Introduction to Lexical constraints in grammar: Minority verb classes and restricted alternations

1 Introduction

In many languages, grammatical relations are subject to lexical constraints; that is, “certain lexical verbs or verb classes take different argument coding than others” (Van Lier and Messerschmidt 2022, 1). Lexical constraints can be manifested in different morphosyntactic domains, for instance, through deviation from canonical case frames (e.g. Aikhenvald et al. 2001, Bickel et al. 2014, Malchukov and Comrie 2015, Tsunoda 1985) or different argument indexation patterns (e.g. Fedden et al. 2013, 2014, Bickel et al. 2015). Other constructions that have been studied through this lens are voice and valency constructions (e.g. Polinsky 2013, Vigus 2018, Olthof et al. 2020, Van Lier and Messerschmidt 2022) and some clause-combining constructions (Van Gijn 2011). The types of oppositions established by lexical constraints vary: some absolute restrictions entail the obligatory presence or absence of a grammatical marker, while others entail the ability of a lexical item to alternate. In the latter instance, differences in the statistical preferences for one construction over another may be observed. In some cases, verb classes can be easily identified based on a common semantic feature; however, various other factors can also lead to the formation of minority verb classes and restrictions on alternations.

This collection of four articles investigates lexical constraints in a variety of morphosyntactic domains, adopting different perspectives and methodologies. Two single-language studies take a synchronic perspective, one on voice alternations and differential agent marking in Kanakanavu (Austronesian, Taiwan; Cheng 2023) and the other on fluidity in argument indexing in Komnzo (Yam, Papua New Guinea; Döhler 2023). The third article adopts a diachronic perspective on the passive alternation in Danish (Indo-European, Denmark; Gregersen 2023). The fourth article considers the role of language-contact phenomena by comparing alignment systems in two neighbouring language families – Quechuan and Tukanoan – in the northern Upper Amazon in north-western South America (Van Gijn et al. 2023).

In this introductory article, we first describe the relevance of minority verb classes and restricted alternations for typological research (Section 2). Following this, Section 3 illustrates some common domains in which lexical constraints play a role, and Section 4 sets out a framework for defining different kinds of oppositions that result from lexical constraints in various morphosyntactic domains. Section 5 outlines a number of different factors that motivate lexical classes, followed by a brief summary in Section 6.

2 Lexical constraints in typology

In classic Greenbergian typology, languages are assigned a single value for a particular feature. For instance, a language has SOV (subject, object, verb) or SVO word order, or it has Nominative-Accusative or Ergative-Absolutive alignment. It is increasingly recognised that such coarse-grained labels – while they have led to many useful insights – mask a great deal of language-internal variation. The answer to this has been a move from type-based to token-based typology (Levshina 2019; see also Bickel 2010, 2015, Schnell and Schiborr 2022). This turn towards language-internal variation, enabled by recent methodological developments, allows us to consider a particular domain at any level of specification, right down to the behaviour of a particular lexical item.

To illustrate the various levels, we consider word order in German (Indo-European). At the level of language types, in the WALS datasets (Dryer and Haspelmath 2013), German is labelled as having no single dominant word order in main clauses; more specifically, it has two ‘primary alternating orders’: SOV and SVO (Dryer 2013). This label is assigned based on the fact that the position of the verb in main clauses depends on the verb form, as shown in (1): when the verb is simple (i.e. the lexical verb is inflected and there is no auxiliary), word order is predominantly SVO, as in (1)a, but when the inflected verb is an auxiliary, the order is SOV (i.e. SAuxOV), as in (1)b (Dryer 2013).

Taking into account only dominant word orders means leaving aside the fact that German is in fact quite flexible. Insubordinate and interrogative clauses exhibit VSO order, and a considerable literature exists on the triggers of OVS order, which include given/new status, definiteness, length, and animacy of the arguments (see e.g. Heylen and Speelman 2003, Weber and Müller 2004). To provide further detail to the picture, there is another syntactic constraint in German: subordinate clauses are predominantly verb final (i.e. SOV or SOVAux for verb forms with inflected auxiliaries). But, as shown in (2), this is not absolute and, in some contexts, SVO order is also possible.

