Lexical tectonics: Mapping structural change in patterns of lexification

1.1 Comparing lexical structures: From synchrony to diachrony

Systematic lexicology, pioneered by structuralists (Hjelmslev 1943; Ullmann 1957; Coseriu 1964), was revived recently through studies in lexical typology, bringing to light various ways in which lexicons can be compared across languages (see Vanhove 2008a; Urban 2012; Juvonen and Koptjevskaja-Tamm 2016; Koptjevskaja-Tamm and Liljegren 2017).

When comparing how languages cut up the semantic space, one efficient approach consists in breaking up a given lexical domain into a number of potential meanings, independent of their expression in individual languages. The “etic grid” formed by these potential senses then serves as a background against which emic categories of particular languages can be compared (Koptjevskaja-Tamm et al. 2007; Evans 2010).

Several authors have advocated semantic maps as a powerful way to visualize the results of such cross-linguistic comparison (Haspelmath 2003; François 2008; Georgakopoulos et al. 2016). Until recently, most authors have adopted a synchronic approach to lexical maps (e. g., Rakhilina and Reznikova 2016: 113); however, the field shows increasing interest towards adapting the map approach to diachronic semantics (Georgakopoulos and Polis 2018; 2021).

In recent decades, a number of authors have used diachronic maps as they explored the historical development of a given semantic field in particular languages – e. g., Matisoff (1978) for body parts in Tibeto-Burman; Evans (1992), Wilkins (1996) for semantic reconstruction in various Australian languages; Pawley (2005) for exploring the meanings of a Proto-Oceanic etymon; J. François (2007; 2013) for some polysemies in French and German; and so on. Coming in the wake of these authors, the present study intends to reflect on some general principles and methods pertaining to the use of maps in the study of lexical change, and to propose ways to systematize their use in historical lexicology.

As we attempt to model the evolution of lexical structures, semantic maps will appear as a powerful device to capture graphically the way in which every semantic innovation redesigns lexical domains and their internal organisation.

1.2 This study: Data sources and outline

My observations will focus on two language groups for which the history of semantic change can be solidly reconstructed: Romance and Oceanic. Romance is a family with a well-known history, and a linguistic ancestor that is well understood.

Oceanic languages, in turn, have no written tradition, and are only known in their modern forms. However, their shared ancestor Proto Oceanic (POc) can be reconstructed with reasonable confidence thanks to the Comparative method, based on its 500 modern descendants (Pawley and Ross 2006; Ross et al. 2016). Among these 500 Oceanic languages, 138 (François et al. 2015) are spoken in the Vanuatu archipelago. This study will focus on the 17 languages of the Torres and Banks Islands of north Vanuatu (Map 1 shows their number of speakers and the three-letter abbreviations I use in this paper).

Map 1

The languages of the Torres and Banks Islands, in northern Vanuatu.

My data originates in personal fieldwork carried out on the 17 languages of the area, combining language immersion, recordings of connected speech, and linguistic elicitation by means of a personal questionnaire (François 2019).

Regular sound correspondences for northern Vanuatu languages are exposed in François (2005) for vowels, and François (2016: 31) for consonants. When a linguistic innovation is shared by Torres and Banks languages together, it can be captured by positing an intermediate common ancestor called PTB “Proto Torres–Banks” (François 2011a; 2016) – itself a descendant of Proto Oceanic. Lists of PTB lexical reconstructions appear in François (2005; 2013; 2016).

The present study will explore the diachrony of lexical structures, and propose ways to conceptualize and represent that diachrony. I will focus on patterns of lexification – though it must be acknowledged that other aspects, not dealt with in the present work, also contribute to a lexicon’s organisation: e. g., lexical gaps, combinatorics, phraseological routines (Koptjevskaja-Tamm and Liljegren 2017; Schapper 2021).

Section 2 will first outline the methodology and conceptual framework underlying our approach. Then Section 3 will discuss instances, first of structural stability, and secondly of structural innovation – based on Romance and Oceanic data. Section 4 will discuss the phenomenon of lexical competition, a key for understanding the process of lexical replacement and structural change.

Our final discussion will propose a typology of structural innovations in the lexicon – and provide some of them with empirical illustrations from firsthand data. This will give us the opportunity to discuss a phenomenon that is frequent in northern Vanuatu: the tendency for innovative lexical structures to realign with those of their neighbours, yielding a process of language-internal relexification.

2.1 Patterns of lexification and semantic change

I propose to define patterns of lexification, taken as a synchronic property, as in (1):^[1]

In the lexicon of modern English, a word like trunk may refer to a portion of a tree, but also to a part of the human body, or a quite different part of an elephant; it may also refer to a large box, or a part of a vehicle; etc. The word trunk covers a certain semantic territory, while drawing contrasts with adjacent lexical units of the same semantic fields (e. g., bole, torso, chest, snout, suitcase, boot).

If we now take a diachronic perspective, what we call semantic change happens when those patterns of distribution between senses and forms are modified:

Or, as Ullmann (1957) puts it:

A semantic change will occur whenever a new name becomes attached to a sense and/or a new sense to a name.

(Ullmann 1957: 171)

Thus the noun corn, which used to be a generic term for all sorts of grain, specialized historically in British English as a synonym of wheat, while in US English it now refers to maize. Over time, the correspondence between forms and senses has evolved.

To take a spatial metaphor, we can say that the semantic territory covered by the word corn has moved, as it were, across the semantic space. The present study, titled “lexical tectonics”, intends precisely to build on this spatial view of semantic change, and see how it can inspire a method of data visualization that can make the most of semantic maps.

