Introduction

One of the most important things people do with language is talk about other people. Through reference to ourselves and to other humans we share social information and encode, enact, and alter social relations. Even though human reference is a universal behavior, it is also a system with many choices [1]. The choices that a person has in any given moment in talk (how will I refer to this person who is simultaneously a father, a policeman, a Māori, Jack, my husband, your son?) are greatly diversified across languages in regard to semantics, combinability, grammatical status, and their role in linguistic and cultural practices.

Within the domain of human reference, one fundamental choice is whether or not to employ a kinship term. Every language has words for kin. These typically emerge early in a child’s linguistic development [2]. Kin terms are inherently relational; one cannot be a mother without having a child. The use of kinship terms highlights this relationality as the key to formulating reference to others: a person is identified through one or several other persons (e.g., English Mary’s mother, Dalabon yengkulngu ‘boy who is the nephew of one of us and the son of the other (we being opposite-sex siblings)’, [3]) rather than through an independent characteristic they possess as an individual (e.g., woman, policeman, tall person, Mary). Kinship terms are thus central to sociality, with the power to define, track, regulate and propagate networks of relationships (cf. [4]).

Do all cultures, then, formulate reference to others equally often in terms of kinship? We know there is massive variation in the meaning, complexity, and formal expression of kinship term sets across languages [5], but we know little about what influences the frequency of kinship mentions in spontaneous language use [6], nor what affects the prevalence of kinship as opposed to other formulation strategies in human reference. Here, we test factors that potentially associate with the frequency of kinship formulations in discourse (as compared with other human reference expressions), and we do so across 22 diverse languages.

In particular, we put the spotlight on a central question of diversity and dynamicity in language, cognition, and culture: is the usage frequency of terms from a semantic domain (here: kinship) associated with the integration of this domain into the grammar? For the kinship domain, we operationalise this by measuring kintax [7,8], the use of morphosyntactic means for encoding kinship in grammar. These include, for instance, special kinds of pronouns or agreement marking [9,10] for particular types of kin relationship, or dyadic markers that derive expressions with meanings like ‘mother and child’ from a noun meaning ‘mother’ or ‘child’ [11]. Although all 22 languages in our sample have kinship terms, only fifteen have some form of kintax as part of their grammar.

Unravelling the relationship between frequency of use of a semantic domain and its presence in grammar is central to understanding three key questions. (a) Why are the world’s 7,000+ languages so strikingly diverse? (b) What are the consequences of this diversity for how they are used and learned? (c) How do use and learning interact with the emergence of different grammatical structures? [12–16] Previous studies suggest that frequency of use is associated with the grammaticalisation of lexical items to grammatical elements (e.g., [17,18]), as well as being a factor in the elaboration of lexical distinctions [6,19] and, specifically, the velocity of lexical change within the kinship domain [20].

However, the causal relations are complex (cf. [21]), and the vast majority of the world’s languages lack accessible historical data. This makes it difficult to compare causal trajectories which might show whether greater frequency of use leads to more extensive grammaticalisation (here: greater frequency of kinship formulation leading to the development of kintax) or vice-versa. At this stage we address the first step in investigating questions of causality, namely determining whether any statistical association can be found. Matched contemporary data allow us to measure if the presence of certain grammatical features is a statistical predictor of whether kinship formulation is employed more frequently in the languages that have them. The question we thus ask in this article is whether there is a demonstrable relationship between the grammatical encoding of a semantic domain and how often that domain is evoked in spontaneous discourse. In the discussion, we delve into the direction of causality in such a relationship.

A serious methodological barrier to demonstrating an association between frequency of use and grammatical differences has been the lack of language corpora which are cross-linguistically comparable but which are unbiased by text founder effects. These text founder effects arise in parallel translation corpora (e.g., Bible translations), because the source language is in a privileged position in terms of which semantic categories get expressed, distorting the frequency of occurrence in the target languages. The reverse, comparing dissimilar corpora across languages, can introduce spurious differences. For example, a corpus of marriage and inheritance jurisdictions in one language would probably show a higher proportion of kinship references than a corpus of gardening manuals in another language. This would not reflect cross-linguistic differences in kinship term frequency, but the effects that genre-skewed sampling can have on virtually any semantic domain. Spontaneous conversation can avoid these pitfalls (e.g., [22,23]), but may be impenetrable in regard to features such as human reference without intimate knowledge of the real-life context. To overcome these barriers while ensuring comparability across languages, ‘parallax’ corpora [24] elicit spontaneous natural speech and signing across a range of languages in response to a common, constrained set of picture card stimuli, while allowing participants to choose their own ways of talking, without translation bias. Here we investigate the use of kin- versus non-kin-based formulations in human reference when people are presented with the same stimuli, using the Family Problems Picture Task: two participants describe a series of 16 picture-cards, negotiate to put them into a sequential order, and then narrate them as a story to a third (naïve) participant [25].