It is not the case, however, that all kinds of subordinate clauses exhibit the same alternation. At the level of individual lexemes, we find that word-order variation in these clauses depends on the subordinating conjunction used. It has been noted since the 1970s (see e.g. Farrar 1999 and references therein) that subordinate clauses with weil ‘because’ are undergoing a diachronic shift, being increasingly attested with SVO order in spoken German. Although other subordinating conjunctions – such as obwohl ‘although’ and während ‘while’ (Farrar 1999, Günthner 1996) – are also claimed to show the SOV/SVO alternation, it is weil-clauses where SVO order ‘is most evident’ (Farrer 1999, 1). Indeed, the German dictionary Duden admits only for weil that both orders are possible, calling SVO order in weil-clauses ‘non-standard’.^[2]

Various proposals have been made for why weil should behave differently from other subordinating conjunctions (see e.g. Farrar 1999, Antomo and Steinbach 2010). More recently, Kempen and Harbusch (2016) proposed a cognitive explanation based on data from a quantitative corpus study. To give a brief summary, because weil is highly frequent, speakers select it before they have planned the rest of the clause. This leads to them uttering weil but continuing their utterance as a main clause, and hence to producing the dominant main-clause word order, SVO, instead of the expected subordinate clause order, SOV. Thus, considering lexical constraints in German word-order variation leads to insights into the synchronic functioning of the language, as well as into the mechanisms implicated in diachronic change. It shows that constraints associated with a specific lexeme are relevant in understanding changes in broader domains. Also, given that SVO order in weil-clauses is most prevalent in colloquial spoken German, this example highlights the importance of differentiating between spoken and written language and the relevance of drawing on the corpora of spoken language.

3 Domains

Beyond word order, lexical constraints may play a role in any grammatical domain. One such domain is that of argument coding, which includes ‘flagging’ (case/adpositional marking on arguments) and ‘indexing’ (marking on verbs that indexes features of an argument such as person and number) (Haspelmath 2013). Alignment typology considers flagging and indexing patterns but, like word-order typology, generally focuses only on the dominant patterns. And, like word order, dominant alignment patterns may be subject to lexical constraints. For instance, Chechen (Nakh-Daghestanian) is considered to have ergative-absolutive alignment, as illustrated in (3)a.^[3] This is expressed by both flagging and indexing: the transitive subject (A argument) k’ant ‘boy’ has an ergative case-marker, and the transitive object (P argument) quor ‘pear’ is in the unmarked absolutive case and is indexed on the verb with a B-gender prefix. However, a small group of verbs of cognition and perception (i.e. ‘experiential’ or ‘psychological’ predicates; see Haspelmath 2001, 59) mark the A argument with the dative case rather than the ergative, as shown in (3)b.

The type of lexical restriction illustrated by Chechen is often described as ‘non-canonical case marking’; that is, non-prototypical verbs such as experiential predicates occur with a marking pattern that deviates from a canonical pattern used with prototypical action verbs (Croft 1991, 212–213, Haspelmath 2001, 59). In such systems, there is no alternation for a given verb; that is, go ‘see’ in Chechen always takes a dative-marked A argument. Chechen illustrates a similar phenomenon for argument indexing: go ‘see’ never takes a gender-agreement prefix, while u’u ‘eat’ obligatorily occurs with one (Komen et al. 2021, 343). Such lexical restrictions in indexing systems have been termed ‘sporadic agreement’ (Corbett 2006, 17; Fedden 2019).

Flagging and indexing systems in which different coding patterns alternate for a given verb are termed differential argument marking (DAM). A number of factors have been reported to play a role in different languages, including various semantic and/or pragmatic factors (for an overview, see Witzlack-Makarevich and Seržant 2018). Lexical restrictions are not generally considered to be a trigger for DAM, but they can nevertheless obtain in DAM systems. An example is provided by Kamang (Alor-Pantar), which has differential P indexing.^[5] In (4)a, the verb faafa ‘search for’ indexes the P argument with the prefix ga-, while the same verb in (4)b bears no index. The alternation in (4)a–b is due to the animacy of P; however, in (4)c–d, the same prefix is present on the verb, despite the different animacy values of P. In fact, Kamang has numerous verb classes that behave differently with respect to indexing: faafa ‘search for’ belongs to a small group of verbs that alternate according to the animacy of P, while tak ‘see’ belongs to another group that always indexes P regardless of animacy (see Fedden et al. 2013, 2014, Walker et al. 2023).^[6]

Lexical constraints are also common in voice and valency-changing constructions. A particular strategy for manipulating the voice or valency of a verb may productively apply across a large part of the verbal lexicon, or it may be restricted to certain verbs or verb classes. For instance, Kalamang (Greater West Bomberai) has a causative proclitic di= that occurs only on directional verbs (except era ‘to move up diagonally’), as shown in (5)a. On any other verb, di= is ungrammatical, and a biclausal strategy must be used instead, as shown for the verb ‘come’ in (5)b–c (Visser 2022, 275–7).