2.2 Establishing senses

Discussing semantic change amounts to analyzing how the correspondences of forms and senses evolves over time. One methodological question, then, is how we can identify senses in the first place.

A widespread conception is the notion that semanticists should start their work, prior to any observation, by establishing an ontology of semantic primitives, that would exist independently of their linguistic expressions. Two schools could then be followed:

1. One approach, which may be called top-down, can endeavour to reach the most abstract, generic senses possible (similar to the semantic primes of Wierzbicka 1992, Goddard 2001). Then these ultra-generic semantic primitives would combine to yield specific meanings: e. g., ‘boy’ results from the combination of ‘child’ + ‘male’.

2. One diametrically opposed approach, this time bottom-up, could propose to break up the semantic universe into as many specific meanings as conceivable (e. g., all possible subtypes of snow, of sand, of stone…). This would compose a maximal etic grid of low-level atomistic meanings, which are grouped differently by natural languages.

It is more solid to ground the definition of senses on the empirical observation of natural languages, by focusing only on those semantic contrasts that are actually attested:

The empirical observation of languages thus becomes the key to identifying a list of potential senses, which in turn can help feed their comparison.

To take an example inspired by Hjelmslev (1943), it would be speculative to wonder aprioristically how many distinct senses in total might be assigned to the domain of wood and woody materials. The Danish noun træ may designate a tree in the wild, or the hard matter of which it is made; it may refer to timber used in carpentry, or to the wood added to fuel a fire… If we consider that word on its own, we have no principled way of deciding whether we are dealing with a single vague meaning, or four or more “senses”. But then cross-linguistic comparison shows that Danish træ sometimes corresponds to English tree, and sometimes to wood: this is enough to identify empirically two distinct senses (one covered by Eng. tree, the other one by wood), and just say that Danish expresses them identically. The reasoning can be recursive, as we observe that Spanish not only has a separate word for ‘tree’ (árbol), but two nouns for ‘wood’: leña ‘firewood’ vs. madera ‘wood as material’. Finally, Mwotlap (Oceanic, Vanuatu) uses a single word qētēnge for {‘tree’ + ‘wood as material’} but has a separate word lēt for ‘firewood’.

In sum, we can compare the patterns of lexification across four languages by just taking into account these three attested senses, and by observing how they are distributed across lexical forms – see Table 1. Given that data set, one can account for the observed semantic diversity by acknowledging exactly these three different senses – neither more nor less.

As more lexical systems are brought into the comparison, even more patterns can be established (e. g., many languages in Australia and New Guinea use the same word for ‘firewood’ and ‘fire’ – Schapper et al. 2016), and the list of attested senses is bound to grow. But crucially, the patterns attested are not random or infinite. For every lexical domain, it is generally possible to come up with a finite list of attested senses, relevant to cross-linguistic comparison.

The last column in Table 1 constitutes an etic grid, a list of attested senses for a given semantic field. Just like what happens in phonology, languages then differ in their emic categories, i. e. the language-specific lexemes that encode these senses. This is how lexical typology can incorporate a universal dimension (the etic points in the grid) while acknowledging the unique profile of each lexicon (the emic categories embodied by each lexeme).

2.3 Describing lexical structures through dis- and co-lexification

Languages may be compared in their patterns of colexification – that is, their ability to encode two distinct senses using the same lexeme (François 2008):

For example, English colexifies ‘wood as material’ and ‘firewood’, because it can designate both these senses using the same word wood. (The notion of colexification is not whether a language must express two senses identically, but whether it can.)

Symmetrically, languages can be compared in the way they draw contrasts between close meanings. I propose to call this dislexification, the polar opposite of colexification:^[3]

Thus, Spanish dislexifies the two senses ‘wood as material’ (madera) and ‘firewood’ (leña), as it lacks any term that would encompass both meanings. As for English, it could also express that contrast, using a specific compound like firewood: but crucially it doesn’t have to, since it can resort to the vague term wood in all cases (e. g. Nights are getting cold, let’s go cut some wood). In sum, the two senses in question are dislexified in Spanish, but colexified in English.

The notion of dislexification, just like its counterpart colexification, entails no judgment on what is to be considered expected or normal. When comparing a given lexical domain across two languages, selecting the right descriptor here is essentially a matter of perspective, and is ultimately arbitrary. Likewise, the two terms are to be understood as purely descriptive, and synchronic: co- and dis-lexification do not inherently refer to a change in meaning – contrary to “lexical merger” or “lexical split”, which will be introduced later on in this study.

Figure 1

Two senses s1 and s2 may be colexified (language A) or dislexified (language B).

Together, colexification and dislexification define a structural property which can be named patterns of lexification;^[4] see (1) above.

Lexification patterns can be compared across languages, and shown visually on a simple map such as Figure 1. A semantic domain is projected onto a two-dimensional space, in such a way that its component senses s1 and s2 occupy distinct locations on a two-dimensional plane. Using sets, senses s1 and s2 are here shown to be colexified in language A, but dislexified in language B.

While Figure 1 compares separate languages, nothing prevents us from using the same principle for comparing two historical stages of the same language. A language may dislexify two senses at a certain point in history, but its descendant may colexify them, or vice versa. This will be central to our discussions on historical lexicology.

3.1 Structural stability

If a set of modern languages descend from a single ancestral protolanguage, part of their lexical structures will just be inherited from that ancestor. These structures can prove stable over time.

To take a simple example, we know that Latin distinguished between bracchium ‘arm’ and manus ‘hand’ – contrary to 37 percent of the world’s languages, which colexify these two body parts (Brown 2013).^[5] This pattern of dislexification {arm ≠ hand} has been preserved intact in modern Romance languages, all of which contrast ‘arm’ vs. ‘hand’ (Table 2). In this specific case, the inheritance of lexical structures came along with the preservation of the inherited etyma themselves: such a case can just be seen as an unremarkable instance of linguistic conservation.

