Internally caused change as change by inner predisposition: Comparative evidence from Romance

Abstract

This article investigates the morphosyntax of verbs of internally caused scalar change and the main facets of their meaning. Availing ourselves of primary evidence from Italian, French, and Spanish corpora, we argue that the verbs under scrutiny divide into three subclasses whose common denominator is that they encode ‘change by inner predisposition’ or change that is possible because of inherent propensities of specific entities. We understand inner predisposition as a feature of the content of the verbs investigated, which has relevance to the choice of the undergoer, alongside the cause of two of the subclasses. Apart from this feature verbs of change by inner predisposition are indistinguishable from other verbs of change. We argue against an agentive or generalized causative analysis, and we advance a proposal for the lexical semantics of representatives of each of the three subclasses and for the alternations in which they participate.

1. Introduction

Differently from inherent properties, for example the colour of a flower or the sweetness of honey, changing states of affairs can have, and indeed often do have, contingent causes which are external to the individuals undergoing the change. The degree to which changes are caused externally, however, varies a great deal, as does the range of causers and undergoers that any given change is possible for. Whereas blossoming is due to the genetic predisposition of specific entities to react to specific ambient conditions, opening changes are not similarly restricted.

The linguistic encoding of different kinds of change has fascinated linguists since Jespersen’s (1927: 332–337) identification of Move and Change verbs (see, e.g. Haspelmath 1993). Drawing on Smith (1970) (see also Nedjalkov & Silnitsky 1973; Labelle 1992; Levin 1993), Levin & Rappaport Hovav (1995: 89–98) drew a distinction between external and internal causation. In their analysis, the former type of causation characterizes eventualities which are brought about under immediate control by an external cause. Internally caused eventualities, instead, lack an external cause and ensue from an inherent property of the argument of the verb. In Levin & Rappaport Hovav’s (1995) proposal, the lexical semantic representation of verbs of internal causation does not include a component of cause.

Subsequent scholarship has brought to light non-trivial evidence of transitive–intransitive alternations with verbs which intuitively classify as internally caused (McKoon & Macfarland 2000; Wright 2001, 2002; Rodríguez 2011). A distinction has been drawn between agentive and causative internal causation roots (Alexiadou 2014). It has also been argued that verbs of internally caused change are not characterized by a grammatically relevant property of their own and are indistinguishable from verbs of externally caused change (Rappaport Hovav 2020).

This article seeks to enhance the understanding of the linguistic encoding of internal causation. We focus exclusively on verbs of internally caused scalar change, ¹ addressing the question of whether these verbs can be characterized as a single class and whether they include causation or agentivity as facets of their semantic make-up. Reporting the findings of comparative analysis of Italian, French, and Spanish, we observe that, in the languages under scrutiny, putative verbs of internally caused change divide into three formal groups: (i) non-alternating verbs which do not exhibit the anticausative morpheme se in the intransitive (e.g. verbs for ‘blossom’); (ii) alternating verbs which do not take anticausative se in the intransitive (e.g. Italian and French verbs for ‘ferment’); (iii) verbs which exhibit little propensity for transitivization but are instead characterized by ±se intransitive realizations (e.g. verbs for ‘rust’). The variation in the morphosyntactic behaviour of the verbs under scrutiny, and in their range of meanings and transitive subjects, supports the insight of previous research that verbs of internally caused change are not a homogeneous class (McKoon & Macfarland 2000; Beavers & Koontz-Garboden 2018) and that they include a subclass which can take human instigators (Alexiadou 2014). Adducing evidence from both the range of arguments admitted by the verbs under scrutiny and periphrastic causation (Donazzan, Raffy & von Heusinger 2020; Floricic 2022), we contend that the common denominator of the meaning of the verbs under analysis is ‘change by inner predisposition’, i.e. change that is driven by the inherent propensity of an entity to change, reacting to, or interacting with, specific triggers. Comparative evidence from agentive and external causation verbs rules out agentivity as a facet of the meaning of the verbs in our sample, suggesting that one subset is also not causative.

We thus identify three subclasses: we argue that the verbs of the one class are not causative, while the verbs of the other two classes alternate as non-causative/causative pairs in the lexicon. The behavioural differences between the last two classes depend on the causes that the content component of their meaning is compatible with. We claim that a key component of the content of one of the two classes is understood as a process. This class thus has duration and manner inferences and can describe complex causation chains, whereby human causers manipulate inanimate triggers, consistently with the tendency for manner verbs to take animate actors (Beavers & Koontz-Garboden 2012). The animacy of the manipulator results in agentivity inferences (Van Valin & Wilkins 1996) and the propensity towards transitivization (Hopper & Thompson 1980). In contrast, the content of the other causative class is construed in terms of the result of the change, and thus this class does not lend itself to manner inferences or the description of complex causation chains. This class encodes changes which are brought about by the interaction of the undergoer with specific inanimate triggers, which are licensed by the result in the content. The different causes of the two classes are a factor in their different propensities towards se anticausativization. Assuming se marks the silent realization of the causer in the lexical phase of the semantics–syntax linking (understood in Van Valin’s 2023 terms), and allowance being made for non-trivial differences between the languages investigated, our analysis of this contrast rests on the claim that this strategy is sensitive to the satisfaction of Wolff’s (2003) conditions on direct causation.

In the sections that follow, we make some preliminary observations on verbs of internally caused change, explaining which expectations and predictions informed our research (§2). We then discuss our methodology and findings (§3). In Section 4, we introduce the notion of change by inner predisposition and we analyse the lexico-semantic structure of the subclasses emerging from the investigation. Se anticausativization is discussed in Section 5. We draw our conclusions in Section 6.

2. Research context and predictions

Chierchia (1989) characterized the contrast between the verbs that participate in the causative alternation and those that do not in terms of the divide between unaccusatives, which are expected to alternate, and unergatives, which are not. Noting that some unaccusatives (i.e. verbs of change of state, e.g. bloom, blossom, flower, etc.) lack a transitive counterpart, Levin & Rappaport Hovav (1995) departed from Chierchia’s proposal and instead based their treatment of non-alternating verbs on the absence of an external cause. They proposed the lexical semantic templates in (1) and (2) for internally and externally caused verbs, respectively.

Levin & Rappaport Hovav (1995: 110–119) did note that some putative internally caused verbs – specifically, agentive verbs of manner of motion and verbs of emission – have transitive counterparts. For Levin & Rappaport Hovav (1995), the transitive counterpart of agentive verbs of manner of motion derives from the intransitive form. As for verbs of emission, the authors suggest that some such verbs must be compatible with a dual classification as internally or externally caused. A relevant example is English burn, which means ‘emit light, heat, or flames’ or ‘consume by fire or combustion’ and only alternates with the latter meaning.

