4.1 Phonology

Even if we exclude phonologically conditioned variation in morphological form from our picture of inflection class systems, we still find phonological correlations that fall outside the bounds of morphophonology. Consider the paradigm in Table 4.2 from Sobei, a Western Oceanic (Austronesian) language of New Guinea. Subject prefixes fall into two classes, which more or less evenly split the verbal lexicon (43%:57%). The difference is largely predictable from root phonology: the prefix set characteristic of ‘go’ is used with monosyllabic C-initial verbs, whereas the other set is found with polysyllabic or CC-initial verbs. There is no discernible relationship between the phonological properties of the stems and those of the prefix allomorphs they select: how could the syllable structure of the root affect an alternation in vowel quality? And indeed, as suggested by Sterner (Reference Sterner1987), the probable origin of the distinction is not phonological but, rather, morphological. The ‘go’ type represents monomorphemic verb roots, the ‘hang’ type represents polymorphemic verb roots with traces of a derivational prefix. On this reconstruction, derived verbs would have taken a different set of mood prefixes (the vowel immediately preceding the root) from underived verbs. As a result, the two classes, though arbitrary in one sense, are nevertheless phonologically predictable.

4.2 Morphology

One obvious way that morphology can influence inflection class membership is by defining properties of the stem that determine inflection class membership. Consider aspectual derivation in Russian verbs illustrated in Table 4.3. There are two main processes. One derives perfective verbs from imperfective through prefixation. The prefixes often add some degree of semantic content, but do not affect membership in the two inflection classes, which is based on the behaviour of the suffixes, as well as associated stem alternations. These prefixed perfective verbs can in turn serve as the base for imperfective verbs derived through a suffix -yva-/-iva- (the alternation in the initial vowel is a by-product of the quality of the preceding consonant). Verbs derived with this suffix always belong to the first conjugation. In effect, inflection class membership here is a property of the suffix itself.

Aspects of paradigmatic behaviour can also determine inflection class membership. Consider the noun paradigms in Table 4.4, from Nuer, a Western Nilotic language of South Sudan. Three patterns of suffixation are found in the plural: no suffixation, the suffix -ni in the oblique cases, and the suffix -ni for all cases. These different suffix patterns are related to the pattern of stem alternation. The first two patterns are only possible if there is a singular ~ plural stem alternation (the second pattern is by far the more frequent), whereas the third pattern is obligatory if the stem lacks a number-based alternation (Baerman Reference Baerman2012). The relationship is not entirely deterministic; thus, the third pattern is still possible with nouns that undergo a number-based stem alternation, though it is infrequent. Thus, the suffix classes of the Nuer noun are largely though not completely predictable given the paradigmatic behaviour of stems.

4.3 Semantics

Semantic conditioning presents two interpretative challenges. The first is definitional: since semantic categories may cross the threshold into being features, if a given semantic category reliably correlates with a given inflectional pattern, should it not be incorporated into the underlying feature profile of the paradigm? In that case, we have in effect distinct word classes and not distinct inflection classes. The second challenge comes from the nature of the evidence: sometimes the only overt indication of a semantic contrast is the very morphological contrast we are investigating, so the supposition of semantic predictability can neither be proven nor disproven. We survey here some representative cases.

Morphological variation in nouns in particular often appears to reflect some kind of semantic categorization. For example, nouns in Mali, an East New Britain language of New Guinea, fall into several different classes on the basis of their number marking suffixes. Stebbins (Reference Stebbins2011) illustrates the semantic contrasts by comparing the same root with different series of number markers. Thus in Table 4.5, the root amēng is shared by various words having some connection with wood, differing with respect to such properties as shape and size. (There are two gaps in its distribution in the table: the ‘teacher’ type is reserved for male humans, whereas the failure of amēng to appear in the ‘feather’ pattern is presumably a lexical accident.) The system clearly has some productivity as a derivational device, particularly apparent when we look at borrowings – for example, plastika ‘medium-sized plastic bottle or bag’ ~ plastiki ‘large-sized plastic bottle or bag’ ~ plastikini ‘small-sized plastic bottle or bag’. Still, it is also lexicalized to some extent, subject at least at times to what in synchronic terms can only be arbitrary principles. Thus, the size contrast implicit in plastika vs. plastiki is not enough to account for such pairs as galēp-ka ‘type of fish’ ~ galēp-ki ‘type of palm tree’, or chael-ka ‘wallaby’ ~ chael-ki ‘stingray’ (Stebbins & Tayul Reference Stebbins and Tayul2012).