Interestingly, lexical structures may remain stable even when the word itself changes its form due to lexical replacement. Consider, in Table 3, the terms for ‘leg’ and ‘foot’ in Romance languages. The word crūs of Classical Latin was lost in modern Romance languages, which have replaced it with new lexical material – either from Lat. perna ‘ham’, or from Late Lat. gamba < Gr. κάμπη ‘flexure’. And yet, in doing so, they never lost the original structural pattern whereby ‘leg’ is dislexified from ‘foot’.

Table 3

‘Leg’ and ‘foot’ in Romance languages.

The only exception here is Romanian, where picior ‘foot’ can also be used for ‘leg’.^[6] In modern Romanian, the inherited form gambă is now mostly used for the lower leg (calf, shank), while nowadays the default term for the lower limb as a whole is picior. Examples such as picioare strâmbe ‘crooked legs’, or un picior mai scurt decât celălalt ‘one leg longer than the other’, clearly refer to the leg rather than just the foot.

One can thus say that Romanian has innovated a colexification {leg = foot} which was not present in its ancestor. This structural innovation reflects areal influence, as Romanian calqued the colexification {leg = foot} characteristic of its neighbours: Hungarian láb, Bulgarian крак, Russian нога, Greek πόδι. Romanian is here the only exception to the general observation that Romance languages generally show structural stability in the domain of body parts for upper and lower limbs.

Many other examples could be found, in the world’s languages, of inherited lexical structures being preserved over time. While this preservation sometimes goes along with the conservation of the original lexemes, we saw that it can also occur even in the case of lexical replacement – as though the structural properties of the lexicon were somehow independent of the specific words that embody them (François 2010; 2011a: 226). And indeed, our final discussion [Section 5] will examine cases of “language-internal relexification”, where lexical structures remain stable in spite of lexical forms being renewed.

The structuralist Coseriu (1964) made the same point: “a semantic structure can be maintained even when the word forms have been replaced”.^[7] Such observations prompted him to promote what I would call semantic lexicology (“lexicologie du contenu”, Coseriu 1964: 164) as a science of the lexicon dedicated to semantic structures per se, independent of changes in their phonological material.

3.2 Structural innovations: Mergers and splits

Languages commonly go through structural innovation. A given lexification pattern can be lost or redesigned over time. Ultimately, any semantic change in the lexicon comes with a change – whether minor or major, temporary or permanent – in the structural organisation of its semantic domain.

I will illustrate the notion of structural innovation by observing two lexical domains in the Oceanic languages of northern Vanuatu: body parts and kinship terms.

3.2.1 Body part terms in northern Vanuatu

Contrary to Romance (Table 2 above), Oceanic languages generally colexify ‘hand’ with ‘arm’. This pattern can be reconstructed for the Proto-Oceanic ancestor with a form *lima ‘arm, hand; five’ (Osmond and Ross 2016: 160).

Within our northern Vanuatu sample (Table 4), all languages have preserved the colexification pattern {arm = hand}, whether they reflect the etymon *lima (white cells) or an innovative form (coloured cells).^[8] This is another case of structural conservation in spite of lexical change (Section 3.1).

Table 4

Some upper body parts in a sample of northern Vanuatu languages.

Nine languages in the central area (five of which are represented in Table 4) went through lexical replacement. They lost their reflexes of *lima, and replaced them with *bani, whose original meaning was ‘[bird] wing’ (Osmond and Ross 2016: 162). For the innovative languages shown in Table 4, we can reconstruct a process of semantic extension that gave rise to a new pattern of colexification:

1. stage 1: *bani {wing} ≠ *lima {arm = hand}

2. stage 2: *bani {wing = arm = hand}

This structural change can be visualized on a semantic map (Figure 2), showing the two stages successively. Compared to Proto-Oceanic, or to the eight conservative languages that have preserved Stage 1, the nine languages that went through the lexical merger illustrated in Stage 2 have modified their lexification structures.

Figure 2

Terms for upper limbs in northern Vanuatu: a lexical merger.

I propose to describe this type of change as a lexical merger:

(6)

A lexical merger is a historical process of semantic change whereby two sets of senses that used to be dislexified (encoded by different lexemes) end up being colexified (encoded by the same lexeme).

Coseriu (1964) discusses similar cases of “fusion fonctionnelle”, which he defines as the loss of a contrast^[9] – drawing explicit analogies with structural mergers in phonology or morphology. As an example, he cites the Latin kin terms patruus ‘father’s brother’ vs. avunculus ‘mother’s brother’ – a contrast which modern Romance languages later merged into a single category ‘uncle’ (Fr. oncle, Ital. zio, Cast. tío…).

The opposite of a lexical merger is a lexical split. Table 5 lists the same sample of ten languages, this time looking at the lexification of ‘leg’ and ‘foot’.

Table 5

Some lower body parts in a sample of northern Vanuatu languages.

The general pattern in the region is the colexification {leg=foot} – a pattern itself parallel with {arm=hand} described above. Setting aside sound change, all modern forms point to the same etymon ptb *raŋo, which can safely be reconstructed as having both meanings ‘leg, foot’ (François 2005:​ 498). But one language, Lo-Toga, stands out in the region: it is the only one that has a special word /təple/ for ‘foot’, distinct from ‘leg’ /rəŋo/. Clearly, this word results from a local innovation, which took place only in Lo-Toga and nowhere else.

