Constant Entailments suggests a taxonomy of interpretive language change that involves pragmatic components. We expect that in language change, shifts are possible between truth conditional content, presuppositions (PSPs), and implicatures or inferences. We will consider the following types of shifts in the next subsections:

This systematics invites a series of research questions. We may ask, for example, which of the types of change above are typical. Can we find putative examples/classes of expressions for types of change that have not been reported or are understudied? We can investigate different types of implicatures in diachronic processes (generalized vs. particularized, scalar etc.); and so on. The case studies below begin this project.

The attentive reader will notice that (23) does not exhaust the logical possibilities. There is of course also the possibility of interpretive language change within the realm of truth conditional content. Moreover, a meaning component could change its status from PSP to implicature or vice versa, and similarly for potential further kinds of meaning component. We will not discuss putative examples of such further types of interpretive change. This is not to suggest, of course, that such changes do not exist (see for example Eckardt (2001; 2006) on the diachronic development of the German focus particle selbst ‘even’, Deo (2015b) on change in the domain of aspect, Gergel (2010) on the development of rather, Beck (forthcoming) on universal quantification, and Eckardt (2011), Deo (2014; 2015a) for overviews). Our goal in this paper is not to be exhaustive; rather, we want to introduce a set of case studies that support the pragmatic version of Constant Entailments (17) and the resulting systematics based on standard semantic and pragmatic analyses of the phenomena they involve.

Before we get started with concrete examples of semantic change in search of answers to the systematic questions we raise, let’s clarify what a compositional semantic and pragmatic analysis of language change comprises. The general task of formal semantic analysis of interpretive language change is summarized in (24), after Beck (forthcoming). For any given instance of interpretive change, we ideally want to answer Q1 Source and goal and Q2 Steps of change (remember that CE is the critical expression, the expression whose meaning undergoes change).

The strategy shown in (24) is what we will apply to the case studies below. We will present four case studies which we consider plausible instantiations of the different types of interpretive change in (23). In each case study, we rely on established semantic analyses of the phenomena under investigation. In subsection 4.2 we comment on interpretive change from implicature to truth conditional content. We use connectives, in particular, to provide a specific analysis in terms of Constant Entailments. In subsection 4.3 we discuss the development of size noun-based quantifiers as a potential candidate for the opposite development, i.e. from truth conditional to implicated meaning. Subsection 4.4 examines change from assertion to PSP in the rise of a focus particle, and subsection 4.5 investigates an interrogative complementizer as an instance of change in the reverse direction, from PSP to assertion. At the end of section 4, the taxonomy in (23) will have gained both concreteness and plausibility.

Among the four logical possibilities presented in subsection 4.1, only meaning changes driven by implicatures have been discussed extensively as one major type of interpretative change. Implicature-based accounts have been proposed for interpretative changes in many different types of expressions. In fact, it was the founder of implicature theory himself, Paul Grice, who cursorily noted that “it may not be impossible for what starts life, so to speak, as a conversational implicature to become conventionalized” (Grice 1975: 58). While Grice did not expand on this putative diachronic dimension of implicature theory, Levinson in his early work went as far as to propose that

A phenomenon that has been cited again and again as a typical and wide-spread instance of the conventionalization of an implicature are meaning changes in connectives. One major subtype is the development of causal connectives out of temporal ones. The World Lexicon of Grammaticalization lists the following cases – many more have been discussed cross-linguistically:

The frequency of this particular meaning change is generally attributed to the cum/post hoc ergo propter hoc fallacy (‘with/after this, therefore because of this.’). The basic idea is that connectives indicating (partial) temporal overlap or temporal posteriority between two sentences in many contexts give rise to causal conversational implicature.

As mentioned in the World Lexicon of Grammaticalization, this type of meaning change can be observed in the diachronic development of the German causal subjunction weil, which ultimately goes back to Old High German dia wîla so, an adverbially used accusative DP dia wîla ‘the-ACC while-ACC’ modified by a relative clause introduced by the particle so. In Early New High German, the nominal expression was eventually reanalyzed as a complementizer (C°) (cf. Eberhardt & Axel-Tober 2023 for details). In Early New High German, the complementizer weyl was still used with its original temporal meaning ‘at the time/lit. while that; during’. The purely temporal reading is present in example (27) from Luthers Taufbüchlein:

In the often-cited example (28) from the Grimms’ dictionary a causal interpretation can be inferred: ‘Because the master craftsman had left the workshop, the apprentice worked with less care’.