Although verbs of external causation unsurprisingly received the most attention in subsequent literature, McKoon & Macfarland (2000) and Wright (2001, 2002) brought to light some intriguing properties of verbs of internally caused change. Particularly relevant to our present purposes is McKoon & Macfarland (2000)’s observation that such verbs offer robust evidence of intransitive–transitive alternations, although their transitive exhibits a more limited range of causes than that of verbs of externally caused change. Their explanation for this contrast relies on the notion that the lexical semantic template of internally caused verbs only licenses one participant (cf. 1). Another participant may be licensed by the content component of the template and must thus be compatible with the change of state that this describes. Instead, externally caused templates license two participants (cf. 2), the one being independent of the change of state and hence free from selectional restrictions. Similarly, Wright suggested that verbs of internally caused change select for inanimate, non-agentive, and controllable causes that are ‘inherent to the meaning associated with the verbs themselves’ (2002: 344).

Subsequent literature shed light on key differences between subclasses of putative internal causation verbs. Rappaport Hovav (2020) drew a distinction between verbs meaning ‘blossom’ and ‘flower’, which she analysed as non-causative verbs of emission, and the remaining ones, which she claimed to be causative. Alexiadou (2014) noted that a subclass of Greek internal causation roots (e.g. ‘ferment’) takes agentive transitive subjects, in contrast with another subclass (e.g. ‘blossom’, see also Lavidas 2007: 118), which takes causer transitive subjects. Alexiadou (2014) proposed a syntactic analysis which predicted that only the former subclass would occur in the passive. ²

Based on the above literature, the expectation was that Romance verbs of internally caused change either do not alternate or select [-animate] causes, exception being made for a class represented by ‘ferment’, which should take [+animate] transitive subjects. We thus examined the behaviour of a sample of putative verbs of internally caused change in the Romance languages under scrutiny, comparing it to that of verbs of external causation, which normally alternate (cf. 3), passivize (cf. 4), and take [±animate] transitive subjects and by-phrases (cf. 3–4). ³

We also considered the distribution of the morpheme se with the verbs in our sample, since this morpheme can figure in the intransitive member of the causative alternation in Romance (cf. 3) and other languages (Haspelmath 1993).

Anticausative se has, on the one hand, been analysed as an argument realization strategy: the marker of an unexpressed causer (Rothemberg 1974; Labelle 1992; Cennamo 1995; Centineo 1995; Van Valin & LaPolla 1997: 416–418, etc.), of an external argument position in a syntactic Voice projection (Alexiadou et al. 2006, 2015; Schäfer 2008), or, alternatively, a reflexivizer (Koontz-Garboden 2009). On the other hand, se has also been analysed as the marker of aspectual properties of verbs or constructions normally described in terms of telicity, resultativity, or more fine-grained scalar notions (Zribi-Hertz 1987; Cennamo 1999, 2012, 2015; Folli 2002; Jezek 2003; Manente 2008; Labelle & Doron 2010; Cennamo & Jezek 2011; Vivanco 2021, etc.).

The aspectual approach explains the language-specific distribution of se with some non-alternating unaccusatives (De Miguel & Fernández Lagunilla 2000) and the diachronic development of se in some Romance languages (Cennamo 2021). Nonetheless, this approach is challenged by evidence of se marking of verbs or constructions lacking the relevant aspectual properties (see Schäfer 2008; Martin & Schäfer 2014; Alexiadou et al. 2015; and, for counterarguments regarding Spanish, Vivanco 2021). While we do not define se in aspectual terms, we shall note the relevance of aspectual considerations in the discussion of the -se attestations of the ±se verbs.

The analysis of se as an argument realization strategy has the non-trivial advantage that it can capture the syncretism of anticausative, passive, reflexive, and impersonal se (see Bentley 2006 for Italian; Alexiadou et al. 2015). It is the approach adopted in this article, where we assume that se marks a silent argument in the highest-ranking argument position in the lexical semantic structure of the predicate in the clause. In the case of causatives, this is the position of the causer, which cannot then be realized overtly as an argument of the verb. A prediction of this analysis is that all causatives should, in principle, permit se anticausativization. The failure for any causatives to anticausativize with se requires an explanation.

3. A comparative analysis of verbs of internally caused change

3.1 Methodology

The findings reported in this article were obtained as follows. To begin with, we gathered and examined the first 500 hits of a sample of verbs in the itTenTen16, frTenTen17, and esTenTen18 corpora of Sketch Engine (Jakubíček et al. 2013). ⁴ The verbs were searched as Lemmas, with Verb being specified under Part of Speech. Some clarification on the verbs chosen for the investigation is necessary. These were deemed to be rough translational equivalents of eight verbs of internally caused change listed by Levin & Rappaport Hovav (1995: 282) (blossom, burn, ferment, flower, germinate, rot, rust, wither), to which we added Romance verbs for ‘(cause to) become mouldy’, henceforth named ‘mould-v(erb(s))’, and age. ⁵ To identify appropriate counterparts of the English verbs, we relied on the advice of native speaker consultants and the use of dictionaries. ⁶ Despite working with genetically related languages, we did not find exact synonyms for all the verbs to be included in the sample. This should not come as a surprise. Lexical items which share the same etyma may undergo divergent semantic developments or become obsolete, or stylistically marked, in one language, but not its sisters. We attempted to select verbs that are ‘characteristic’ in Talmy’s (2007[1985]: 72) sense, i.e. the most colloquial in style and frequent in speech among quasi-synonymous alternatives, although this was not possible to the same extent for all the verbs (see, for example, Spanish ‘mould-v’ or two of the ‘burn’ verbs, which are not colloquial).

With respect to the specific verbs that we chose for the investigation, we should note, first, that in accordance with insights discussed in Section 2, we distinguished two meanings of ‘burn’: ‘consume or destroy with fire, heat, or a metabolic process’ and ‘be incandescent, produce heat and flame’. We thus included two Italian and Spanish translations of English burn, namely ardere/arder, labelled ‘burn_i’, and bruciare/quemar, labelled ‘burn_ii’. The first pair was expected to more closely fit the meaning of ‘burn from within’, and the second pair was expected to translate both meanings. Having included two translations for burn, we had 11 forms in our Italian and Spanish samples. Although French has a participial adjective ardent ‘very hot, burning’, the cognate verb hardly occurs in the modern language (Franck Floricic, p.c.). When searched, this verb only yielded 22 hits, which had very limited geolinguistic and textual distribution. Therefore, we did not include it in the French sample.