The latter type of change (structurally the opposite of a lexical merger) can be called a lexical split:

(7)

A lexical split is a historical process of semantic change whereby two senses that used to be colexified (encoded by the same lexeme) end up being dislexified (encoded by different lexemes).

Coseriu (1964) discusses similar structural processes, and defines them as the creation of a new emic contrast (“l’apparition d’un trait distinctif nouveau”, 1964: 174). He illustrates this with Lat. avis ‘bird’, a generic category which Portuguese divided in two – with pássaro ‘small bird’ vs. ave ‘(non-small) bird’.

3.2.2 Words for siblings in northern Vanuatu

The previous section defined the diachronic notions of lexical merger and lexical split, using simple examples taken from the domain of body parts in Vanuatu languages.

While remaining in Vanuatu, we can now turn to a more complex case of structural reshaping in the domain of kinship terms – namely, terms for siblings.

English divides this lexical domain based on the target’s gender, contrasting sister vs. brother; but other languages use different criteria to organise the different types of siblings (Evans 2010).

A system with three categories of siblings

Let us start by observing how the field is structured synchronically in one language of the Banks islands, e. g., Lakon [lkn] from the island of Gaua. In this language, the sibling domain is organised according to two semantic features:

1. the relative sex of the target compared with that of its anchor (‘ego’):

   1. tata ‘opposite-sex sibling’ is the term used

      for a male’s sister [♂Z]

      or for a female’s brother [♀B]^[10]

   2. tua and tahi (see below) both refer to a ‘same-sex sibling’,

      whether a male’s brother [♂B]

      or a female’s sister [♀Z]

2. [for same-sex siblings] the relative age of the target compared with its anchor:

   1. tua ‘older same-sex sibling’ is the term used

      for a male’s elder brother [♂eB]

      or a female’s elder sister [♀eZ]

   2. tahi ‘younger same-sex sibling’ is the term used

      for a male’s younger brother [♂yB]

      or a female’s younger sister [♀yZ]

Lakon and English thus differ drastically in the way they categorize sibling relations. Their differences can be shown on a semantic map whose background is an etic grid listing a number of specific kin relations – e. g., ⟨♀yB⟩ ‘a female’s younger brother’. English and Lakon group these atomic senses according to their respective emic categories – see Figure 3.

Figure 3

Terms for siblings in the Lakon language of Gaua island (Banks group).

The ternary structure at the source of modern Torres–Banks systems

The ternary pattern of Lakon is also found in several other languages of the Torres–Banks area, and even in several Oceanic languages spoken outside that area. This semantic organisation can in fact be safely reconstructed for PTB (‘Proto-Torres–Banks’), the region’s shared linguistic ancestor:

1. ptb *tuatua ‘opposite-sex sibling’ (François 2005: 500)

2. ptb *tuaɣa ‘older same-sex sibling’ (Clark 2009: 203)

   < POc *tuqaka (Pawley and Ross 2006: 53)

3. ptb *tasi ‘younger same-sex sibling’

   < POc *taci (Pawley and Ross 2006: 53).

Like Lakon, several modern languages of northern Vanuatu have preserved the protoform of their common ancestor (albeit with sound change) as well as its semantic organisation. Forms are given here for a 1sg possessor (suffix *-gu > -k):

(8)

*tuatua (-gu)  ‘(my) opposite-sex sibling’ >

Hiw tʉtvɔ-k; Lo-Toga səse-k; Löyöp ʧøʧø-k; Mwotlap tɪtɛ-k; Lemerig

ʔœwœ-k; Vera’a ʔɔwɔ-k; Vurës tytyœ-k; Mwesen tutuɔ-k; Mota tutua-k; Dorig tutaː-k; Koro tɛtɛ͡a-k; Olrat tati-k; Lakon tata-k; Mwerlap tɞtɞ-k.

(9)

*tuaɣa (-gu)  ‘(my) older same-sex sibling’ >

Vera’a ʔuwa-k; Vurës toɣœ-k;^[11] Mota tuaɣa-k; Nume tua-k; Olrat tua-k; Lakon tua-k; Mwerlap tuɔ-k.

(10)

*tasi (-gu)  ‘(my) younger same-sex sibling’ >

Vera’a ʔisi-k; Vurës tɪsi-k; Mota tasi-k; Nume tisi-k; Olrat tisi-k; Lakon tahi-k; Mwerlap tɛsi-k.

All the forms cited in (8)–(10) have preserved the ternary system of their ancestor – the one illustrated in Figure 3 for Lakon. But the rest of the Torres–Banks languages have gone through a couple of innovations, which I will present now.

Semantic change and restructuring

The first innovation is the generalisation of *tasi to refer to any ‘same-sex sibling’ regardless of relative age. This is a new example of a lexical merger (Section 3.2.1). As a corollary, the root *tuaɣa was eliminated:

(11)

*tasi (-gu)  * ‘(my) younger same-sex sibling’ > ‘(my) same-sex sibling’

Mwotlap ithi-k; Lemerig ʔisi-k; Mwesen tisi-k; Dorig tsi-k; Koro tsi-k.

The five languages listed in (11) have lost the feature of relative age as a structuring principle in their sibling domain. The only relevant criterion is now relative sex: these languages contrast only two terms for siblings, namely *tuatua ‘opposite-sex sibling’ vs. *tasi ‘same-sex sibling’. This innovation likely reflects a cognitive pressure for symmetry in the system.