Such a causal reading frequently arises in sentences with temporal connectives, see for instance the modern German and English examples in (28’) and (28”).3 But as the cancelability test shows, the causal meaning is often merely a conversational implicature.4

The sentence with the goal meaning Weil der Meister die Werkstatt verlassen hatte, arbeitete der Lehrling nachlässiger ‘because the master left the workshop, the apprentice worked with less care’ clearly differs in truth conditions from its putative diachronic source meaning ‘While/When the master craftsman had left the workshop, the apprentice worked with less care’. Standard causal adverbials can be questioned or negated, which shows that they contribute to the truth conditional, at-issue content of the host clause:

Thus the old (=temporal) and the new (=causal) usage are semantically different; but there is pragmatic equivalence in many contexts and the same context update of the common ground is performed. The cancelability test (27), (28) is designed to provide a context in which this pragmatic equivalence does not arise, but such contexts are not typical in natural language use.

In order to analyse the semantic change, we sketch a semantic and pragmatic analysis of a simplified example (30) below.

(31a) and (31b) paraphrase the temporal ‘source’ and the causal ‘goal’ interpretation respectively.

Both interpretations involve two propositions, the one expressed by the subordinate clause and the one expressed by the main clause.

The critical expression CE in the example is weyl. The two different meanings for the connective, the old temporal one and the new causal one, are given in (33a) and (33b) (see e.g. Fabricius-Hansen (1991), Romero & von Stechow (2009), Lewis (1973) for a basic semantics of temporal and causal subordinate clauses).

This leads us to the following meaning components and their sources for the two analyses:

We see that once more, there is pragmatic equivalence with different sources for some meaning components. A former implicature (34d) is incorporated into the lexical semantics (35c). The change in meaning falls under the Constant Entailments hypothesis as stated in the previous section.

Our analysis answers (36).

If a meaning component of an expression contributes to or constitutes its truth conditional content, this is the most typical instance of ‘coded/conventionalized’ meaning. The famous case of causal connectives developing out of temporal ones thus typically involves change along pathway (23; 1a) (implicature > truth conditional).

In the previous subsection we dealt with the widely-discussed phenomenon that meaning components from conversational implicature develop into truth conditional content. To our knowledge, the opposite case, i.e. the development of merely implicated from truth conditional content (cf. pathway (23; 1b) in the taxonomy above), has not been addressed explicitly in the previous literature. In this section, we shall argue that the evolution of quantifier expressions from size nouns is a potential candidate for this pathway – in English and in further languages with quantifiers from comparable sources.

Quantifiers have special pragmatic properties. They have been argued to systematically evoke pragmatic enrichment through scalar implicatures (Grice 1975). In the following example, the use of the quantifier a lot of/lots of triggers the conversational implicature that not all of the students fell sick during the excursion:

The meaning component that not all of the students fell sick during the excursion is not part of the truth conditional contribution of the quantifier. Its status as a conversational implicature is confirmed by the standard tests: it can be cancelled, suspended or reinforced:

The quantifiers are part of an entailment scale whose different positions are lexicalized by a set of specific quantifying expressions – a Horn scale (Horn 1972), here:

According to the classical analysis of implicatures, the speaker chooses the lexical item from this scale in accordance with Grice’s (1975) second maxim of Quantity, which requires them to be as informative as is required (no more, no less). Since the scalar implicature is systematically (or ‘generally’) evoked by the quantifier, it is not tied to a specific context. Grice (1975; 1989) thus calls inferences of this type ‘generalized’ conversational implicatures. (More modern incarnations would make use of an exhaustivity operator to derive this meaning component; see e.g. Chierchia & Fox & Spector 2012).