Some of the verbs included in the sample are broader in meaning than their counterparts in the sister languages. Thus, the French verb chosen for ‘blossom’, éclore, describes not only the opening of a bud into a flower, like Italian sbocciare, but also the hatching of chicks. As for Spanish brotar, it can describe the appearance of a flower on a plant, but also, more broadly, the appearance of leaves and greens, or the gushing of water from rocks or fluids from the body. In addition to translating rust, Spanish oxidar describes a wider range of processes involving oxidation. Other verbs included in the Spanish sample (enmohecer ‘mould-v’; pudrir ‘rot’) can translate rust, alongside those that we considered to be their primary meanings. It is therefore essential to note that in our investigation, we recorded and analysed the profile of individual verbs rather than sets of putatively synonymous verbs taken together. We list in Table 1 the verbs included in our sample.

Table 1 The verbs included in the sample.

We measured the propensity of the verbs to occur as transitives, passives, and +se or -se intransitives. ⁷ We examined the range of transitive subjects occurring with the verbs that were found to alternate, and we compared the meanings of the verbs when transitive and intransitive. Following McNally & Spalek (2022), we included in the investigation the metaphorical or extended meanings of the verbs, assuming that grammatically relevant aspects of meaning can be carried over in figurative polysemy. The consideration of this evidence shed light on important aspects of the analysis, including the notion of dual classification (§4.1). The full gamut of percentages recorded for each verb in each construction is provided in Appendix 1. ⁸

The contrastive analysis of the morphosyntax of the verbs in three languages was expected to shed light on the typology of verbs of internally caused change. Specifically, we assumed that verbs would fall into the same formal subgroups by virtue of shared aspects of meaning. We also hypothesized that should verbs with comparable meanings fall into different formal groups, this would be indicative of facets of meaning that only a subset of the grammars is sensitive to, or, alternatively, of grammatically relevant semantic changes that have affected a given verb but not its counterparts in the sister languages. We should add that key microvariational differences between the sister languages (regarding perfect auxiliary selection and periphrastic causation) allowed us to consider a wider range of diagnostics than we would have been able to rely on, had we worked on a single language.

3.2 The corpus findings

3.2.1 The transitive–intransitive alternations

Although most verbs turned out to participate in transitive–intransitive alternations, they did so to different extents. This can be seen in Figure 1, which displays the transitive percentages obtained for each verb in each of the three languages. We remind the reader that there are no results for ‘burn_i’ in French.

Figure 1 Transitive percentages per verb.

The lowest transitive percentages (<1%) were found with ‘blossom’ in the three languages; ‘mould-v’ in Italian and French; ‘flower’ in Italian and Spanish; ‘rot’ in Italian; ‘burn_i’ in Spanish; and, lastly, ‘germinate’ in French. We classified the few transitive occurrences of these verbs as metonymic, non-literal or non-conventional. The metonymic examples are characterized by a part-whole relationship between the undergoer and the instigator of the change, which are both affected (cf. 5). The non-literal examples describe changes which do not occur in the physical world (cf. 6). Lastly, the non-conventional occurrences departed from the meanings expected from the dictionary descriptions (cf. 7).

The verbs which yielded transitive percentages that were lower than 1% will henceforth be referred to as [-tr(ansitive)], which is not meant to suggest that verbs with these meanings cannot transitivize in principle, but rather that the verbs in question showed very little propensity to occur as transitives in our survey.

The verbs with higher transitive percentages will be labelled as [+tr]. ‘Burn_ii’ turned out to have the highest transitive frequency (>37%), which is unsurprising, given that it not only describes internally caused change but also an externally caused change of ‘consuming or destroying with fire, heat, or a metabolic process’ (recall that externally caused verbs transitivize (cf. 3)). In the three languages, this verb occurred with [±animate] causes, describing various processes of combustion. For brevity, we only provide Italian examples.

The remaining [+tr] verbs subdivided into three subgroups, characterizable as follows:

1. a. verbs which were attested with [±human] transitive subjects with the same meanings as in the intransitive (‘ferment’ and ‘age’ in the three languages; ‘germinate’ in Italian and Spanish);

2. b. verbs which occurred with [±human] transitive subjects with meanings which could extend, or be a subset of, the meanings of the intransitive (Italian ‘burn_i’ and ‘wither’, French ‘flower’, French and Spanish ‘rot’, Spanish ‘wither’);

3. c. verbs which only occurred with [-animate] transitive subjects (‘rust’ in the three languages; ‘mould-v’ in Spanish; French ‘wither’).

The key property of the (a) group is that these verbs take human transitive subjects when describing the same physical changes as they describe in the intransitive. In (9)–(11), we provide examples with ‘ferment’.

Similarly, the transitive attestations of Italian and Spanish ‘germinate’ can describe processes of germination which have a human instigator. We did not record the same results with French ‘germinate’.

The case of ‘age’ is comparable, insofar as in the three languages the relevant verbs combine with human instigators ‘making someone/something look/become older’.

In the three languages, the ‘age’ verbs were also attested with [-animate] transitive subjects, and the ‘ferment’ ones with [-human] ones, with the same meaning as in the intransitive.

The (b) group was characterized by idiosyncratic meaning differences between intransitive and transitive and by discrepancies between the sister languages. Thus, Italian ardere ‘burn_i’, in contrast with its Spanish cognate, was found to alternate like burn_ii’.

French ‘flower’ occurred transitively in the sense of ‘make flowery’ or ‘embellish with flowers’, a meaning that was not attested with the Italian and Spanish cognates.

When transitive, French pourrir was only attested in the sense of ‘ruin’ (cf. 22), which is one of its intransitive meanings, ‘rot’ and ‘ruin’ (cf. 23–24).

Similar observations were made with Spanish pudrir ‘rot’/‘rust’/‘spoil’/‘ruin’ and marchitar ‘wither’/‘spoil’/‘ruin’, although we found some examples meaning ‘rot’ and, respectively, ‘wither’ with inanimate transitive subjects. Italian appassire occurred both in the sense of ‘wither’ and of ‘soften or brown from cooking or exposure to heat’ in the intransitive (cf. 25–26). In contrast, only the latter meaning was attested in the transitive (cf. 27).

In sum, the verbs of group (b) only tend to transitivize with a selection of their meanings, and the analysis of the transitive occurrences highlighted differences between the three languages.

Lastly, group (c) gathers ‘rust’, Spanish ‘mould-v.’ and French ‘wither’, which were found to take [-animate] transitive subjects with no meaning differences between transitive and intransitive. The following examples illustrate the alternation with Spanish enmohecer ‘mould-v’ (but also ‘rust’ or ‘tarnish’).