Finally, the four northernmost languages in the area show evidence of the same change (loss of *tuaɣa), but have added to it a further innovation. In addition to the feature relative sex which provides the sibling system with its overall structuring principle, these languages split the subdomain “same-sex sibling” according to the referents’ absolute sex. This lexical split (Section 3.2.1) was made possible by allowing a new lexeme into the kinship domain. This root *tagele (whose original meaning must have been ‘other half, counterpart’)^[12] became used specifically for the female instances of ‘same-sex sibling’ – that is, a woman’s sister:

(12)

*tagele (-gu)  * ‘(my) counterpart’

> ‘female same-sex sibling = (woman’s) sister’

Hiw takje-k; Lo Toga takəle-k; Lehali takle-k; Löyöp takle-k.

As a corollary, these four languages restricted their root *tasi to male referents, i. e., a man’s brother:

(13)

*tasi (-gu)  * ‘same-sex sibling’

> ‘male same-sex sibling = (man’s) brother’

[Hiw ŋʷati-k];^[13] Lo Toga təɣi-k; Lehali tihi-k; Löyöp ʧisi-k.

In sum, the root *tasi (POc *taci) has been assigned three different meanings in the course of its evolution – from its original sense ‘younger same-sex sibling’ to a broader meaning (11) ‘same sex sibling [= woman’s sister/​man’s brother]’, and then to an innovative gendered gloss (13) ‘man’s brother’.

Synthesis: A diachronic map

This complex history of sibling terms in northern Vanuatu can be synthesized in Figure 4 – a diachronic map showing the three stages of lexical change. The first innovation was a lexical merger; the second one was a lexical split.

Figure 4

Terms for siblings in northern Vanuatu: three stages of evolution.

Stage 1, the initial organisation of meaning that can be reconstructed for Proto-Torres–Banks, is preserved in seven languages (including Lakon, shown in Figure 3); the two other stages illustrate successive innovations in a subset of the Torres–Banks group. Among its 17 members, ten languages went through the lexical merger of Stage 2, losing the dimension of relative age. Among these, four languages later went through the lexical split of Stage 3, as they divided the ‘same-sex sibling’ category according to the referent’s absolute sex.

While the history of the sibling system in the region can appear complex to describe, it becomes much easier to conceive through the use of a map like Figure 4. Over time, the semantic range of each root has redesigned its contour on the map.

4.1 Lexical replacement and the markedness shift

Based on the examples and analyses above, we can propose an abstract representation of lexical change. Let us represent a lexical domain Δ as a set of attested senses numbered ⟨A B C … H I⟩. These meanings can be spread out on a two-dimensional semantic map. Taking such a map as its background, Figure 5 illustrates three successive stages of lexical change.

Figure 5

The three stages of lexical replacement.

In an initial Stage 1, a form f1 colexifies together the three senses ⟨A B C⟩, whereas a form f2 encodes two neighbouring yet distinct meanings ⟨D E⟩. Then, through a process of semantic extension – whether due to external influence (language contact) or to internal change – f2 starts encroaching upon f1’s territory, as it becomes capable of also lexifying the sense C. Plotted on a map, such a change translates visually through the reshaping of each line – i. e. the blob encompassing all the senses of a polysemous word.

Stage 2 represents a phase of lexical competition between the words f1 and f2 when expressing the sense C. (By way of illustration, consider how some varieties of English show variation between film and movie, between tap and faucet, between if it were and if it was.) The variation can be a matter of stylistic register, where typically f1 goes from being the norm to becoming marked and archaic, while f2, which started as a marked synonym for sense C, ends up becoming the normal, unmarked form to express it. This reversal process is called markedness shift (Dik 1989: 44); it can be captured by the formula in (14), where ‘S’ refers to a specific sense. (The forms in bold constitute the default terms for a given meaning, while brackets flag the synonyms that are stylistically or pragmatically “marked”.)

If the new usage takes root and spreads through the speech community (via social diffusion), eventually the old term becomes obsolete, and ends up being replaced by the new term. The final result, shown here as Stage 3, is one where sense C is not lexified by f1 anymore, but exclusively by f2. As far as this sense is concerned, the language has gone through the whole process of lexical replacement.

As Figure 5 shows, the process of lexical replacement affecting the sense C can technically be broken into two subphases:

Lexical replacement, in principle, always consists of a lexical merger (resulting in lexical competition) followed by a lexical split (resolving the lexical competition).

4.2 Lexical competition as a slow, emergent process

In the reality of linguistic use, the sort of change illustrated here would take the form of a continuous process of variation between two strategies f1 and f2 to express a given sense (see Traugott and Dasher 2001: 11–12; Enfield 2003: 29): variation over time, variation across speakers, and even intra-speaker.

The lexical competition between variants can span over several generations, sometimes centuries. Analyses based on diachronic corpora, when they are available, show that the process of language change manifests itself as a slow evolution in the distribution of the two variants, such that f1 remains the preferred option for a long period, while f2 increases its prevalence over time. Rather than being a sudden shift, the emergence of f2 is incremental; and what is transmitted from one generation to the next is really a particular statistical distribution of f1 vs. f2, together with a sense of the directionality of change – as f1 decreases towards obsolescence while f2 grows in frequency, and evolves towards becoming the new norm. Rather than forming a linear, steady increase, the trajectory of the distribution along the time axis typically takes the form of an “S curve” (Chambers 2002: 361; Blythe and Croft 2012). The competition between the two variants begins slowly, then accelerates for a short while, before it finally fades off during another long period. This trajectory, drawn along a time axis, is visually reminiscent of the shape of the letter S or (hence its name ‘S curve’, or ‘sigmoid curve’).^[15]

This point should be kept in mind whenever a linguistic innovation (whether phonological, syntactic or lexical) is represented by an abstract formula, of the type {f1 → f2} “f1 was replaced by f2”. Just like geological activity, lexical tectonics must be understood as a slow process, which can span over many generations and still be imperceptibly active. Competition in the lexicon involves long periods when the two variants coexist in discourse. During that transitional period, the perception of norms and markedness fluctuates among speakers, before it finally stabilises.