In our comparison, we include data from 21 spoken languages and one signed language. The languages come from all over the world (see Fig 1 for a map of the languages’ locations) and from 16 different language families, thus reducing the risk of biasing our results by repeated sampling from language families or areas (see also SI-2.5 in S1 Appendix). Table 1 includes the languages’ names (and their ISO codes), language clades (families), countries, macro-regions, and details of which researchers collected and annotated the data.

Download:

• PNG
  larger image
• TIFF
  original image

Table 1. Languages in the present study.

https://doi.org/10.1371/journal.pone.0355784.t001

Download:

• PNG
  larger image
• TIFF
  original image

Fig 1. Map of approximate locations where data were collected.

The map was created with the R packages rnaturalearth [26] and rnaturalearthdata [27], which draw on public domain data from Natural Earth.

https://doi.org/10.1371/journal.pone.0355784.g001

Since each language in the study has a range of terms available for talking about people, including kinship terms (aunt), generic terms (woman), role nouns (policewoman) and names (Maria), the choice of reference at any given moment is not forced by the task, but remains each participant’s (conscious or unconscious) decision. At the same time, by hypothesis, users of different languages may weight these choices in different ways. As an initial (and typical) illustration, contrast the examples from Kogi in (1) and German in (2), which both describe the first card of the task, reproduced in Fig 2.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 2. ‘Homecoming’ card, stimuli example.

(The figure is in the public domain and available at https://scopicproject.wordpress.com/run-the-task/ and https://scopicproject.wordpress.com/wp-content/uploads/2018/01/stimuli-pictures.pdf).

https://doi.org/10.1371/journal.pone.0355784.g002

Example (1) shows a kin-rich string of utterances from Kogi: every human character is identified through a kinship term. The speaker first describes the old man character as a grandfather, reckoning his identity from the perspective of the child character; then identifies the young man as a son in relation to the grandfather; and then switches perspective again to that of the young man, as he refines his assessment of their relationship, before identifying the young woman as a relative (the wife) of the young man.

(Examples in this paper are displayed using interlinear glossing; see at the end for a list of abbreviations used. Information in square brackets indicates the recording of the picture task and the time interval of the example. The recordings are available in [28].)

1. [1] hẽki    hate-dweba      hẽki   a-skwá
2. [2]  dem=foc  grandfather-old      dem=foc  3sg.poss-son
3. [3]  ezhi    a-hwäsgwi        hálde=ki  ahí      munzhi
4. [4]  or     3sg.poss-father.in.law   dem=foc   3sg.poss    woman/wife

‘This is the grandfather. This is his [the old man’s] son. Or his [the young man’s] father in law. This one is his [the young man’s] wife.’

[SocCog_kog01-CNC_130619_1 - 00:00:11-00:00:18]

Example (2) illustrates a German language description of the same scene that uses common nouns, descriptors for age differences and a role (dieser Besuch ‘this visitor’) rather than kin terms. The speaker relates characters to each other through comparative age, and no family ties are made explicit.

1. (2) Da    drauf     sind    ein     älterer   Mann
2. med    on.top      be.prs.pl  indf.sg.m   elderly.sg.m man
3. eine    Frau     ein     Kind     und   dieser    Besuch,
4. indf.sg.f    woman    indf.sg.n  child     and   dem.sg.m   visitor
5. ein     junger    Mann
6. indf.sg.m   young.sg.m     man

‘On it (the picture) are an elderly man, a woman, a child, and this visitor, a young man.’

[SocCog-deu01-hs_ks_HR_RS_PV2019-10-30 - 00:00:42-00:00:47]

The lexical options for human reference across languages are broadly comparable. For example, both German and Kogi have kin and non-kin nouns available in the lexicon, but the speakers of (1) and (2) made different choices about when to deploy which. We examine whether particular patterns of semantic construal (for example, whether one says the father as opposed to the man) are associated with the presence of grammatical complexity within the semantic domain of kinship. If grammatical structure is independent of the frequency with which words in the same semantic domain are used, we will not observe a relationship between the presence of kintax and the prevalence of kin-based formulations in our sample. However, if grammatical structure directly relates to formulation frequency, we expect to see statistical advantages from including kintax as a predictor in our model.

