Partial colexifications reveal directional tendencies in object naming

1.1 Directionality in meaning extensions

According to Lakoff and Johnson (1999, 18), “[t]he same neural and cognitive mechanisms that allow us to perceive and move around also create our conceptual systems and modes of reason.” On the basis of this embodiment hypothesis, the body is the source of our mental representations. Therefore, the direction of meaning extension between the semantic domains of the human body and objects should proceed from the body part term to the object term. However, only a few studies in cognitive linguistics exist that have investigated the extension of meanings between two concrete domains (e.g., Winter and Srinivasan 2022). In comparison, historical linguists have documented many examples of meaning extensions between two concrete domains over time and across diverse languages (e.g., Wilkins 1996; Brown and Witkowski 1981; Urban 2011).

Semantic change refers to the change of the meaning of a word. The processes underlying this phenomenon are interpreted as being based on cognitive mechanisms. For example, the perception of contiguity between the concept hood expressed in the Latin word capitium is the basis for the development of the Spanish word cabeza for head (Steinberg 2014, 258). Studies by Traugott and Dasher (2001) and Wilkins (1996) indicate that semantic changes are more regular than previously assumed. Wilkins (1996, 272) documented many examples of semantic changes between the domains of the human body and objects. Apart from body part terms being the sources, these examples include semantic changes in which the direction of a semantic change goes from the object domain to the human body domain. This includes metonymic semantic changes of words for clothing items in Dravidian languages: footling → foot, earring → ear, and pubic tassel → penis. Similarly, Brown and Witkowski (1981, 603-604) showed that words expressing the concept egg are extended to include the concept testicle due to the similarity of the round shape. Furthermore, Koch (2008, 128) demonstrated that the words for egg or ball frequently extend their meanings to eyeball. These examples are in contrast to the embodiment hypothesis. However, no systematic study investigating the directionality of meaning extensions between the two concrete domains of the human body and objects exists to date.

A systematic investigation of directionality consists of two parts. First, the source and target domains need to be determined. Second, the causes of the meaning extensions need to be analyzed. Since both parts of the analysis are laborious, studies have so far concentrated on only one of the two parts. Urban (2011) examined a selection of pairs of concepts, including human body parts and objects such as skin-bark, mouth-estuary, and tongue-flame. Based on an analysis of the asymmetry in the morphological complexity of word forms for each of the concepts across 149 languages, Urban (2011) determined the target and source concept of semantic changes. For example, many languages have a morphologically complex word form for bark (e.g., Yuki Ɂ ol šil ‘tree skin’), whereas only a few languages have a morphologically complex word form for skin including the word for bark Urban (2011, 8). Based on this observation, Urban (2011) concluded that skin is the source concept and bark is the target concept. Winter and Srinivasan (2022) used Urban’s judgments of source and target concepts and investigated the factors that lead to meaning extensions between two domains. They tested the claim that the direction of meaning extension is from concrete to abstract domains by comparing English concreteness ratings and word frequency norms. Their results showed that word frequency is a better predictor of semantic change than concreteness. Based on a large-scale historical corpus, Xu et al. (2017) showed that asymmetry in other scales between two concepts leads to meaning extensions from embodied to disembodied, external to internal, less valenced to more valenced in English. The studies suggest that it is likely to find more meaning extensions from the domain of the human body to the domain of objects and that multiple factors play a role in the emergence of semantic changes.

1.2 Partial colexifications

With the advancement of computer-assisted methods, the comparison of vocabularies across different languages has become more efficient. Especially the theoretical construct “colexification”, introduced by François (2008), has attracted the interest of cognitive scientists and led to studies investigating the variation of words expressing distinct concepts with large-scale datasets including many diverse languages (e.g., Jackson et al. 2019; Xu et al. 2020; Brochhagen and Boleda 2022). The basis of these studies is the implementation of colexification networks, which show the connections between concepts in a semantic space. A prominent implementation of this approach is the Database of Cross-Linguistic Colexifications (CLICS, Mayer et al. 2014; List et al. 2018; Rzymski et al. 2020). CLICS includes methods and tools for data representation and analysis of colexifications across 3,156 language varieties. It was the first use case to demonstrate the benefits of curating datasets in a standardized format, i.e., the Cross-Linguistic Data Formats (CLDF, Forkel et al. 2018). Building on this achievement, List et al. (2022) presented a large collection of multilingual word lists: Lexibank. This resource was used to identify 78 body-object colexifications in 396 language varieties by Tjuka (2024b). These colexification networks were based on one-to-one matches where a single word form was colexified with two concepts.

