When children are learning to speak they often use a word to refer to other objects or living beings whose name they do not know, for example, they might refer to a balloon by using the term ‘ball’. Beyond children, the speakers of a language also generate new meanings of words with the passage of time. This explains why the term ‘mouse’ is currently used to refer to the device with which we interact with the computer, in addition to denominating a rodent. But what patterns does so-called lexical creativity follow?

Recently, an article analysing this has been published in the journal Science, stemming from research conducted by UPF, with the support of the University of Toronto (Canada). The main conclusion of the study is that children’s lexical creativity and that of the speakers of a language (which explains the historical evolution of the meanings of words) follow common patterns. The study has gone even further and shown that these patterns are shared by communities of speakers of up to 1,400 different languages.  Thus, for example, the research considers European languages such as Spanish, Catalan, Basque, Galician, German, French, Portuguese, Dutch, Danish or Norwegian. It also includes languages from other parts of the world such as Swahili, Arabic, Mandarin Chinese, Hindi and Korean.

Thomas Brochhagen (UPF), first author of the article:  “This possible relationship between individual learning and the evolution of languages in terms of how meaning is organized had not been demonstrated and this study does so on a large scale and in a generalized way”

The article, titled From language development to language evolution: A unified view of human lexical creativity, was written jointly by Thomas Brochhagen (UPF), Gemma Boleda (UPF and ICREA researcher), Eleonora Gualdoni (UPF), and Yang Xu (University of Toronto). The UPF researchers belong to the Computational Linguistics and Linguistic Theory (COLT) research group of the UPF Department of Translation and Language Sciences. The researcher from the University of Toronto is attached to the Cognitive Science programme at its Department of Computational Science.

For the first author of the paper, Thomas Brochhagen, the research “suggests that such diverse processes as children’s language learning during their first months of life, on the one hand, and the evolution of the world’s languages over hundreds of years, on the other, share a common cognitive basis”. “This possible relationship between individual learning  and the evolution of languages in terms of how meaning is organized had not been demonstrated and this study does so on a large scale and in a generalized way”, Brochhagen adds.

Computational models to measure the factors that predict lexical creativity

The research team has used computational models to analyse lexical creativity during the process of linguistic development of children and of society as a whole. These computational models are trained to detect which factors come into play on the different linguistic phenomena that may explain the creation of new meanings of words. The three phenomena that the research has taken into account are the following.

The first is overextension, the process whereby the meaning of a word is extended, so that it is used to name other objects or human beings. It serves to explain children’s lexical creativity (for example, when they use the term ‘ball’ to denote a balloon). 

The second phenomenon is colexification, whereby a word serves to name what is expressed by two different words. For example, the Catalan word “poble” is used to name both a settlement and the group of people who live in a given territory. It has a greater influence on the historical evolution of the lexicon of a language.

The third phenomenon is semantic change, the modification of one of the meanings of a word that can take place during a language’s historical evolution. For example, this phenomenon explains that “cold” or “bitter” have extended their meaning and that, in addition to referring to temperature or taste, respectively, they can also allude to a person’s character.

Conceptual association, the most predictive factor of lexical creativity

The research set off from a theoretical framework that considers that four factors affect the previous linguistic phenomena:

• Affective or emotional factors: for example, a child may call a cat a dog or vice versa because it arouses a similar emotion in them.
• The visual similarity of two objects or living beings:  for example, we can refer to a lamp by using the word “light”.
• The association between two concepts: for example, when we say “mobile” for mobile phone and omit the word “phone”. This phenomenon occurs when, upon hearing one word, we mentally associate it with another.
• The taxonomy, the hierarchical classification of the terms of the same family: for example, by ‘way’ I may refer to a road (paved), because a road is a type of way.

Affective factors only explain children’s lexical creativity

The results of the research show that conceptual association is the factor that has the greatest predictive value over the three previous phenomena (overextension, colexification and semantic change). They also show that affectivity only has a high predictive value in children’s lexical creativity through overextension, while its influence is irrelevant to explain the other two phenomena. In contrast, visual similarity and taxonomy have greater predictive value over colexification and semantic change than overextension.

Verification systems of the predictive model used for 1,400 languages are applied

Despite the differences between the reasons that explain the lexical creativity of children and that of the community of speakers as a whole, many of the predictive factors are common in all cases. Hence, the computational models designed to measure the predictive factors of one of the above phenomena (colexification, for example) are also effective to measure the factors that influence the other two phenomena (overextension and semantic change). This has been confirmed by the research through an estimation of the predictive factors in which the different models have been crossed.

It should also be noted that the research has applied several verification systems in order to check the robustness of these predictive models in various languages, other than English. As a result of applying these verification systems, the robustness of these predictive models has been verified in 1,400 languages.

Reference article:

Brochhagen, T., Boleda, G., Gualdoni, E., Xu, Y. (2023). From language development to language evolution: A unified view of human lexical creativity. Science, Vol. 381, N. 6656. https://doi.org/10.1126/science.ade7981