Italian and French ‘rust’ can describe the rusting of metals or, alternatively, inactivity and the loss of physical or mental abilities. Below, we exemplify the former meaning with Italian arrugginire and the latter with French rouiller.

To recapitulate, non-trivial contrasts were found both in the respective transitive propensities of the verbs in our sample and in whether the transitive attestations had the same meanings as the intransitives. Only some of the [+tr] verbs turned out to take human transitive subjects when describing the same physical changes as are described in the intransitive: ‘ferment’ and ‘age’ in the three languages and ‘germinate’ in Italian and Spanish. Where we observed meaning differences between transitive and intransitive, we also found contrasts between the sister languages. Lastly, when [+tr], ‘rust’ and ‘mould-v’ stood out as verbs which occur with the same meanings in transitive and intransitive domains, exclusively taking [-animate] transitive subjects.

3.2.2 The passive

Figure 2 shows the percentages of passives recorded with each of the verbs in our sample.

Figure 2 Passive percentages per verb.

Since the languages under scrutiny have a passive formed with se and an active form of the verb, we should clarify that the se passives, included in the passive counts, were distinguished from the non-passive +se intransitives in terms of the presence of another unambiguous passive in the previous, the same, or the following sentence. To give but one example, since se cosecha ‘it is harvested’ is unambiguously passive, the verb cosechar being agentive rather than causative, the form se germina was also analysed as a passive in (34).

Admittedly, some se passives may not have been detected with this criterion and may instead have been classified as non-passive +se intransitives (see note xv).

A difference emerged between the (a) and (b) groups of [+tr] verbs (those attested with [±animate] transitive subjects) and the (c) group (attested with [-animate] transitive subjects). The verbs of the last group were only found as stative passives with auxiliary estar in Spanish (Cabredo Hofherr 2017: 242) or as participles lacking an auxiliary but followed by a by-phrase. The actors in these by-phrases pertained to the conceptual domains of ‘water’ (humidity, rain, etc.) or ‘time’ (age, boredom, death).

The verbs of the first two groups were not restricted in terms of the passive structures they occurred in, although they were not all attested in the passive. Examples (37) and (38) feature verbs from group (a), (39) a verb from group (b).

3.2.3 The distribution of se

The distribution of se in non-passive intransitives brought to light a divide between two verb groups, the one being characterized by the absence of se, the other exhibiting non-trivial percentages of +se intransitives, though more conspicuously so in Spanish than in the other languages (see also Heidinger 2015). More specifically, while ‘blossom’, ‘flower’, ‘germinate’, ‘burn_i’, and ‘age’ only yielded -se intransitives in the three languages, ‘wither’, ‘mould-v’, ‘rust’, and ‘burn_ii’ provided evidence of ±se variation. These findings are shown in Figures 3 and 4.⁹ Differently from the transitive and passive percentages, which were calculated over the whole set of examples collected for each verb, the ±se percentages pertain to the non-passive intransitive subsets (for the non-passive intransitive counts, see Appendix 1). We should also note that 8.4% of the 143 non-passive intransitives of Spanish fermentar ‘ferment’ and 98.8% of the 162 non-passive intransitives of Spanish pudrir ‘rot’ exhibited se. These verbs thus behave differently from their counterparts in Italian and French, which belong to the -se group.

Figure 3 -se marking of intransitives.

Figure 4 +se marking of intransitives.

Recall now that se has been analysed as the marker of aspectual properties of verbs or constructions normally described in terms of telicity or resultativity (§2). Consideration of the -se verbs in our sample suggests that the contrast between the -se and the ±se groups cannot be reduced to the former group being non-resultative, i.e. lacking a result state in their semantic make-up. Two resultativity diagnostics challenge this hypothesis: the selection of the perfect auxiliary essere ‘be’ in Italian (Sorace 2000; Cennamo 2008; Bentley and Eythórsson 2004) and the occurrence of the -se verbs as resultative adjectives or absolute participles (van Hout 2004; Bentley 2006: 355; Legendre 2017: 283–284). ¹⁰ With respect to perfect auxiliary alternation, we refer the reader to Appendix 2, which shows a clear preference for essere ‘be’ over avere ‘have’ for all verbs, except fermentare ‘ferment’, with which avere slightly outnumbered essere, and ardere ‘burn_i’, which was not attested in the perfect. ¹¹ As for the second diagnostic, in (40)–(42) we provide authentic examples of the -se verbs, bar ‘burn_i’, which did not yield any evidence of resultativity and will be assumed to be non-resultative (§4.1).

To be sure, this evidence is not indicative of the precise scalar properties of the verbs under analysis. While not being activities in Vendlerian terms, the verbs in (40)–(42) could be degree achievements (Dowty 1979: 88–90), or verbs of non-quantized change, i.e. verbs that describe the progression of a theme along a scale, introducing a measure of the amount to which the theme varies along the scale (Hay, Kennedy & Levin 1999), but failing to lexicalize a specific final state (Beavers 2011).

Taking ‘blossom’ and ‘ferment’ as representatives of the -se verbs (‘ferment’ only in Italian and French) and ‘rust’ as a ±se verb, we found that the notion of quantized change cuts across the -se vs. ±se distinction, with ‘blossom’ describing quantized change and the other two verbs failing to do so. Known tests to diagnose degree achievements are the entailment of the perfect by the progressive and the compatibility with additions indicating change by some amount (Bertinetto & Squartini 1995). All the ‘blossom’, ‘ferment’, and ‘rust’ verbs were attested in the progressive form or in aspectual predication with ‘begin’, which suggests that they are not punctual or allow readings whereby the event is extended. ¹² For brevity, we only give Italian examples here.

We tested the compatibility with ‘a little’/‘a lot’ and whether the progressive entails the perfect. The Italian and Spanish speakers (see note vi) judged the verbs to split into two groups: ‘blossom’ failed the tests, whereas ‘rust’ and ‘ferment’ did not. The judgements of the French speakers were more nuanced, but we found evidence of modification of both rouiller ‘rust’ and fermenter ‘ferment’ with ‘a lot’, while we did not find any comparable evidence with éclore ‘blossom’.

We shall thus assume that ‘rust’ and ‘ferment’ are degree achievements, whereas the ‘blossom’ verbs are not. ¹³ If the -se group is aspectually varied as discussed above and there are verbs of the same lexical aspectual type on both sides of the -se vs. ±se split, then this cannot be reduced to a ±result contrast. ¹⁴

Turning now to the ±se variation with individual verbs (see Figure 4), this lends support to the hypothesis that the +se intransitives involve a cause that is not expressed as an argument of the verb, while also indicating that the aspectual properties of the construction have a bearing on the intransitive realization of the ±se verbs (§2). To begin with, the +se intransitives of ‘burn_ii’ mean ‘scald oneself’ or ‘be burned by something’, which are externally caused meanings.