4.3 An example of lexical replacement

To take a classic example, Late Latin progressively replaced its inherited term caput ‘head’ with another noun testa, originally ‘earthen pot’ (cf. Blank and Koch 1999). The four stages followed by this case of lexical replacement – in line with the formula in (14) – can be shown in Table 6.

Table 6

When markedness shift drives lexical replacement: words for ‘head’ in Late Latin / Romance.

The first innovation (Stage 1→2a) was to use the noun ‘pot’ as a figurative, slang word for ‘head’, in competition with the inherited form caput – in a way similar to early 20th c. French slang carafe ‘jug’ used sometimes for ‘head’. Eventually, testa lost its jocular connotations, and ended up as the standard term for this body part (Stage 2b), as evidenced by Italian testa, and French tête. The older term caput (> Italian capo, O.Fr. chief) resisted for some time, but eventually became the marked term in the pair.

Standard Italian still belongs to Stage 2b. While testa has become the default name of the body part, capo still survives as a marked synonym – i. e. archaic, regional or medical – and in idioms (e. g., mal di capo ‘headache’). Yet the word’s main modern meaning is ‘leader, chief’ (e. g., capo dello stato ‘head of State’).

French has gone one step further. In Classical French, chef was still used as an archaic synonym of teste, until it disappeared in this sense. In modern French, if one puts aside a couple of vestigial compounds (e. g., couvre-chef ‘hat’), the only living meaning of chef is ‘leader, chief’: the language has reached Stage 3, and the lexical split is now complete.

If a typologist wanted to list the languages that colexify the meanings ‘head’ and ‘leader’, Standard Italian would still qualify (with capo), but modern French would not: le chef is no longer used to refer to the head, and la tête can hardly refer to a chief.^[16]

One could summarise the whole process by stating that French “replaced” its noun caput (Stage 1) with an innovative form testa (Stage 3). While not untrue, this simplified formula encapsulates what was really a gradual sequence involving lexical competition and markedness shift between polysemous terms (see Sweetser 1990: 9; Evans and Wilkins 2000: 549), spanning – in this case – over almost two millennia. Rather than a leap from Stage 1 to Stage 3, the key turning point in this evolution was really the shift from Stage 2a to Stage 2b, and the reversal of markedness.

4.4 Subtypes of lexical replacement

The process of lexical replacement, which we just examined from the angle of a single sense, can be revisited in the broader perspective of its impact upon the system’s lexical structures. As a language goes through lexical replacement, a corollary is a new configuration of its lexical patterns.

As far as the time axis is concerned, we just saw that lexical replacement can be complete (cf. *chef → tête in French for ‘head’) or it can take the form of an ongoing competition (cf. Italian capo ∼ testa). This contrast can be visually represented on a map by the presence or absence of an overlap: compare Stage 2 and Stage 3 in Figure 5.

On a different dimension – namely, the distribution of meaning on the semantic map – lexical replacement can be local or total:

1. Local replacement: f2 replaces f1 for only some of its senses.

2. Total replacement: f2 takes over f1 in all of its senses, resulting in the elimination of f1 from the lexicon.

Conversely, we saw in Section 3.2.1 (Figure 2) that POc *lima ‘arm, hand’ was replaced by *bani ‘wing’ for all its senses; this resulted in the total replacement – or relexification – of *lima by *bani, and the loss of *lima.

4.5 A typology of innovations in patterns of lexification

Let us recapitulate the different types of innovation that can affect lexical structures.

In line with Figure 5, let us posit two word-forms f1 and f2, and a segment of the semantic space defined by three senses A B C. In stage 1, senses {A B} are lexified by the form f1, and {C} by f2. Assuming these words undergo semantic change, we can characterise Stage 2 by how this initial pattern of lexification is modified over time. Table 7 defines five main types of diachronic scenarios: lexical competition, lexical split, lexical merger, lexical shift, and relexification.

In Table 7, the sign ‘—’ means that the form in question no longer exists in the subdomain defined by senses A B C: either it moves on to take up a separate meaning D, or it disappears altogether from the language.

Here is a short description for each innovation type:

1. Lexical competition:

   Sense B is initially lexified by a word f1; but another word f2, by semantic extension, acquires that same sense B. The two words f1 and f2 compete for some time with respect to lexifying B.

2. Lexical split:

   A word f1 initially colexifies two senses {A B}; but another word f2 acquires the sense B, and after a period of lexical competition, f2 ends up replacing f1 for sense B. As a result, the initial pattern of colexification {A B} is split apart.

3. Lexical merger:

   Senses A B C used to be dislexified across two separate forms f1 and f2. Following lexical replacement of f2 by f1 (for sense C), these senses become colexified.

4. Lexical shift:

   A word f1 loses some senses (e. g., A) and gains others (e. g., C) which used to be lexified with another word f2. Such a lexical shift combines a split ({A=B} → {A≠B}) and a merger ({B≠C} → {B=C}).

5. Relexification:

   Senses A B used to be lexified by a word f1. Following wholesale lexical replacement, they become lexified by another word f2, while f1 exits the domain.

5.1 Moon, month

The Proto-Oceanic (POc) etymon *pulan colexified {moon=month}, and most languages in northern Vanuatu have preserved that colexification.^[17]

The language Lakon keeps its reflex of *pulan, namely /vʊl/, for the sense ‘month’; yet it now lexifies the moon using an innovative form /siŋaː/, from ptb *siŋaRi ‘shine’ (François 2011b: 186). This is a neat case of a lexical split – see Table 8. In this instance, Lakon shows structural divergence compared to its linguistic neighbours.