This “inferential meaning […] constructed around the basis of lexical or coded meaning” (Levinson 2010: 174) is of special importance to us when we consider the diachronic development of certain quantified expressions. There is a well-attested diachronic path from binominal constructions with lexical ‘part-of’ meanings to quantifiers (e.g. Traugott 2008a; b; Brems 2011; 2012 on English quantifiers such as a bit of or a lot of,/lots of, heaps of, a bunch of; cf. also Seržant 2021 for a typological survey). In the following, we will illustrate this development by looking at the English large quantity quantifier a lot of/lots of. We shall thus argue that the meaning component ‘part of, not all’ is originally lexically encoded as part of a binominal construction (= NP of NP-syntagm, cf. Brems 2011; 2012). But when the first noun, i.e. the ‘size noun’ (ibid.), structurally develops into a quantifier, it is integrated into a lexical scale of the type in (39). Consequently, this meaning component becomes a pragmatic inference, a generalized conversational implicature.

In the binominal construction, in which a lot/lot(s) does not function as a quantifier, and which is marginally still in use in modern English, a lot/lot(s) can be modified or even be preceded by a quantifier, cf. (40a). In the plural (=lots), it triggers verbal agreement if used as a subject, cf. (40b).

Originally, lot(s) denoted objects of random selection that were used in decision making or in the resolution of disputes (cf. Oxford English Dictionary [OED], sense I.1.a) and from this the more abstract meaning in which it denotes something that falls to someone by fate developed that is still sporadically used in contemporary English (cf. the lof of the working man = ‘the fate’ of the working man, Brems 2012: 203). In (41a) and (41b) it is used with an even more abstract sense. According to the OED, lot may refer to “[a] part or portion of something; a number of things or people forming part of a larger whole.” (sense II.8). This sense is also attested quite early, e.g. in the following examples from the Ormulum:

Brems (2011: 211) cites (41b) and further examples from the Ormulum and argues that these are instances of an early ‘part of’ use where lot refers to a portion or part and the second NP in the of-phrase (cf. Þiss lott off all crisstene follc in (41a) or an lott off tatt Judisshenn follc in (41b)) can either contain animate or inanimate concrete nouns or even more abstract nouns. In these uses, the noun lot lexically contributes a ‘part of’ meaning.

The quantifier use, in which a lot of and lots of denote “[a] large number or amount” (OED, sense V.18), seems to occur only several hundred years later. In Brems (2011; 2012) data ranging from 1780 to 1850, examples are attested that are ambiguous between the old lexical use and the new quantifier reading, cf. (42). Such examples can be paraphrased either by ‘a group of’ or by ‘much, many’ (Brems 2011: 212; 2012: 218).

Brems (2012: 217) notes that in the ambiguous examples the entity that is portioned of or quantified over is typically something concrete, either animate or inanimate. Somewhat later unambiguous attestations of plural lots of quantifying the abstract room emerge, cf. (43a). Traugott (2008a) provides examples where a lot is modified by a degree modifier such as quite, (43b). In examples of this type, a lot is unambiguously used as a quantifier. They occur from around 1900 on.

We now turn to our semantic analysis of the change. We present an analysis of the origin, choosing the lexical ‘part of’ construction as our point of departure (OED sense II.8 from above). We relate this analysis to an analysis of the goal of the change, the quantifier construction, by way of a bridging context. The bridging context is a predicative use of the ‘part of’ construction as exemplified by (41b). Instead of actual examples like (41), we use the simpler (44).

We focus on the interpretation in terms of mereological parts of this ‘part of’ construction (while this may not be the only possible meaning, it is certainly prominent). The meaning of (44) is represented in (45). The concepts of plural ontology that (45) uses are introduced in (46) for the interested reader.

Next, let’s look at the analysis of the quantificational counterpart of (44), (47). The standard semantics in terms of Generalized Quantifier theory (e.g. Barwise & Cooper 1981) is given in (48).

(48) is intended to be neutral with regard to the particular reading that the quantifier a lot gives rise to. Parallel to many, several readings are possible (cf. e.g. Partee (1989), Westerståhl (1985), Penka (2018) among many others; see in particular Penka for examples with many parallel to (48’) and the terminology). Penka (2018) offers a compositional derivation of the three readings without assuming any lexical ambiguity for many. Although we cannot do justice to her proposal here,5 we conjecture that similar to her analysis for many, there are compositional routes to calculating different values for n in (48) that are instrumental to deriving the different readings. Thus we will work with one interpretation for a lot, (49b) below.

(49) puts the ‘old’ contribution of lot and the new contribution of lot side by side. Clearly, the lexical change in isolation appears unmotivated.