The -se intransitives, however, primarily mean ‘be incandescent, be in flames’, being synonymous with intransitive ‘burn_i’.

Recall that, different from its cognates, Spanish fermentar ‘ferment’ yielded a non-trivial +se percentage (8.4% of the non-passive intransitives). Interestingly, a cause is frequently mentioned with the +se intransitives of this verb. ¹⁵

Similar considerations can be made for Italian and French ‘mould-v’ and ‘rust’, which were unmarked when describing the condition of ‘being abandoned or stuck somewhere’. In contrast, they exhibited se when describing physical changes of ‘becoming mouldy’ and ‘rusting’, the cause occurring in the surrounding context. Italian examples are provided here.

Given that the +se intransitives presuppose a cause or exhibit one in the context, these findings lend support to the analysis of se as an argument realization strategy (§2). At the same time, the examples in (55)–(58) are suggestive of a distinction between the description of processes or conditions (cf. 55, 57) and completed events of change (cf. 56, 58). The same contrast is found with Italian and French ‘wither’. Observe how readings of potentiality or inception, coerced by sans ‘without’ in (60) and inizia a ‘begins to’ in (62), correlate negatively with se.

We conclude that the ±se variation is found with verbs which lexicalize both cause and result when the cause fails to be expressed as an argument of the verb (cf. 48–50; 54; 56; 58–59; 61). When one such verb is polysemous, one of its meanings may not encompass a cause and hence fail to be marked by se. This is the case with ‘burn_ii’, which has both externally and internally caused meanings (cf. 48–50 vs. 51–53), and with the figurative extensions of ‘rust’ and ‘mould-v’ (cf. 55, 57). Lastly, the durative component of the verb meaning may be the only grammatically relevant component because of contextual coercion, with the consequent lack of se marking (cf. 57, 60, 62). Our proposal on se will be formalized in Section 5.2.

3.3 The verbs under investigation: Formal groups

The results of our investigation support the insight of previous literature that the verbs of internally caused change (henceforth VICCs) constitute a composite class. Adopting as above a membership threshold of 1%, the morphosyntax of the verbs in our sample divides them into the groups shown in Table 2.

Table 2 The verbs under scrutiny: Formal groups.

The finding that the Romance VICCs in the top left cell show no propensity to transitivize, passivize, or se intransitivize suggests that, in the languages under investigation, these verbs must be distinguished from the verbs of externally caused change (VECCs). Some evidence on the behaviour of the VECCs was given in Section 2 (cf. 3–4). In addition, in Figures 5 and 6, we contrast the transitive percentages and, respectively, the +se percentages found with three VECCs and a representative from each of the three columns in Table 2. The data on the VECCs were drawn from the same Sketch Engine corpora as mentioned in Section 3.1, using the same methodology. ¹⁶ As above, the transitive percentages were calculated over the whole set of examples collected per each verb, the +se ones over the subset of intransitive examples.

Figure 5 Transitive percentages with VECCs and VICCs.

Figure 6 +se marking of intransitive VECCs and VICCs.

Figures 5 and 6 bring to light noticeable behavioural differences between VECCs and VICCs, with the ‘blossom’ verbs failing to transitivize or se intransitivize altogether.

Rappaport Hovav (2020) claimed that English blossom and flower should be analysed as verbs of emission. However, this analysis is problematic for Romance ‘blossom’ and ‘flower’, which occur as resultative participles modifying not only the putative emittee, but also the putative emitter (cf. 63 and 40–42). This behaviour is unknown to verbs of emission (cf. 64) (Levin 1993: 238).

In addition, the perfect auxiliary essere ‘be’ is prevalent with Italian ‘blossom’ and ‘flower’, which further suggests that these verbs are resultative (Appendix 2, §3.2.3). Instead, Rappaport Hovav’s (2020) analysis suits intransitive ‘burn_i’, which does describe emission (cf. 20) and provided no evidence of resultativity (§3.2.3). In the last analysis, the results captured in Table 2 and Figures 5 and 6 do not warrant a generalized account of the Romance VICCs as VECCs, while also suggesting that many VICCs share behavioural properties of the VECCs (see the middle and right-hand columns of Table 2). In the next sections, we make proposals for the common denominator of the meaning of verbs of internally caused change first, and then for the meaning differences between the formal groups emerging from Table 2.

4. Internally caused change as ‘change by inner predisposition’

4.1 Change by inner predisposition and dual classification

Although the literature has placed emphasis on the selectional restrictions on the transitive subjects of internally caused verbs of change (§2), we note, in line with Levin’s (1993: 246–247) analysis of English, that most of the verbs under discussion here take a limited range of undergoers. In their literal meanings, ‘blossom’ verbs describe changes that occur to flowers or buds; ‘flower’ verbs are predicated of plants, flowers, or places where flowers grow (e.g. grassland); ‘germinate’ verbs describe processes affecting seeds, grains, or bulbs; ‘ferment’ verbs are predicated of fruit, dairy, and other organic products; ‘rust’ verbs are predicated of metals, etc. To be sure, some of the verbs included in our sample are broader in meaning than their rough translational counterparts (§3.1). We also noted that some of the verbs have figurative extensions, in which case they do not impose as strict selectional restrictions on their undergoer as their literal counterparts (see also McKoon & Macfarland 2000; Rappaport Hovav 2020). Nonetheless, it can be argued that the common denominator of the verbs in the sample, in their non-figurative senses, i.e. when describing scalar change in the physical world, is that they describe ‘change by inner predisposition’ or change that is only possible because of the inherent propensity of specific entities to change. This claim builds on, and at the same time departs from, Levin & Rappaport Hovav’s (1995) proposal that internally caused eventualities are triggered by properties of the entities undergoing the change. First, the said properties are qualified as a propensity, which can be genetic programming or a tendency towards reacting to specific triggers. We believe that this addresses the objection that externally caused change can also be enabled by properties of the undergoer, as with the flexibility of certain metals enabling bending or the friability of certain rocks enabling crumbling (Rodríguez 2011: 245). Second, in light of the differences between VICCs and VECCs (§ 3.3), our proposal is not formulated in terms of causation. ¹⁷

Having introduced the notion of change by inner predisposition, we can reconsider the mismatches between the intransitive and transitive meanings of the [+tr] verbs that we referred to as group (b) in Section 3.2.1 (‘flower’, ‘wither’, ‘rot’, alongside the two ‘burn’ verbs). Exception being made for a handful of examples of Spanish ‘wither’ and ‘rot’, the transitives of these verbs do not describe changes for which specific entities are predisposed: vegetables are not genetically programmed to be softened by cooking, buildings do not have a propensity towards being embellished with flowers, and so on. Thus, these are verbs of inner predisposition when intransitive, whereas most of their transitive occurrences describe external causation. Because of this polysemy, these verbs require dual classification as verbs of inner predisposition and external causation (§2). In this perspective, the verbs whose meanings do not change in accordance with transitivity, including those that allow [+human] transitive subjects (group a of Section 3.2.1), do not require dual classification.