The study by List (2023) presented the first methodical approach for automatically computing partial colexifications. Partial colexifications occur when a word consists of more than one morpheme, one of which expresses a different concept, for example, river mouth. It is not straightforward to infer partial colexifications from word lists because shared morphemes between words may reflect grammatical distinctions, resulting in a noisy network with many coincidental colexifications. The methods and workflows provided by List (2023) allow researchers to extract different types of partial colexification patterns from multilingual word lists. The most relevant type of partial colexifications for the present study are affix colexifications, which occur when one word form denoting a certain concept recurs at the beginning or the end of another word form expressing another concept. Note that the term “affix colexification” corresponds to the definition of “affix” in computer science, not linguistics, since the method does not identify morphemes but merely detects formal similarities. The methods presented by List (2023) infer affix colexifications based on a common substring between two sequences. For example, the substring ABC is a common substring of XYZABC. If two sequences overlap in at least one common substring of length 3, the algorithm defines it as an affix colexification. In a recent study, Bocklage et al. (2024) used affix colexifications to test the predictions of directionality in semantic change made by Urban (2011) on a larger language sample and found support for the less strict version of his hypothesis. The study showed that overt marking as reflected in automatically inferred affix colexifications seems to predict semantic change to some degree, provided that full colexifications are also attested for the concept pairs in question. Bocklage et al. (2024) also included a comparison with data from the Database of Semantic Shifts (Zalizniak et al. 2024), which contains a collection of semantic shifts documented in the world’s languages.

The underlying assumption of the present study is that directional tendencies can be predicted by analyzing cross-linguistic patterns of affix colexifications. This assumption is based on the hypothesis that the mechanisms driving semantic motivation in word formation exhibit similar directional preferences as those governing semantic change (Koch and Marzo 2007). Based on the results by Bocklage et al. (2024), we assume that there is a positive correlation between semantic motivation and semantic change and that we can infer the directions in a colexification network with the methods described in List (2023). The present study uses the methods and data presented in List (2023) and Tjuka (2024b) to investigate the directionality of partial colexifications between the semantic domains of the human body and objects. We test the prediction that meaning extensions move predominantly from the body domain to the object domain.

3.1 Directional tendencies

The frequency across languages with which a body-object colexification occurred either in the direction that the word form for a body concept is used to express the object concept or vice versa, was the basis for the analysis of directionality. Table 1 shows the directional tendencies between body and object concepts including the number of languages with an affix colexification for each direction and the total number of occurrences. The dominant direction was determined by evaluating the counts in each direction, and if there was a direction with at least two instances more in one direction, that direction was determined to be the dominant one. Twenty-one out of 39 colexifications show a directional tendency from body to object (blue arrows), while 16 colexifications show a directional tendency from object to body (orange arrows). In two cases – intestines-sausage and lip-shore – no directional tendency exists (yellow arrows).

The most frequent body-object colexification is ear-earring in which the word form for ear is used to express the concept earring in 66 language varieties. Examples include kula-pepeiao lit. ‘gold-ear’ in Hawaiian or sau faliɳa lit. ‘king ear’ meaning earring in Rotuman, an Austronesian language. The second most frequent body-object colexification is skin-bark where the word form for skin is used to express bark in 42 language varieties, for example, in Kalamang, a West Bomberai language, ror kulun lit. ‘tree/wood skin’. The third most frequent body-object colexification with a directional tendency from body to object is neck-collar, as in sipluw tor͂ lit. ‘neck cloth’ in Mansi, a Uralic language, or ynĩ teɁ lit. ‘neck clothing’ in Chatino, a Zacatepec variety of the Otomanguean language family.