But in the context of a predicative sentence like (44), the path of the reanalysis of lot becomes transparent. In (50), we take a closer look at the entailments of the semantics of (44), i.e. of plain (50a). (50b) presents an entailment of (50a) in (frequent) contexts in which the two groups related by the ‘part of’ relation are genuine pluralities. In that case, the sets described by the two predicates have an intersection with multiple members. Thus (50b) puts us on track towards the truth conditional content of the (future) quantifier.6 (50c) examines another entailment of (50a): if the smaller group is a proper part of the larger group,7 then not all entities that have the property of being in the larger group belong to the smaller group.

This entailment is familiar from the quantifier construction as well: it is the well-known scalar implicature ‘not all’. What we observe in this semantic change, then, is a shift in the origin of an entailment between the source and the goal, and another shift to what we might call a pragmatically equivalent restatement of the truth conditions. The overall sentence meaning in the bridging contexts remains stable.

Of particular interest in the present context is the shift in the origin of the meaning component ‘not all’. It was an entailment of the truth conditional content of the source construction. In the goal construction, it is an implicature.

A reviewer points out to us that in cardinal interpretations of examples like (48’c), the implicature systematically fails to arise (e.g. in (48’c) there is no implicature ‘there aren’t all apples’). This is compatible with our analysis, because for PDE a lot the meaning component is simply an implicature and doesn’t always have to be there. The implicature is relevant for the application of Constant Entailments in the bridging contexts at the time of the change: In those contexts, Constant Entailments will be met if the implicature is present. This is the case for proportional interpretations.

The proportional interpretation is one possible interpretation of the newly emerging quantifier. In the source construction where lot is still non-quantificational, it lexically conveys ‘part of a whole’ (in combination with the of-NP2-phrase). Simplifying a bit, ‘a lot of NP2’ originally meant something like ‘a part of NP2’, as shown in (45), (51). This type of part-of-reading is interpretively close to the proportional reading of the quantifier cf. (48), (52). The readings have similar truth conditions and therefore lead to bridging contexts with stable common ground updates containing the meaning component ‘not all’. The crucial difference is that in the original construction, this component is part of the assertion, whereas in the newly emerging quantifier construction8 it is an implicature triggered by the quantifier, a new functional item that stands in paradigmatic relations with other quantifiers (some, all etc.) and has come to lexicalize a position on the Horn scale in (39).

If Penka’s (2018) analysis is right that the cardinal reading and the proportional one can be derived compositionally from the same lexical entry, we would expect that the reanalysis of a lot as a quantifier also enables uses with cardinal interpretations, but of course in different types of co- and contexts. And this is what the diachronic studies by Traugott (2008a, b) and Brems (2011; 2012) indicate: Their examples for unambiguously quantificational uses are the ones with cardinal interpretations, cf. (43a) above. And those are attested no earlier than the proportional readings in the bridging contexts. cf. (41b) – which is what our analysis predicts.

While we have pointed out some interesting issues in this change that call for further investigation, our case study provides a plausible example of the shift from truth conditional content to implicature, (23;1b).

A further issue we want to briefly address is the following: The development of ‘size’ nouns into quantifiers can also be regarded as a case of grammaticalization as it involves the evolution of a functional from a lexical element. There is a lot of controversy concerning the semantic developments in grammaticalization. In functionalist approaches, grammaticalization is often taken to be characterized by a ‘weakening’ or ‘bleaching’ of meaning or the like. We cannot do justice to this discussion here, but cf. Eckardt (2006: 33–34) for a critical evaluation of some widely-used terms and conceptuatlizations. Alternatively, it has been proposed by von Fintel (1995: 185) that in grammaticalization the sentence as a whole “remains (largely) constant in meaning” and that “Grammaticalization is a re-arrangement of meaning, not a change of meaning. Many examples of grammaticalization have always been analyzed in a similar, ‘musical chairs’ kind of way. Jespersen’s famous cycle of negation is a prime example.” (p. 186). This is an important predecessor to our proposal that reanalysis may be enabled by Constant Entailments.