4.2 Change by inner predisposition and periphrastic causation

The conceptual shift to change by inner predisposition is supported by the observation that in our dataset ‘let’ causatives were prevalent with French and Spanish ‘ferment’, ‘age’, ‘rot’, ‘wither’ and ‘mould-v’, as well as French ‘rust’.

We did not find equally robust evidence of ‘let’ causatives in Italian, in accordance with Floricic’s (2022) claim that Italian periphrastic causatives tend to generalize fare ‘make’, in contrast with their French counterparts, which alternate faire ‘make’ with laisser ‘let’ depending on the meaning of the embedded verb (see also Cerbasi 1998).

These results can be understood with reference to Donazzan, Raffy & von Heusinger’s (2020) treatment of the French laisser ‘let’ causative. The authors analyse this construction in terms of a property of the causee, which is a disposition towards the behaviour described by the embedded verb, and a property of the causer, which is a disposition to prevent the causee from following their tendency. The laisser causative is felicitous if the causer refrains from stopping the causee, whilst being understood to be able to do so. In the case of the VICCs, the disposition of the causee is the predisposition of specific entities to react to specific external triggers.

‘Blossom’ and ‘flower’ verbs hardly occurred in ‘let’ causatives and were instead found in non-literal or metonymic ‘make’ causatives. A non-literal example is given here.

These findings suggest that the two verbs resist causation, presumably because they describe processes which specific entities are genetically programmed to undergo. An anonymous reviewer provided evidence of ‘make’ causatives with literal ‘blossom’ and ‘flower’, an example being Italian Come far fiorire l’orchidea? ‘How can one enable (lit. make) orchids to flower?’ (http//www.ideegreen.it). We note that the ‘make’ “causer” of this example cannot be understood to be the effector of the ‘flowering’ change. Instead, it must be construed as a potential facilitator of the ambient conditions which the undergoer is programmed to respond to. Therefore, not only does periphrastic causation differentiate ‘blossom’ and ‘flower’ from the other verbs of inner predisposition, which lend themselves to embedding in ‘let’ causatives, but also from the VECCs, which, when embedded in the ‘how to make’ structure, admit a single-cause and a two-cause reading; in the single-cause reading, the ‘make’ causer is the effector of the change (Italian Come far rompere il guscio? ‘How can one break the shell / make someone else break the shell?’).

4.3 Classes of verbs of change by inner predisposition

Factoring out the external causation meanings of the verbs that require dual classification (§4.1), we propose the following tripartite classification.

Prime examples of class (i) are ‘blossom’ and ‘flower’. Recall that the morphosyntax of these verbs offers no evidence for a causative analysis, in that they do not exhibit a propensity towards transitivization, passivization, or se intransitivization, which are key behavioural properties of external causation verbs (see (3)–(4) and Figures 5–6). Periphrastic causation also sets ‘blossom’ and ‘flower’ apart from other verbs of inner predisposition and from VECCs (§4.2). We thus assume that these verbs are not causative. With respect to ‘blossom’ verbs, considering that they describe events that can be extended but include a specific result state (§ 3.2.3), we represent them as non-causative verbs of quantized change.

Elaborating Van Valin’s (2023: 94–102) lexical decomposition system, ‘blossom’ is broken down into a process (PROC becoming.higher.on.[blossom]scale´ (x)), or the progression of a theme along the scale of becoming blossomed, and a specific final result (FIN blossomed´), which is reached with the culmination of the process (the symbol ∧ indicates such concomitance). In this representation, the capitalized operators identify the Aktionsart class which the predicate belongs to (see note xiii), whereas the parts in bold combine grammatically relevant and idiosyncratic aspects of meaning: the one part indicates that the relevant scale is one of becoming blossomed, the other that the specific result state is that of being blossomed. We shall henceforth refer to the parts in bold of the lexical representations as ‘the content’. It is the [blossom] component of the content that makes ‘blossom’ a verb of inner predisposition because it is this component that selects specific undergoers.

Although Italian ‘rot’ occurs in the left-hand column of Table 2, alongside ‘blossom’ and ‘flower’, it does not squarely fit the characterization of class (i) because 1% of its attestations were passive (§3.2.2). It is possible that this verb is a member of another class for some speakers or is being reanalysed as such, an issue to which we return below. Lastly, intransitive ‘burn_i’ belongs to class (i) insofar as it is not causative. However, the fact that it describes emission, and that it provided no evidence of resultativity, sets it apart from the other verbs under investigation. In fact, following Rappaport Hovav (2020), it should not be included in the verb class under investigation here.

Examples of class (ii) are ‘ferment’ and ‘age’. Although these verbs can take human transitive subjects, an agentive analysis is challenged by the fact that the ‘age’ verbs were also attested with [-animate] transitive subjects, and the ‘ferment’ ones with [-human] ones (cf. 17–18). In addition, when these verbs are embedded in causative periphrases, they can be understood as mono-argumental, which is in fact the most natural reading in (70a–b) and is the only possible reading when the causer of the periphrasis is [-human] (cf. 71).

In contrast, when embedded in a causative periphrasis, non-causative agentive verbs only admit a reading whereby an agent that is different from the causer is understood to be their higher argument.

The agentive analysis of class (ii) is thus to be ruled out. The quasi-synonymity of these verbs with their periphrastic causative counterparts (cf. 70a) is a hallmark of causative behaviour (Zribi-Hertz 1987: 26–27), which, together with the transitivization and passivization facts (see Figures 1, 2, 5), supports a causative analysis. At the same time, since the periphrastic causative can be understood to have a single causer, the fare causer, and indeed this is the most natural reading of (70a–b), we propose that the verbs of class (ii) are not simply causative, but rather stored as non-causative/causative lexical pairs. This analysis follows the framework adopted here (Van Valin 2005: 46, 2023: 95), which takes monoradical process/accomplishment pairs or state/process/accomplishment clusters to be stored as derivationally related members of complex lexical entries. ¹⁸ The lexico-semantic representations for ‘ferment’ are given here.