While many different body concepts are used to express object concepts, there are particular body concepts for which the source are object names. The concept testicles has predominantly affix colexifications consisting of names for objects such as egg, fruit, seed, and ball. For example, zaad.bal lit. ‘seed.ball’ in Dutch. This directional tendency is due to taboo conventions with respect to reproductive body parts in many cultures. Another body concept that is expressed with words for objects in many languages is shoulder blade. However, these body-object colexifications are an areal phenomenon that occurs mainly in the Nakh-Daghestanian languages (see also Tjuka 2024b).

Within the 15 most frequent body-object colexifications, which occur in at least ten language varieties, nine body-object colexifications show a directional tendency from body to object concepts. In comparison, only five frequent body-object colexifications show a directional tendency from object to body concepts. It appears that the more frequently a body-object colexification is, the more likely a transfer from body to object occurs. However, this generalization should be treated with caution because in many cases the scarcity of data can lead to infrequent patterns. Object concepts such as shovel or root and body concepts such as blood vessel or throat are not commonly represented in the word lists. The interpretation of the frequencies must therefore be understood as a tendency rather than a universal.

3.2 Network representations

The discussion of frequencies suggests a slight tendency for body concepts to be the source of object names. However, the picture becomes more complex when examining the network of affix colexifications. Here, we present three subgraphs of the affix network based on List (2023) to illustrate the different ways in which words for body and object concepts are used to express other concepts. In the graphs, body concepts are represented by blue ellipses and object concepts by orange rectangles.The thickness of the edges indicates the frequency across languages and the arrows indicate the target concepts.

Figure 1:

Subgraph with the body concepts skin and body. Body concepts = blue ellipsis; object concepts = orange rectangle; thickness of edge = frequency across languages; arrow = directionality.

Figure 1 shows the subgraph with the body concepts skin and body and the object concepts bark, tree, and tree trunk. The network reveals that there are many languages in which the word forms for the concepts skin and tree are used for the concept bark. The word forms for tree are also frequently used to express tree trunk and for example, in Ende, a Pahoturi language, llo pätt lit. ‘tree body’. There are other cases in which the word forms for body are used for bark. In these cases, it is likely that skin and body are colexified and then the word form is used to express bark. Interestingly, the network shows that there are a few affix colexifications in which the word forms for skin express body, but not the other way around.

Figure 2:

Subgraph with the body concepts testicles. Body concepts = blue ellipsis; object concepts = orange rectangle; thickness of edge = frequency across languages; arrow = directionality.

Figure 2 illustrates the different directional tendencies between the body concept testicles and the object concepts seed, egg, and fruit. Word forms for object concepts are predominantly the basis for expressing the concept testicles, while the word forms for testicles occur less frequently in the expression of the object concepts. In addition, the word forms for the object concepts are also used to express the other object concepts, such as in Hawaiian hua-Ɂai lit.’ egg-food’ for fruit. There are also many languages that have full colexifications between the three object concepts.^[2] The semantic motivations that lead to particular affix colexifications can therefore not straightforwardly be distinguished.

Figure 3:

Subgraph with the body concepts mouth and lip. Body concepts = blue ellipsis; object concepts = orange rectangle; thickness of edge = frequency across languages; arrow = directionality.

Figure 3 shows the subgraph with the colexifications between the body concepts mouth and lip and the object concepts edge and shore. It becomes apparent that the word forms for mouth are frequently used to express lip and that the word forms for edge are frequently used to express shore. An example of the former is wal bii lit. ‘leaf mouth’ in Polci, an Afro-Asiatic language, and an example of the latter is wini. ^Ɂdanum lit. ‘water.edge’ in Wapishana, an Arawakan language. In some cases, the word forms for edge are used to express lip, as in vom dor͂ lit. ‘mouth edge’ in Komi, a Uralic language. Although general tendencies arise, there are multiple cases in which word forms are sources for either body or object concepts.