In the Mali system, the categories go hand-in-hand with particular morphological exponents – that is, on the whole, each number marker is at the same time representative of its class.Footnote ¹ Contrast this with the noun paradigms in Table 4.6 from Diyari, a Pama Nyungan language of Australia. There is a sharp division of inflectional pattern according to semantic principles: common nouns versus women’s personal names versus men’s personal names. But the contrasting noun classes cannot be encapsulated in contrasting inflectional exponents in the same way as in Mali. For example, both male and female personal names share a suffix -nha, but differ in the way it is distributed in the paradigm (absolutive/accusative with male names, accusative, and the base for locative, allative, and dative with female names). Further, the non-singular paradigm of common nouns (here represented by a plural, but the dual is the same, just with a different stem) patterns with personal names: the oblique case suffixes are those of male names, whereas the behaviour of the core cases (contrast of absolutive and accusative) is that of female names. This means there is no inherent association between the morphological exponents and the lexical categories they are associated with. Instead, the mapping is between the lexical categories and the entire paradigm, which means that the notion of inflection class – as a morphological object – is still useful here, however semantically transparent the assignment rules are.

Cross-linguistically, the question of semantically based noun categorization arises most frequently in connection with gender systems. In many languages the morphological marking of gender agreement overlaps with that of noun morphology. And if gender assignment is at least in part semantically motivated – which it is in any gender system (Corbett Reference Corbett1991: 7–8) – it is natural to try to equate any inflection classes displayed by nouns with the categories described by gender values that have the same morphological expression. In other words, to construe inflection classes as direct exponents of the gender feature. We would argue that this conflates two notions that should be kept distinct, but at the same time shows that otherwise arbitrary inflection class systems can harbour a sizeable element of semantic motivation.

Consider the nouns in Table 4.7 from Eegimaa, an Atlantic language (Niger-Congo) of Senegal. They fall into several inflection classes based on their number prefixes. These in turn map in a non-trivial way onto gender values, which are realized through agreement morphology on verbs and various modifiers (Table 4.8). Sagna (Reference Sagna2012) argues that genders I-VI have a definable semantic basis – that is, gender I is for humans, gender II embraces roundness, thickness, and associated notions, and so on. Gender VII is a default category, devoid of specific semantic content. If we take the semantic coherence of the first six genders as a given, then in those cases where there is a one-to-one correspondence between the noun’s inflection class and gender (gender I, and III-VI), the choice of inflection class for a given noun stem could be said to have a semantic basis. But there are still more inflection classes than there are gender values; thus, the inflection-class distinctions within genders II and VII are underdetermined by the semantic principles observable in the gender system, and hence they are arbitrary.

The question of semantic categorization as a component of inflection class also arises with verbs, but because of the nature of verbal paradigms – or at least, our conventional interpretations of them – the issues are somewhat different. The semantic categories involved in inflectional contrasts in verbal paradigms are often ones that implicate the morphosyntactic paradigm, such as argument structure and role marking. The result is categories that line up approximately with syntactic distinctions, but not seamlessly; so it is not clear, at least under a modular conception of language, which component of grammar is responsible.

As an example, consider the inflectional contrast that pervades the verbal system of Rotokas, a language of the North Bougainville family of New Guinea. Table 4.9 shows a representative fragment of the larger paradigm, involving dual and plural subject number and further TAM distinctions. (The first suffix here marks subject person-number, the second marks TAM.) Following Robinson (Reference Robinson2011), we call these the alpha and beta class.

Examples (7)–(10) illustrate the contrast in use between the two inflectional types. In (7) a transitive verb with a direct object inflects according to the beta-class. When the same verb is detransitivized (taking an incorporated object, so that the notional object is now marked as an oblique, with the locative case), it inflects according to the alpha class (8). The intransitive verb ‘go’ in (9) also inflects according to the alpha class. So far, this suggests we interpret the apha-class suffixes as markers of intransitivity, and the beta-class suffixes as markers of transitivity. But we also get verbs like ‘walk’ in (10), which are intransitive but inflect according to the ‘transitive’ class.