As outlined above, we would argue that meaning changes (along with morpho-syntactic changes) are reanalyses that occur in acquisition (=‘imperfect acquisition’). In usage-based accounts, by contrast, it has been claimed that frequency of use and routinization may affect the meanings of expressions.9 As objected by Haspelmath (1999), however, “semantic bleaching or generalization that is so often observed in the development from a lexical to a functional category […] does not seem to be a consequence of routinization, unlike the phonological and syntactic changes of grammaticalized items, but a pre-requisite for it.” (p. 1062). Applying this argumentation to our case study on the development of quantifiers, for example, Haspelmath would probably argue that a lot must have first undergone a shift from a more concrete lexical meaning to a pure quantificational meaning to enable to fulfill a “basic discourse function” (ibid.), which is in turn the prerequisite for habituation or routinization. We would argue that this shift occurs through imperfect acquisition in bridging contexts with stable common ground updates.

Presupposition is another ‘conventional’ meaning component (in the sense of Grice and Levinson above) besides truth conditions. The role of presuppositions has not received much explicit attention in research on meaning change (but see Eckardt 2001; 2009).10 Beck & Gergel (2015) analyze the diachronic development of English again and view one of its transitions as an instance of pathway (23;2b) (from truth conditional to presuppositional) in our taxonomy above. It would be an interesting question for future research whether the transition of truth conditional into presuppositional meaning is typical for the diachronic development of iterative expressions. Discourse particles have been argued to be a further class of lexical elements that trigger presuppositions (cf. Kratzer 2004; Grosz 2020 for presuppositional analyses). In German, which has a very rich inventory, discourse particles derive from various diachronic sources. Many have been shown to go back to adverbially used adjectives (e.g. wohl ‘well’, ruhig ‘calm’, eben ‘even’ etc.) and their development would thus be a further typical example for pathway (23;2b). Besides iteratives and discourse particles, focus particles (FPs) such nur ‘only’, allein ‘solely’ or also or too or German auch ‘also’ are widely considered to be classical cases of lexical presupposition triggers. According to the standard analysis, they associate with the focused constituent in the sentence (Rooth 1985; 1992, a.m.o). This is the instance of interpretive change we will concentrate on here. We investigate, concretely, the CE allein ‘alone’.

The source of the change is the adverbial use of the adjective allein ‘by oneself, alone, sole, solitary, without company/assistance’, which goes back to Middle High German all + ein(e/alein(e), lit. ‘all one’ (Pfeifer et al. 1993). The adjective is attested from the 12^th century onwards. The focus-particle meaning use develops almost simultaneously (DWB2, lemma ALLEIN). A modern German example for the adjectival use is given in (53).

In this use, the adjective can be modified by intensifiers as in ganz allein (‘all alone’) or by a FP like auch ‘also’ or sogar ‘even’ (cf. Die Intendantin kann auch/sogar allein arbeiten ‘The director can also/even work alone’).

The target of the change is the focus particle with the meaning of only/solely, as exemplified by the modern German example in (54a). When used as an FP, allein cannot be intensified (#ganz allein ‘all/very only’). In the sentence in (54b) allein can thus only be interpreted adjectivally and not as the focus particle. (Syntactically, the change arguably involves a reanalysis of the syntactic category of the CE from Adj > F(ocus) P, but this is not what interests us here).

We hypothesize that the bridging context for this change (see section 3) are ambiguous sentence structures. The version of the example given in (55) is such a context. The ambiguity exists because the FP can syntactically occur in a post-posed position relative to their associate.

Let us examine (55), the bridging context, semantically. The two readings – adjective and focus particle reading – are paraphrased in (56).

Once more, the lexical contribution of the adjective and that of the focus particle are strikingly different. We sketch a standard semantics for both in (57) (see e.g. Lasersohn 1995; Rooth 1985; 1992, a.m.o).

Things make more sense in the overall sentence context. The interpretations of (55) according to the two analyses (57a) and (57b) are compared in (58) and (59).

Assuming that the context provides a group of relevant individuals – say, the director, the manager and the sponsor – then to say that the director sat at home without the other two amounts to the same thing as saying that out of the three alternatives {that the director sat at home, that the manager sat at home, that the sponsor sat at home}, the single true one is that the director sat at home. So contextual information can make (58) and (59) pragmatically equivalent – in line with Constant Entailments.

But let’s also consider the status of the meaning components, in particular (58a), (59a). It is asserted in the old analysis, but presupposed (e.g. Rooth 1985) in the new analysis. (60a) vs. (60b) demonstrate that this is accurate for allein specifically.

