, though not deterministic, by using data-driven machine learning techniques.

We show in this paper how the computer science concept of ‘continuations’, together with categorial grammars and a type shifting mechanism, is able to account for a wide range of natural language semantic phenomena, such as hierarchical discourse structure, ellipses, accommodation and free-focus and bound-focus anaphora. The merit of continuations in the dynamic semantics framework is that they abstract away from assignment functions that are essential to the formulations of Dynamic Intensional Logic, Dynamic Montague Grammar, Dynamic Predicate Logic and Discourse Representation Theory, Thus, continuation style semantic do not pose problems such as the destructive assignment problem in Dynamic Predicate Logic or the variable clash problem in Discourse Representation Theory. We argue that continuations are a versatile and powerful tool, particularly well suited to manipulate scope and long distance dependencies, phenomena that abound in natural language semantics.