Having a multilingual corpus with many occurrences of human reference allows us to examine a preponderance of evidence: for each instance of reference to a human, we assess the factors that may influence whether the token is a kinship word or not. We build a generalized linear statistical model to show which factors matter most often in determining this (subconscious or conscious) choice by language users. In addition to kintax, we examine three further variables as possible predictors of kinship expression in the data, described and motivated in more detail in the section ‘Candidate predictors of kin term frequency and statistical model details’ and S1 Appendix SI-2: complexity of a language’s sibling-term system, discourse type (conversation vs narration), and scene content (type of picture-card being viewed, e.g., solitary character vs intergenerational group of characters). Evaluating the contribution of these factors helps to identify other potential influences on reference practice, and gives an indication as to what other features may regulate any observed relationship between grammaticalisation and lexical use frequency.

Materials and methods

The languages and their features

Linguists with expertise in the relevant languages coded the corpus data for the presence of lexical noun-phrase reference to humans, and each instance was annotated as a kin or non-kin expression in ELAN [30,31] (n = 18,841 tokens after exclusions: datapoints from participants who produced less than four lexical human referent terms during their session were removed – these were usually tokens from the third naïve participant who was present for the latter phase of the task). Overall, more than half (53%) of the references to humans are carried out with kinship terms. Token numbers by language vary from 55 (Idi – one collected run of the task) to 5,319 (Balinese – 25 collected runs of the task). This variability is a feature of documentary data collection, where access to participants is often limited, expensive and dependent on the kinds of data that communities want to collect. Especially for under-described languages, we see even limited data points as a contribution to our knowledge of different languages and communities, so they are included in our analysis.

Table 2 shows the numeric and proportional uses of kin terms in the corpus data for each language, along with a measure of the structural (kintax) and lexical (sibling term complexity) designation for each language (see the section ‘Candidate predictors of kin term frequency and statistical model details’ for a discussion).

Download:

• PNG
  larger image
• TIFF
  original image

Table 2. Kin term usage by language (ordered by proportion of reference) and measures of language-dependent predictors tested. (Kin proportion calculated based on total terms, not as average of languages' kin proportions).

https://doi.org/10.1371/journal.pone.0355784.t002

Results

Fig 3 shows plots of the model probabilities for each predictor variable plus distributions for each language in our sample.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 3. Kin-based formulations model probabilities by each predictor variable plus distributions for languages.

https://doi.org/10.1371/journal.pone.0355784.g003

Our model has significant (at p < 0.05) predictors (fixed effects) due to structural kintax score (Z = 3.56, p < 0.001), but not due to sibling term lexical complexity (Z = 1.48, p = 0.14). Scenes with intergenerational configurations of people (Z = 3.69, p < 0.001) are more likely to be described or narrated with kinship terms than solitary depictions, showing that most task participants placed different-aged characters into kinship relationships with each other. The largest effect we see is from discourse type, where narrative phases of the task are much more likely to use kinship terms (Z = 13.87, p < 0.0001).

Our results clearly show that people who use languages which have more kinship-based grammar (kintax) use kinship terms more often than others, even taking into account other plausible predictors. Therefore, we can feel confident that having kintax is predictive of higher usage of kinship-based formulations in a language, establishing an empirical link between the possession of a grammatical structure and frequency of formulation.

Kintax is not the only determining factor, though: Narrative discourse was strongly associated with higher kinship reference frequency. We also see the frequency of kinship formulations being influenced by the number and kinds of characters visible in the picture stimuli. This suggests an influence of discourse type on reference patterns, and a general propensity for people to infer and express kin relationships in relation to depictions of co-present individuals, whereas solitary individuals are less likely to be put in relational configurations. We do not, however, see a significant link between the complexity of a language’s sibling term vocabulary and users’ frequency of kinship formulations. Therefore, while structural elaborateness in the kinship domain is a predictor for kinship formulation frequency, lexical elaborateness in this domain is not (given our operationalisations).

Discussion

What is very clear from our results is that having at least some kinship grammar is associated with a higher frequency of use of kin-based formulations, and having more elaborated kintax is likewise associated with a higher rate of kin-based formulations (Fig 3). Thus lexical usage patterns in spontaneous discourse do correlate with structural patterns in grammar.