Table 8

‘Moon’ and ‘month’ in a sample of northern Vanuatu languages.

The lexical innovation of Lakon is also attested in the other languages of Gaua, illustrated here by Dorig and Nume. Yet interestingly, these languages went one step further. They generalised the innovative root *siŋaRi for the two meanings ‘moon’ and ‘month’, thereby eliminating *pulan altogether from the language: e. g. Dorig now has /sŋar/ ‘moon, month’. In other terms, after diverging from the other Banks languages in a way similar to Lakon, the two languages Nume and Dorig eventually “reconverged” towards the colexification pattern {moon=​month} that used to be associated with *pulan.

The result of this two-step process (divergence + re-convergence) was wholesale relexification. The polysemy {moon = month}, which had once been split apart, was eventually retrieved, but associated with a different etymon – see Figure 6.

Figure 6

Language-internal relexification via semantic realignment. In Gaua languages, the colexified pair {moon = month} underwent a lexical split; then both senses were reunited through a process of contact-induced realignment.

But how can we account for such a process of semantic realignment, whereby lexicons would first lose the lexical structures of their ancestors, and then mysteriously retrieve them? Why didn’t Nume and Dorig just keep their new dislexification {moon ≠ ​month} just like Lakon did? I believe the process must be explained as an effect of areal pressure. In a region characterised by widespread multilingualism (François 2012), the islanders of Gaua remain familiar with the lexical structures that are common in the Banks archipelago. And indeed, a general phenomenon observed in north Vanuatu is one of perpetual alignment of structures (whether syntactic or semantic), regardless of the form of words (François 2011a).

In that part of the world, when a language goes through a local case of lexical replacement, the resulting structural disruption usually does not last long. Once a newly introduced word has started an incursion into a given semantic domain, it is just a matter of time before it becomes harnessed into the region’s dominant lexification patterns. As if by a sleight-of-hand, the end result is a complete renewal of word forms, and yet a perfect preservation of inherited polysemies. This paradox (replacement of forms, preservation of ancestral lexical structures) is what I call language-internal relexification.

An examination of the lexicons of north Vanuatu languages shows that language-internal relexification is there a pervasive phenomenon (François 2010; 2011a:​ 226–8). The typical scenario is for lexical replacement [stage 2] to be followed by semantic realignment [stage 3], just like we saw for Dorig in Figure 6. Admittedly, lexical innovation does sometimes bring about structural disruption (as in the case of Lakon here) yet that situation happens much less often. When the lexical patterns of a language are at odds with its neighbours, the units of one system become less readily translatable from one language to another; such cross-linguistic discrepancies tend to become ironed out sooner or later, so as to facilitate the handling of several lexicons in the minds of multilingual speakers.^[18]

5.2 Village, island, country

A similar example is provided by the words for ‘village’. The Proto-Oceanic etymon *panua ‘inhabited territory’ (Pawley 2005) is well reflected in Torres–Banks languages, where it commonly colexifies {village = island = country}.

In Hiw, the regular reflex of *panua, namely /vənjɵ/, had to concede some ground to an innovative compound /mətevənjɵ/ (etym. ‘island spot’) for the sole meaning ‘village’. As a result, Hiw now splits the semantic array of *panua across two lexical items: /vənjɵ/ ‘inhabited territory, country, island, (*village)’ vs. /mətevənjɵ/ ‘village’. This is a case of lexical split.

Further south, the language Dorig has also innovated a new word for ‘village’, from a root *varea ‘outside, public space’ (cf. Clark 2009: 218). This innovative noun must have entered the territory of *panua through the sense ‘open space in the middle of the village’ → ‘village’. In principle, this could have resulted in some form of lexical split (like in Hiw), with a new contrast (*varea ‘village’ vs. *panua ‘island, country’).

Yet what happened here, once again, was a process of semantic realignment. The lexification pattern {village = island = country} is so entrenched in the region that the new root *varea ended up covering the whole territory of former *panua (Figure 7). Modern Dorig has lost all reflexes of *panua, and replaced them with a single noun /vrɛ/ ‘village; island; country’. This is another perfect case of language-internal relexification.

Figure 7

Language-internal relexification via semantic realignment. In Dorig, the root *varea entered the territory of *panua via the sense ‘village’, but eventually replaced *panua in its whole polysemy ‘village, island, country’.

5.3 Sacred, forbidden, haunted, cemetery

The Proto-Oceanic etymon *tabu – the origin of Eng. taboo – was polysemous. It included such senses as ‘sacred, holy; haunted by spiritual forces; forbidden, secret; ban, proscription; reserved to initiated men’ (François 2022).

In Mwotlap, the semantic territory of *tabu has shrunk considerably. This root only survives in the noun /nɛ.tɛkp͡ʷ/ ‘graveyard, cemetery’ (François 2013: ​225). While arguably related to the sense ‘haunted place’, that meaning ‘graveyard’ is itself innovative, as it did not even belong to the root’s meanings in Proto-Oceanic.

As for the root’s original polysemy, it has now been split across two roots. An etymon *salaɣoro (literally ‘forbidden path’) has replaced *tabu for all the senses reflecting a social proscription {‘forbidden, secret; ban, proscription; secret society, reserved to initiated men’}. As for the spiritual meanings of *tabu, which pertained to the notion of holiness or divine presence, they were relexified with a separate root *roŋo (‘sacred, holy, numinous’ – originally ‘quiet, silent’).