4 Opposition types

When comparing the behaviour of subgroups (e.g. verb classes) in any grammatical domain, different kinds of oppositions are revealed. The most straightforward case of opposition is when there is no overlap between the behaviour of different classes: subgroup X always behaves in one way, and subgroup Y always behaves in another. In the case of argument marking variation, this means that members of a verb class obligatorily occur in a single construction. Case-marking of A arguments in Chechen, as illustrated above, exemplifies this kind of fixed opposition: (some) experiencer verbs obligatorily occur with a dative-marked A, while most other verbs obligatorily mark A with the ergative.^[7]

In Chechen, the opposition is between two overt case markers, which can be considered a ‘symmetrical’ system. Where the opposition is between an overt marker and zero, the system may be labelled ‘asymmetrical’ (Iemmolo 2013). An asymmetrical system is exemplified by Mian (Trans New Guinea). Most transitive verbs in Mian index only the A argument with a suffix and do not index P at all, as shown in (6)a. This contrasts with a small number of verbs that require the P argument to be indexed with a prefix, as shown in (6)b–d. However, the class of P-indexing verbs itself contains a symmetrical opposition: a group of seven verbs indexes the person/number of P, as shown in (6)b, while a group of approximately 35 verbs indexes P using a variable classifier prefix, as shown in (6)c–d (Fedden 2011, 185, 260). Thus, Mian has an asymmetrical opposition between verbs that do not index P and those that do, and a symmetrical opposition between the P-indexing verbs, giving a total of three verb classes.

Oppositions that are less straightforward involve a many-to-one correspondence between subgroups and behavioural possibilities. In other words, contrary to the fixed oppositions above, more than one behaviour is available for a given subgroup; hence, members of a verb class can alternate between two or more constructions. There may also be some overlap such that multiple subgroups share some options. Alternations, like fixed oppositions, can also be symmetrical (alternation between two different markers) or asymmetrical (alternation between an overt marker and zero). For instance, faafa ‘search for’ in Kamang (example (4)) shows an asymmetrical alternation between an overt P prefix and zero indexing.

Alternations vary in terms of their predictability. Following literature on DAM, alternants in a ‘split’ system are in ‘obligatory complementary distribution’, while a fluid (or optional) alternation ‘works solely according to probabilistic rules’ (Witzlack-Makarevich and Seržant 2018, 28). That is, alternations in a split system are absolute in the sense that the factors conditioning the alternation provide clearly delineated contexts of use. The occurrence of an alternant in a fluid system is a matter of (statistical) preference. Much quantitative research into these statistical preferences exists for well-known alternations such as the English dative alternation (e.g. I give you the book vs I give the book to you; Wasow 2002, Gries and Stefanowitsch 2004, Bresnan et al. 2007, Perek 2015, Diessel 2020, among others), but the increasing availability of corpora has allowed fluid alternations to be investigated in an ever-wider array of languages for various syntactic alternation phenomena (e.g. McDonnell 2016, Faghiri 2016, Riesberg et al. 2022, Just and Witzlack-Makarevich 2022, Walker et al. 2023).

Split and fluid alternations are frequently subject to constraints of varying types, including lexical. This results in oppositions between classes that do alternate and those that do not, or between classes with different types of alternations. For instance, S flagging (case marking of the single argument of intransitives) in Hindu/Urdu (Indo-European) is a system that has an opposition between a fixed zero class and an alternating class. Most intransitive verbs are ‘fixed zero’, which obligatorily take nominative (i.e. unmarked) S arguments. Alternating verbs comprise a small group of ‘bodily emission’ verbs (e.g. ‘cough’, ‘scream’) that can occur either with an unmarked S or an ergative-marked S to signal a distinction in volitionality, as reflected in the translations in (7).

Languages may have multiple classes, each exhibiting different types of alternations, obligatory markers, or zero. Kamang, introduced above, is one such language. It has a lexically conditioned split between verbs that obligatorily index P (see (4)c–d) and verbs that alternate according to the animacy of P (shown in (4)a–b). In addition, some verbs never index P, such as bo'na ‘hit’ in (8)a. Kamang also has further subdivisions within both the ‘obligatorily present’ and ‘alternating’ categories, since several prefix paradigms are available. Example (4) illustrates the obligatory presence and alternation of the /a/-series prefix; example (8)b–d demonstrates the same categories for the /e/-series prefix. For instance, n ‘marry’ obligatorily indexes the P argument with an /e/-series prefix in (8)b. In (8)c–d, the /e/-series prefix indexes P on the verb beta ‘bump into’, which alternates with zero indexing in order to express a difference in affectedness (Schapper 2014, 325–6, Walker forthcoming).