At this stage, it is not clear whether the relationship between having kintax and increased kinship word usage is causal nor, if it is, what the direction of causation would be. Does the presence of grammatical categories that pick out kinship make the concept of kinship more salient and more readily selected, a result of ‘thinking for speaking’ [36]? Or is it that the frequent choice to formulate reference in terms of kinship rather than some other way, iterated over many generations, drives the emergence of kintax – that ‘grammar is restless and earned’ ([37], p. 60) and that ‘grammars code best what speakers do most’ ([38], p. 363)? There is also a third possibility: that some further variable, like the greater cultural importance of kinship, could influence both usage frequency and the emergence of kintax, with no direct causal relationship in either direction between frequency and kintax.

Proper evaluation of the first two causal scenarios would require longitudinal studies, unfolding over centuries or even millennia, for a good sample of languages. To evaluate the third, we would need some independent way of evaluating the cultural importance of kinship (cf. [39]), something that would require careful operationalisation and more fieldwork. Yet, we can suggest the potential importance of feedback loops (cf. [23]), with differential frequency driving the emergence of different grammars, which then in their turn influence language users’ choices of formulation. The first two causal scenarios might therefore reinforce each other.

Even more complex causal mechanisms might be at work, and we now mention two of these briefly.

Firstly, it is likely that the effects of formulation frequency are stochastic rather than absolute, and are path-dependent in the sense of interacting with other features of the language. For example, see [8] on the intricacies of how Murrinhpatha sibling-dual pronouns evolved, and [7] for arguments that kinship-sensitive pronouns may only develop when kin-dyad constructions combine with inclusory constructions [40], in which ‘me and dad’ is expressed as ‘we dad’ (i.e., ‘dad’ is included in the referent set of ‘we’); the conjunction of these two conditions is a far from common situation.

Secondly, we might also expect other discourse practices, such as the existence of special registers sensitive to kinship relations like ‘mother-in-law’ or ‘avoidance’ registers (e.g., [8,41,42]), to influence attention to kinship relations. However, social practices which look superficially similar may have quite different effects on usage frequency of kin-based formulations. For example, many societies have name taboos, and this can promote the use of kinship terms as an alternative to proper names, potentially accelerating the development of kintax (see [41] on Murrinhpatha and [43] on Yélî Dnye). However, Ku Waru, which also has name taboos, has very low rates of kinship formulation and kintax. A possible reason is that Ku Waru has a different solution to the collision of name taboos with the challenge of how to clearly refer to others, namely the use of ‘food names’. ‘Food names’ link pairs of individuals based on some event where they shared a food item, so that one may say, e.g., ‘your banana’ in place of ‘your uncle’, thus allowing clear reference without invoking kin. Although there are no uses of these ‘food names’ in the Ku Waru picture task data, the fact that they are frequently used in everyday life by Ku Waru people in place of kin terms may be relevant to the otherwise unexpectedly low use of kin terms in our Ku Waru corpus, where, instead, generic referents such as words meaning ‘man’ or ‘woman’ are frequent (see also, e.g., [44,45] on other New Guinea groups). Such particularistic cultural taboos may thus affect usage frequency, and underline the importance of distinguishing the degree to which kinship is an organising factor in daily life (a fact about culture) from the frequency with which kinship relations are mentioned in talk (a fact about discourse).

Increasingly, the diversity of the world’s languages is seen not just as a kaleidoscopic reshuffling of the same factors, but as the result of numerous selection pressures, varying across language communities, that differentially favour the emergence of certain linguistic phenomena in certain language communities [46–49]. These factors may be genetic (e.g., palate shape), climatic (e.g., altitude or humidity), demographic (e.g., population size, proportion and influence of mother-tongue users vs immigrants), but also cultural, as we have started to show here via a cross-linguistic study of human reference. Cultural influence operates by favouring different patterns of discourse, themselves reflecting different cultural conventions in how humans talk about the world, then generating frequency patterns which eventually precipitate into grammatical differences. The methods employed in this article show us that deep differences exist in how members of different cultures formulate their talk about the social world, and that this correlates significantly with which grammatical structures are found.

Interlinear glossing

The first line in language examples (such as (1) from Kogi) is a transcription in the orthography of the language, with hyphens to show breaks between morphemes (word parts). The gloss in the second line provides information about the meaning of each morpheme. The final part of the example is a translation into English. Abbreviations used in the glosses are:

dem  demonstrative

f   feminine gender

foc  focus marker

indf  indefinite

m   masculine gender

med  mediative

n   neuter gender

pl   plural

poss  possessive

prs  present tense

sg   singular

Supporting information

Acknowledgments

We would like to gratefully acknowledge the community members and consultants who participated in the data collection and facilitated fieldwork.