In this case, there was no case of wholesale relexification, nor preservation of earlier lexical structures. On the contrary, the history of *tabu in Mwotlap reflects a rich history of semantic shifts, splits, and relexifications. The root *tabu itself has now migrated to a new location of the semantic space. As for the initial polysemy it used to have in Proto-Oceanic, it has been split apart across two other roots (*salaɣoro and *roŋo), resulting in quite different patterns of lexification between the modern language and its POc ancestor.^[19]

5.4 Up, uphill, inland, southeast

In the domain of space directionals (François 2004; 2015), the inherited word *sake ‘up’ reconstructs (for Proto Torres–Banks) as encoding a number of different directions: {⟨1⟩ [vertical] up; ⟨2⟩ [on hilly islands] uphill; ⟨3⟩ [on flat islands] inland; ⟨4⟩ [on sea] landwards’; ⟨5⟩ upwind > [inter island] southeast; ⟨6⟩ [on land] parallel to the shore, towards south}.

Now, the language Hiw has split the lexical domain of *sake in two. An innovative directional *vene (‘climb → [go] up’) has evicted *sake – but only for the meanings most clearly linked to the vertical dimension, namely ⟨1⟩–⟨2⟩. As for senses ⟨3⟩–⟨4⟩–⟨5⟩–⟨6⟩, they are now lexified by a vestigial form /aɣ/ (François 2015: 184–185). Even though it reflects *sake ‘[go] up’, this form has now lost any semantic connection with verticality. The historical outcome is a neat lexical split, represented in Figure 8.

Figure 8

A lexical split. In Hiw, the original polysemy of *sake ‘up+’ was split in two. Some senses are now lexified by *sake, others by *vene.

The change resulted in the dislexification between ‘inland’ (on gentle slopes) and ‘uphill’ (on steeper ground) – a rare pattern among the geocentric systems of north Vanuatu (François 2015: 151, 176). In this particular case, there was no process of semantic realignment. And indeed, the lexification patterns of space directionals in modern Hiw are quite original compared to its neighbours.

5.5 Give, take, get, understand

One last example concerns the polysemy of ‘take’, and the wholesale relexification that took place in the language Lehali.

The Proto-Oceanic root *alap-i ‘get, take’ (Ross 2016: 426–427) is reflected in most of the Banks languages, via a root *lavi or *lai. That verb means ‘take, carry’; combined with directionals, it encodes meanings such as ‘bring’ (take hither), ‘take away’, ‘raise’, ‘lower’, ‘remove’, and so on. Besides these basic meanings, reflexes of *lavi can also mean ‘harvest’, ‘steal’, ‘abduct’, ‘adopt [a child]’, ‘marry [a woman]’. The same verb can also take up cognitive meanings, such as ‘commit s.th. to memory’ or ‘understand’.^[20]

Finally, the combination of *lavi ‘take’ with a dative argument or a participant-oriented directional (François 2003: 413, 2015: 146) forms the equivalent of ‘give’: ‘he gave it to me’ literally reads he took it hither. In other words, Northern Vanuatu languages are characterised by the colexification {take=​give}, a pattern that is otherwise found mostly in languages of New Guinea (Gil 2017: 80).

Now one language, Lehali, has lost *lavi, and instead uses a non-related form hö /hɒ/. This verb hö is cognate with Mwotlap /hɔɰ/ ‘hold out o.’s arm; hand out, give’, Vurës /sʊɰ/ ‘give generously’, Mota /soɣo/ ‘bring, give’. We can reconstruct a protoform for Proto Torres–Banks, *soɣo ‘hand out, give’.

Within the whole semantic set of *lavi, the sense ‘give’ was probably the entry point whence *soɣo launched its lexical intrusion, and where the competition must have begun. That sense ‘give’ was then to act as the “Trojan horse” in the relexification of *lavi by *soɣo – see Figure 9.

Figure 9

Language-internal relexification via semantic realignment. In Lehali, the polysemous root *lavi ‘take+’ was entirely replaced by *soɣo (etym. ‘hand out’).

Figure 9 illustrates a full process of internal relexification. The word hö (< *soɣo) has now replaced *lavi in all of its senses and combinations: compare the semantic set of *soɣo in Stage 4 with that of *lavi in Stage 1. Besides the sense ‘give’, hö means ‘take, get’ in modern Lehali; it forms compounds meaning ‘bring’ (hö ma), ‘remove’ (hö yak), etc. The same verb can mean ‘steal’, ‘adopt [a child]’, ‘marry [a woman]’, ‘understand’…:

Through the process of language-internal relexification, a root whose original meaning was ‘hold out o.’s arm, hand out’ has reshaped its semantic array so as to include such diverse senses as ‘carry’, ‘marry’, or ‘understand’.

In doing so, it appears as though the root *soɣo had cast itself into the mould of *lavi – like a lake that borrows its shape from the crater of an extinct volcano.

The structural properties of Lehali hö are now perfectly parallel with those of Mwotlap lep, in spite of their distinct etymology. And indeed, bilingual speakers can sometimes be heard commenting that “whenever Mwotlap says lep [lɛp], Lehali says hö [hɒ]”.

* * *

In all the cases we reviewed, sequential maps proved helpful in visualising the process of lexical change as a whole.

To take a geological analogy, Figure 9 could be described as a collision between two tectonic plates: as the *soɣo plate expanded its semantic range, it must have gone through a period of friction at the boundary, in the form of lexical competition (Section 4.1) between *soɣo and *lavi for the meaning ‘give’ (stage 2). Eventually, the boundary friction was resolved, so to speak, by the “subduction” of *lavi, giving way to *soɣo for that particular zone. As a result of these tectonic movements, the two plates *soɣo and *lavi changed their shapes, and redesigned the local semantic landscape.