These representations assume that the ‘ferment’ verbs are degree achievement/causative degree achievement pairs (§3.2.3), thus encoding the progress of a theme along the inner predisposition scale [ferment] without encoding a specific result state (Van Valin 2023: 98). The representation of the causer as an argument of [do´(x,… )] in (74) indicates that this argument is an effector (Van Valin & Wilkins 1996) rather than an agent. The symbol Ø in the square brackets indicates that the lexicon does not specify exactly how the cause brings about the change.

Italian and Spanish ‘germinate’ patterned with ‘ferment’ and ‘age’, while French ‘germinate’ was classified as [-tr] (§3.2.1). The mismatches in the behaviour of cognate verbs suggest that verbs may oscillate between non-causative and causative realizations in different varieties or over time (see also the case of Italian ‘rot’). This hypothesis is further supported by the observation that the meaning of some transitives differs from that of the intransitive counterparts (group (b) of Section 3.2.1). Assuming the transitive can derive by labile causativization, the evidence suggests that this derivation may occur diachronically alongside idiosyncratic meaning changes. The same changes need not affect the cognate verbs.

Let us now compare the verbs of class (ii) with ‘mould-v.’, ‘rust’, alongside inner predisposition ‘wither’ and ‘rot’ (in Spanish), which we take to be members of class (iii). These verbs can also be paraphrased with periphrastic causatives. In fact, since they only take inanimate subjects, qua inner predisposition verbs, the single cause reading is the only possible one.

We thus propose that class (iii) verbs are also stored as non-causative/causative pairs in the lexicon and, in light of the results discussed in Section 3.2.3, we represent ‘rust’ as an alternating degree achievement.

Having discussed how classes (ii) and (iii) are similar, and how they differ from class (i), we now turn to their differences.

4.3.1 Class (ii) vis-à-vis class (iii)

In morphosyntactic terms, class (ii) transitivizes and passivizes but rejects se (we shall return to Spanish ‘ferment’). In contrast, class (iii) does not give as much evidence of transitivization and passivization but exhibits the ±se variation. Starting from the consideration that class (ii) verbs admit human causers, while class (iii) verbs do not, as inner predisposition verbs, the higher propensity towards transitivization and passivization (Hopper & Thompson 1980) of the former class could be explained by the agentivity inferences of the animate manipulator (Van Valin & Wilkins 1996).

The question that arises is why only class (ii) verbs allow human causers. We suggest that the idiosyncratic component of the content of class (ii) verbs ([ferment] in 73–74) is understood in terms of the process in the scale of change and, thus, lends itself to manner inferences. The idiosyncratic component of class (iii) verbs is construed in terms of the result ([rust] in 76–77) and is not as compatible with such inferences. We have two pieces of evidence in support of this hypothesis. First, the class (ii) verbs are more readily embedded in the frame take x time verbing than the class (iii) ones. This test is cited by Beavers & Koontz-Garboden (2012: 343–349) as a manner diagnostic because of the association of manner with duration and complexity. Compare this wine takes ten years to age with #this bike takes ten years to rust. There is no grammaticality contrast, but the different felicitousness of the examples depends on encyclopaedic knowledge about human attitudes towards the relevant processes. Humans normally facilitate the aging of wine but not the rusting of bikes. Second, Italian fermentare ‘ferment’ stood out for the alternation of the two perfect auxiliaries (Appendix 2), the possibility of selection of ‘have’ being suggestive of construals that bring duration and manner to the fore (see Section 3.2.3 for relevant references). ‘Age’ and ‘germinate’ did not exhibit a similar alternation, and, indeed, we do not analyse class (ii) as manner verbs. Instead, we claim that they lend themselves to manner inferences because of a key component of their content.

Verbs that have a manner component in their meaning tend to select human transitive actors (Van Valin & Wilkins 1996; Beavers & Koontz-Garboden 2012). Similarly, the class (ii) verbs are readily compatible with human causers: the relevant change can be acted upon by humans providing the necessary triggers for it to obtain.

Instead, the idiosyncratic meaning of class (iii) verbs is construed in terms of the result and is not as easily combined with causation chains of the type described above. The relevant verbs select inanimate causes, which are the kinds of trigger that can interact with the selected undergoer in the way that yields the result encoded in the content. In this perspective, the class (iii) verbs are true verbs of internal causation because their causes are licensed in terms of the result in the content. The incompatibility of inanimates with inferences of agentivity results in low transitivity (§3.2.1). Our hypothesis on the salience of the result in the content of these verbs also explains the occurrence of these verbs in passive structures that describe the result of a change rather than an event of change (§3.2.2). ¹⁹

Recall now that we take se anticausativization to be a strategy of realization of the causer (§§2, 3.2.3). We propose that the different type of causation described by the verbs of the two classes is a factor in se anticausativization. In accordance with Wolff’s (2003: 4–5) no-intervening-cause criterion, which spells out the conceptual–perceptual conditions to distinguish direct from indirect causation, direct causation obtains if:

Our proposal is that se anticausativization is sensitive to the conditions in (78a–b). Compare ‘ferment’ with ‘break’ changes. Although an animate causer of the latter type of change can avail herself of an instrument (e.g. she broke the door with a hammer), the participation of the instrument in the event satisfies the intention of the causer, and the instrument cannot effect the change by itself. Therefore, the instrument is an enabler and the link between the initial causer and the causee is construed as direct. ‘Ferment’ changes are different, not only because they cannot be effected by a human cause without specific triggers, but also because such triggers are not instruments, since they can bring about the change by themselves (as with lactobacilli fermenting milk). They are thus causers, not enablers, which means that condition (78b) is violated. In contrast, the cause of class (iii) verbs does not violate either of the conditions on direct causation and can be marked by se as an inherent facet of the event of change (§2, 3.2.3, 5). In Section 5.2, we will reformulate this proposal as a constructional constraint on se anticausativization.

5. Se anticausativization: a linking account

5.1 The linking

The framework adopted here (Van Valin and LaPolla 1997; Van Valin 2005, 2023) assumes a direct mapping between syntactic and semantic representation, called linking, which is an idealization of what speaker and hearer do in actual communication. The speaker’s perspective is represented in the semantics to syntax linking, whereas the hearer’s perspective is represented in syntax to semantics linking. It would be beyond the scope of this article to discuss the universal and language specific steps of the algorithm which governs the linking. We will, instead, briefly go over the steps in the semantics–syntax linking which are relevant to se anticausativization. The first phase, called ‘lexical’, involves the construction of the semantic representation of the clause, starting from the semantic representation, or logical structure, of the predicate. The latter is usually a verb, which is retrieved from the lexicon, where it is stored as the mapping of a form with a logical structure. The second step involves the assignment of values to the argument variables figuring in the logical structure of the predicate. Actor or undergoer assignment follows (cf. 79), a key step in the interface with syntax, which for brevity will not be discussed here. The morphosyntactic coding of argument–predicate relations abides by alignment principles: in accusative alignment the default controller of verb agreement is the highest-ranking argument in logical structure.