Hence this is an example of interpretive change involving change of a meaning component from truth conditional content to PSP. The boldfaced meaning component in the source meaning of allein in (57a) (repeated below) changes its status from assertion to PSP.11

This subsection considers the source of the English complementizer whether and its diachronic trajectory. Etymologically, whether comes from the Old English indeterminate pronoun hwæþer ‘which of the two’ (Eckardt & Walkden 2022). This pronoun had interrogative uses (61a), but also quantificational uses (61b).

The semantics of phrases with hwæþer (like which of the evils in (61a)) is given in (62) (Hamblin 1973; Kotek 2019; Beck 2020 and specifically Eckardt & Walkden 2022). The phrase can only be used appropriately if the restrictor property P is true of exactly two entities (e.g. there are exactly two evils). Then, the phrase denotes the set of alternative entities that have the property P. Note that the semantics is presuppositional. (62’) sketches a standard Hamblin/Karttunen question semantics of (61a) (Hamblin 1973; Karttunen 1977). According to such a semantics, a question denotes the set of its possible answers. The hwæðer-phrase contributes a set of alternatives (62’b). In the example, the two alternatives are entities (type <e>), but other types of alternatives are possible as well.12 Those alternatives project to the propositional level, (62’c). The propositional alternatives are the possible answers to the question, i.e. the question meaning.

We follow Eckardt & Walkden (2022) in how they trace the diachronic development of the OE indeterminate pronoun hwæþer to the marker of Yes/No-questions. Their case study provides an example of interpretive change of the type (23, 2a): from PSP to truth conditional content. In addition to well-understood examples like (61a) on the one hand (the source of the interpretive change), and data like (63) on the other hand (the goal of the change, parallel to PDE whether), Eckardt & Walkden observe less expected occurrences like (64) and (65):

Eckardt & Walkden show how (64) and (65) may pave the way towards the analysis of hwæðer as the interrogative complementizer for polar questions, i.e. the semantics of PDE whether, given in (66) (Hamblin 1973; Karttunen 1977).

Simplifying their discussion somewhat, an analysis of (64) is sketched in (64’). Two assumptions are crucial: first, the domain of the indeterminate pronoun hwæðer is the embedded proposition and its negation (with the negated proposition elided or not explicitly given at all), as shown in (64’b). Hwæðer is interpreted in its underlying position in the embedded clause, not in its surface position in the matrix clause, cf. the structure in (64’a). Secondly, the alternatives calculated in this way in (64’b) project beyond the level of the embedded clause and into the matrix, making (64) a matrix question, cf. (64c).

Association with an embedded proposition enables reanalysis to complementizer whether, which has the two alternatives {p,¬p} built in. Example (65) takes us another step closer to the goal analysis. In (65), hwæðer occurs in embedded position. The embedded clause may thus be analyzed as in (65’b) using the complementizer meaning in (66a). The example is still unusual in that this question meaning occurs in the position of the argument of a belief – ininterpretable by way of continuing projecting the alternatives beyond the embedding verb to the matrix, yielding the matrix question in (65’c).

In a last step, the original embedding structure with belief-type verbs is lost. The structures with hwæðer may then fully take on their present day syntactic and semantic properties, already attested in (63). See Eckard & Walkden (2022) for details.

For present purposes, we are particularly interested in the change in the word hwæðer itself. The two analyses are put side by side in (67).

It is immediately obvious that the PSP from the source analysis is lost in the goal analysis. It is slightly less obvious that it is shifted to the level of truth conditional content because of the detour that the PSP takes: the two alternatives are identified, in bridging contexts like (64), (65), as {p,¬p} by Eckardt & Walkden (2022). By way of this identification the former PSP shows up in the truth conditional contribution ‘q=p ∨ q=¬p’ in the complementizer meaning.

In (68) and (69) we demonstrate that Eckardt & Walkden’s (65) is indeed a bridging context in that the old and the new analysis, applied to the embedded clause, lead to the same interpretation. The change is in line with a reanalysis relying on a combinatory ambiguity under Constant Entailments.

In sum, Eckardt & Walkden’s analysis presents us with an example of an interpretive change from PSP to truth conditional content. This is an indication that the pathway (23; 2a) may be available too.