Rotokas is thus a classic example of ‘split intransitivity’. Transitive verbs display uniform behaviour, but intransitives are split: some behave like transitives, others behave inflectionally according to a pattern distinct from transitives. Robinson (Reference Robinson2011), along with many others, speculates that there is a semantic basis for this split, along a cline of semantic transitivity that is separate from actual syntactic valence. But the actual semantic factors remain elusive, which means that at least provisionally it may be better to treat this as an inflection class contrast that has a predominantly syntactic and/or semantic application, rather than as an imperfectly characterizeable component of the morphosyntactic paradigm. In the following section, we explore these borderline phenomena in more detail.

4.4 Deponency

A modular conception of language is one of the noteworthy achievements of the discipline of linguistics. Being able to divide an analysis into phonological, morphological, syntactic, lexical, and so forth elements lends clarity and cross-linguistic consistency to a description. But, in practice, this model is often an idealization over a body of facts that do not quite line up, something we should bear in mind when considering what exactly it is that lies behind some morphological form.

As an illustration of the issues involved, consider the following examples of case-number inflection from the Baltic (Indo-European) language Latvian. Table 4.10 shows some nouns that are divided into four inflection classes based on their suffixation. All the nouns are inanimate, and assignment to these classes is arbitrary (as far as anyone can tell).

The same inflection classes are found with animate nouns, but in this case there is a clear relationship to the sex of the referent: inflection classes 1 and 2 are masculine, and classes 4 and 5 are feminine (Table 4.11). Does that mean we should treat these inflection classes as markers of sex, at least where the referent can be so characterized? Perhaps, however, the relationship is indirect. Alongside pairs as in Table 4.11, where we can think in terms of a single stem that is inflected according to one or the other inflectional pattern, we have pairs such as amerikāniet-is (m), amerikānet-e (f) ‘American’, which display distinct stems. This suggests that underlying the pairs in Table 4.11 is a null derivational process that toggles between masculine and feminine stems.Footnote ²

These inflection classes in turn have a non-trivial relationship to adjectival inflection: the opposition of 1st ~ 4th declension is what marks the distinction between masculine and feminine in short form (indefinite) adjectives (Table 4.12).Footnote ³ In these paradigms we are forced to say that this morphological opposition realizes the morphosyntactic feature of gender.

So in the end we have a similar set of morphological oppositions operating in three domains: as arbitrary inflection class attributes with non-human nouns, as sex markers with human nouns, and as gender markers with adjectives. Which function is primary, or does it even make any sense to pose a question like that? In practice there is no reigning consensus about how to deal with this question, although it might be reasonable to assume that the primary function would be associated with the word class that is cross-linguistically basic (i.e., nouns in this case).

Probably the most familiar example of this kind of analytical ambiguity are deponent verbs in Latin. Transitive verbs regularly oppose active and passive forms, as the partial paradigms in Table 4.13 shows. Deponent verbs – typically intransitive, though there are a few transitives as well, as shown here – use passive forms (where available in the paradigm), but are not otherwise demonstrably ‘passive’. Whether a verb is deponent or not appears to be lexically specified (though many share semantic properties that suggest they might be treated as ‘middle voice’; Kemmer Reference Kemmer1993). On the face of it, there are two equally satisfying (or unsatisfying) analytic alternatives: (1) deponent verbs are lexically specified as inflecting like passives, even though they are not, or (2) active and passive inflection represents two inflection classes; transitives employ the opposition between these two classes to mark the active ~ passive alternation.

The traditional approach is the first one, probably due to consideration of a further morphological fact. Latin also has a distinction between four major inflection classes (Table 4.14). These were originally due to the stem-final vowel rather than any differences in the inflectional suffixes, and some portions of the paradigm still lend themselves to such an analysis – for example, the imperfect subjunctive third-person singular forms in Table 4.14 could easily be segmented as amā-ret, monē-ret, rege-ret, and audī-ret, with four different stem-final vowels and an invariant suffix, just as in Wintu (section 2.2). But elsewhere in the paradigm matters become more opaque in synchronic terms. For example, on the basis of the last three classes, the present subjunctive 3sg appears to involve shortening of a long or deletion of a short stem-final vowel (*monē+at → moneat, *audī-at → audiat, *rege+at → regat), but ‘love’ would require an ad hoc morphophonological rule transforming a sequence of /a+a/ into /e/. Worse, the future indicative forms of the ‘love’ and ‘advise’ classes take an additional stem augment -b- which is lacking in the ‘rule’ and ‘hear’ classes. Thus, the system has the distinctions in inflectional morphology characteristic of inflection classes.