5.2 Se anticausativization

Building on Centineo (1995), Van Valin & LaPolla (1997: 416–418), and Bentley (2006: 126–136), we analyse se anticausativization as an operation which occurs in the lexical phase of the linking and concerns the realization of the highest argument in the causative logical structure, [do´ (x,…)] in (77). The causer position is saturated with a silent argument, indicated with Ø in [do´ (Ø,…)] in (79). This operation blocks the projection of the causer as an argument of the verb, although se marking will signal that a cause is a facet of the event. Since the highest position is unavailable for morphosyntactic purposes, the controller of person–verb agreement will be linked from the lower argument position.

This analysis of se anticausativization can capture the syncretism of anticausative, passive, reflexive, and impersonal se (§2), assuming that these are variations of a single operation of suppression of the highest-ranking argument in the lexical phase of the linking (see Bentley 2006 for Italian). Treating anticausative se as the spellout of an operation on the causative member of a non-causative/causative pair, it is also possible to explain key aspects of the ±se variation with the verbs of class (iii) and the absence of se on the intransitive of the verbs of class (ii).

The -se attestations of both classes are realizations of the non-causative logical structure (see 73 for ‘ferment’ and 76 for ‘rust’). This is also the lexical input when a durative or inceptive reading is coerced (cf. 57, 60, 62). In such cases, the process component of the non-causative logical structure (PROC becoming.higher.on.[α]scale´ (x)) will alone be grammatically relevant because of the aspectual inferences that arise from the context.

Turning now to the failure of the class-(ii) verbs to anticausativize, recall that we proposed that se anticausativization is sensitive to the conditions in (78). In the terms proposed here, a complex causation chain is not normally realized as a silent argument in the position [do´ (x,…)]. In turn, the verbs that admit such complex causes do not normally participate in se anticausativization, even though a priori they could (see the case of Spanish ‘ferment’).

The comparison of the class (ii) verbs with external causation verbs which are known only to exhibit -se intransitives in Italian (e.g. guarire ‘heal’; affondare ‘sink’; cambiare ‘change’; migliorare ‘improve’ and peggiorare ‘worsen’; aumentare ‘increase’ and diminuire ‘decrease’) provides relevant independent evidence, suggesting that, in this language, a constructional constraint is at work beyond the specific case of inner predisposition verbs. Indeed, the -se verbs listed above also describe complex causation events with potential intervening causes between the initial causer and the final causee (as with doctors healing patients by prescribing drugs or lifestyle changes).

The framework adopted here conceives of the architecture if grammar in terms of cross-linguistically and cross-constructionally valid linking principles (§5.1), which are complemented by constructional instructions stored in language-specific inventories of C(onstructional) S(chemas). We thus propose that in Italian se anticausativization is subject to a construction-specific instruction which discerns between different types of cause. Specifically, the CS of se anticausativization in Italian will include an instruction stating: <se anticausativize iff conditions on direct causation are satisfied>.

We acknowledge that other constraints must be at work when considering se anticausativization more broadly. In cross-linguistic terms, the -se classes do not exactly coincide. In French, the incidence of anticausative se is known to be historically low, albeit on the rise diachronically (Heidinger 2014; Cennamo 2021). It was very low in our sample, exception being made for faner ‘wither’. In Spanish, the incidence of se was generally higher and the morpheme was attested with two verbs that do not take se in the sister languages: fermentar ‘ferment’ (though see note xiv) and pudrir (which took [-animate] transitive subjects when meaning ‘rust’ and ‘rot’ in an inner predisposition sense, thus being unproblematic for our proposal). The development and distribution of se in these languages has been discussed in term of aspectual criteria (Zribi-Hertz 1987; Cennamo 2021; Vivanco 2021) and attempts have been made to reconcile the argument realization approach with the aspectual one (Jiménez-Fernandez & Tubino-Blanco 2017 on Southern Peninsular Spanish). Our claim is, therefore, that although the comparison of the sister languages suggests that different language-specific factors, including aspectual ones (§3.2.3), are at work in the distribution and frequency of se, the evidence of inner predisposition verbs supports the analysis of se as an argument realization strategy, in that the type of cause which different subclasses of verbs are compatible with has a bearing on se marking of the intransitive. The other relevant criteria should be factored into this broad understanding of se.

6. Conclusion

Verbs of internally caused change were first characterized by Levin & Rappaport Hovav (1995) as monovalent verbs whose argument has an inherent property that is responsible for bringing about the eventuality of change. Subsequently, the same verbs were claimed to take inanimate causes or agent subjects (Alexiadou 2014), or to be causative verbs with a lower propensity towards transitivization than the known verbs of external causation (Rappaport Hovav 2020).

Availing ourselves of corpus evidence on a sample of putative verbs of internally caused change in Italian, French, and Spanish, we have proposed that the key property of the verbs in question is a predisposition of the undergoer: genetic programming for a given change or a propensity to react to specific triggers in specific ways. Building on Van Valin’s (2005, 2023) system of lexical decomposition, we have proposed a threefold classification of verbs of inner predisposition in Romance, with one class being non-causative and the other two alternating as causative/non-causative pairs in the lexicon. A key observable difference between the two alternating classes is that the one takes [±human] transitive subjects, whilst the other takes [-animate] transitive subjects. We have suggested that the content of the former class lends itself to manner inferences and, hence, to the description of complex causation chains whereby human instigators manipulate the process availing themselves of inanimate triggers. In contrast, the content of the latter class is construed in terms of the result. These verbs do not lend themselves to the description of complex causation chains and select specific causes that interact with specific types of undergoer to yield the result that is salient in the content. Importantly, this was the only subclass that was found to se anticausativize. While the existing literature has placed emphasis on the aspectual conditions on the ±se variation, we noted that the types of cause which verbs are compatible with play a role in the licensing of se anticausativization. With specific reference to Italian, we proposed a construction-specific instruction which constrains se anticausativization to cases which fully satisfy Wolff’s (2003) conditions on direct causation.

The results discussed in this study indicate that inner predisposition is a property of the content of verbs, which has a bearing on the selection of the undergoer, and, in two of the subclasses, the cause. In other respects, verbs of inner predisposition are indistinguishable from other verbs of change.