An example of fluid alternation from outside the domain of flagging and indexing is light-verb alternation in light verb constructions in Persian (Indo-European). Light verb constructions are prevalent in the language and consist of a light verb preceded by a non-verbal element, as in bāzi kardan ‘play’ (lit. ‘game do’). The opposition in Persian is between predicates that obligatorily occur with a particular light verb, such as harf zadan/*kardan ‘talk’ (lit. ‘talking hit’) vs sohbat kardan/*zadan ‘talk’ (lit. ‘talking do’), and those that alternate (with no clear difference in meaning), such as telefon kardan/zadan ‘call by phone’ (lit. ‘phone do/hit’) and imeyl kardan/zadan ‘contact by email’ (lit. ‘email do/hit’). Such light-verb alternation is fluid, in that the two constructions are not in complementary distribution (see e.g. Samvelian and Faghiri 2013).^[8] However, alternating predicates do show (statistical) preferences and can be further subdivided into classes according to which is the preferred light verb.^[9] For instance, while both atse kardan (lit. ‘sneeze do’) and atse zadan (lit. ‘sneeze hit’) can be used (in the same context) to mean ‘sneeze’, there is a clear preference in usage for atse kardan. The opposite preference is shown by the predicate ‘be (freezing) cold’: yax zadan (lit. ‘ice hit’) is the preferred construction, though yax kardan (lit. ‘ice do’) is also possible (Faghiri et al. submitted).

5 Motivating factors

The previous section described how lexical classes can be established on the grounds of opposition types (fixed/alternating) and further subdivided according to the form or presence of a marker (symmetrical/asymmetrical). Here, we outline the different features that members of a lexical class have in common, as well as if or how these features motivate the marking behaviour of that class.

Several of the contributions in this special issue describe lexical constraints based on semantic classes. In Döhler’s (2023) contribution, lexical semantics are found to motivate the opposition in S-indexing in Komnzo: verbs that are unspecified for agentivity have a greater propensity to alternate compared to verbs whose arguments are specified for high or low agentivity. In the North-West Amazon region that is the subject of Van Gijn and colleagues’ (2023) contribution, the P of motion verbs and A of sensation predicates show the most ‘deviant’ flagging. As illustrated above by dative-experiencer constructions in Chechen, deviant flagging (or non-canonical case marking) of experiencer arguments occurs frequently cross-linguistically (e.g. Malchukov 2008, Seržant and Kulikov 2013).

Verb classes defined semantically frequently leak to varying degrees. In Kalamang, shown above in (5), there are two classes in regard to the use of causativizer di=: it is permitted only with directional verbs; other verbs require different causativization strategies. However, one directional verb (era ‘to move up diagonally’) exhibits the same behaviour as non-directionals and may not be causativized with di= (Visser 2022, 260). Leaks may go in both directions. For instance, Mian verbs that take classifier prefixes tend to be verbs of handling or movement, yet some handling verbs do not take classifier prefixes (e.g. mengge ‘pull’) and some verbs take classifiers that are not verbs of handling or movement, such as halin ‘worry’ and suan ‘hate’ (Fedden 2019, 312). Equally, Mian verbs that index the person/number of the object tend to be high on the transitivity scale (Hopper and Thompson 1980); however, low transitivity verbs like têm’ ‘see’ and temê’ ‘look at’ also index their objects, and highly transitive verbs like klutaka ‘smash’ do not (Fedden 2019, 314).

Verb classes may exhibit different markings for reasons of phonotactics. In many Nakh-Daghestanian languages, some verbs obligatorily occur with a prefix indexing the gender of the absolutive argument. All other verbs never occur with a gender prefix. This is illustrated for Chechen in (3), repeated below in (9): in (9)a, the verb u’u ‘eat’ is prefixed with a marker of ‘B’ gender, while in (3)b, go ‘see’ is not prefixed with a gender marker. In almost all cases, verbs with an initial consonant are prohibited from taking a gender prefix, while vowel-initial verbs obligatorily take a prefix (Komen et al. 2021). However, as shown for the verb olxu ‘comb’ in (9)c, some verbs do not take a gender prefix even when they are vowel initial. This is the same situation in a related language, Tsez, which has at least ten vowel-initial verbs that do not agree (Polinsky and Comrie 1999, 111). It is speculated that gender prefixes are blocked on these Tsez verbs due to ‘the presence of an underlying laryngeal’ (Fedden 2019, 307) that is not present synchronically.

The situation in Chechen suggests that frequency may be a mechanism through which minority verb classes persist in a language. Although gender-prefixing verbs are a minority in terms of type frequency (approximately 30% of verb types take a gender prefix), these verbs have an overall higher token frequency; that is, they are over-represented in discourse (approximately 50% of verbs in discourse – i.e. verb tokens – take a gender prefix) (Komen et al. 2021, 324). The same pattern but with a sharper difference in proportions – low type frequency (27% of dictionary entries) vs high token frequency (85%) – has also been shown for Tsez (Gagliardi 2012). Fedden postulates that ‘[t]his frequency effect contributes to the learnability and stability of the Tsez agreement system’ (2019, 319). That is, it is only through frequent use that a restricted class in terms of the number of members is maintained. Beyond the phonotactically conditioned prefixing in Nakh-Daghestanian, the same pattern is attested for Mian verb classes, which have a (leaky) semantic basis: 16% of verb types (in the dictionary) have some kind of object agreement, but this rises to 42% of corpus tokens (Fedden 2022, 292–3).

Extra-linguistic factors may also be implicated in the formation of verb classes and the associated lexical constraints. This is the subject of Gregersen’s (2023) contribution on the role of sociolinguistic factors in Danish: some verbs permit an alternation between a periphrastic and inflectional passive, while others permit only the periphrastic construction. Gregersen proposes that the promotion of the inflectional passive by eighteenth-century grammarians contributed to the present-day situation in which verb lexemes more commonly associated with the formal written register permit the inflectional passive, while verb lexemes that might be expected to occur frequently in spoken discourse allow only the more common periphrastic construction. Here, again, we find frequency of use – here in different contexts – playing a role in the formation and maintenance of verb classes.

The interaction of lexical class and construction type is illustrated in Cheng’s (2023) contribution on Kanakanavu. The semantic class of the verb (dynamic or stative) interacts with the construction type (perfective or imperfective) to influence the interpretation of certain features of an argument. For dynamic verbs, agents omitted from perfective clauses have specific reference, while those omitted from imperfective clauses may be generic or backgrounded. For some stative verbs, perfective predicates have a semantically agentive argument, but an experiencer argument in imperfectives. For other stative verbs, perfectivity does not affect the agentivity of the argument.

6 Summary and outlook

This introductory article first discussed lexical constraints in a typological context before drawing on case studies from contributions to this special issue and beyond to illustrate the different grammatical domains that may be affected by lexical constraints. We set out a framework for considering different opposition types formed by the differing behaviour of different verb classes. Following this, we outlined a number of different factors that motivate the formation of verb classes. We showed that lexical constraints provide fertile ground for typologists adopting a token-based approach seeking to compare languages at ever-greater levels of specification.

In addition, studying the exceptions within a grammatical system – the types of exceptions and the frequency with which they occur (in the lexicon and in discourse) – can also provide insights to those interested in, for instance, language cognition (cf. Faghiri et al. submitted for a psycholinguistic study of light-verb alternation in Persian) or child language acquisition (cf. Gagliardi 2012 on the learnability of Tsez). The suggestion that type vs token frequency plays a role in the maintenance of minority verb classes lends itself to quantitative corpus analysis, for which the increasing availability of diverse (spoken) language corpora can be utilised in the future (e.g. MultiCAST: Haig and Schnell 2022; CORPAFROAS: Mettouchi et al. 2010; DoReCo: Seifart et al. 2022). Such studies could address the question of whether frequency effects are always visible in lexically constrained systems and whether cross-linguistic generalisations can be made in terms of the relationship between type and token frequency and the interaction with other factors that motivate verb classes.

However, quantitative corpus studies will not be able to provide all the answers: the corpora available for less-studied languages may not contain enough material to confirm whether a particular verb can appear in a particular construction (cf. e.g. Molochieva et al. 2022, Walker et al. 2023). Research on lexical constraints will therefore need to continue to be a methodologically diverse pursuit, incorporating insights not only from quantitative corpus studies but also from synchronic language descriptions, studies of diachrony, language contact, and sociolinguistic context, and experimental work.