These distinctions are faithfully mirrored in the passive forms (Table 4.15). That means if we treat active and passive as parallel but separate sets of lexical items in some kind of derivational relationship, the coordination of their inflection classes emerges as a freakish accident. Alternatively, if we allow inflection class membership to be transferred across derivationally related items, we have in effect said that they are not derivationally related in any meaningful sense. This interdependence suggests that we treat active and passive forms as constituting part of a single paradigm, with deponent verbs representing a particular deviation from this (see section 5.5.8 for further discussion).

This is not to say that all superficially similar examples lend themselves to the same interpretation. A case in point is subject suffixation on verbs in Lealao Chinantec, an Oto-Manguean language of Mexico. In this language, animacy is an agreement feature in some parts of the grammar (e.g., determiners, numerals), but with verbs its grammatical status is more uncertain. Most verb roots are restricted to having either an animate or an inanimate ‘absolutive’ argument, i.e., intransitive subject or transitive object. But some roots occur in animacy pairs – that is, their lexical semantics are such as to be compatible with both. Table 4.16 illustrates unpaired and paired verbs, both intransitive and transitive. The contrasting suffixation will be the topic of the following discussion. (Note that there are also plural forms, but these are transparently constructed through affixation to the forms already illustrated here, and so would add nothing to the discussion; the forms shown are future/potential, which is the citation form in Rupp & Rupp’s Reference Rupp and Rupp1996 dictionary.)

Against this background, consider now the three major inflection classes (Table 4.17), which account for 96 percent of the fully inflected verbal lexicon (i.e., verbs that allow inflection for all three subject persons). There is a non-trivial relationship between animacy and inflection class, but it is a complex one. For the sake of simplicity, we look here only at transitive verbs, since intransitive verbs obviously lack complete paradigms in the case of inanimate subjects.

With unpaired verbs, animate verbs usually inflect as class III, whereas inanimate verbs show no particular tendency, as shown in Figure 4.1. With paired verbs, animates always inflect as class III, whereas inanimates are overwhelmingly assigned to classes I or II, as shown in Figure 4.2. Within this particular domain, the assignment of inanimates to class I or II is itself largely predictable by morphological criteria – namely, the behaviour of tone: if both members of the pair have the same tone, the inanimate verb usually inflects as class II. Other verbs undergo a tonal alternation. In that case the inanimate verb usually inflects as class I (Table 4.18).

Figure 4.1. Mapping of animacy type onto inflection class in Lealao Chinantec: unpaired transitives

Figure 4.2. Mapping of animacy type onto inflection class in Lealao Chinantec: paired transitives

Thus within a narrowly defined domain we can say that this morphological opposition realizes the morphosyntactic feature of animacy, but with respect to the verb system as a whole it is perhaps better classed as simply an inflection class opposition. This is not however to suggest that the Latin and Lealao Chinantec examples are fundamentally different; rather it is a necessary evil engendered by our (in general justified) analytic conventions that we are forced to anchor our description somewhere, either in the inflectional categories, or in the morphological paradigms. But we should recognize the fluidity of the boundaries.

In the foregoing examples, evidence for dual patterning comes from a comparison across lexemes – for example, the syntactic and semantic attributes of ‘passive’ morphology with the verb ‘love’ are different from the verb ‘exhort’. In some cases the evidence is actually paradigm-internal. Consider the system of subject prefixes in Santa Ana Keres, a Keresan language of New Mexico (Table 4.19). These follow either of two patterns, one characterized as actor marking, the other as undergoer marking, which is allied with object marking. It is not entirely clear how deterministic the semantic characterization of these two patterns is or whether there is some degree of lexical specification involved. But whatever the case may be, a verb like ‘eat’ in Table 4.19 stands out: it behaves like an ‘actor’ verb in the singular and like an ‘undergoer’ verb in the plural. Whatever semantic principles might underlie the distinction, they should not change along with subject number.

In Takelma, a language formerly spoken in Oregon (classified by some as a member of the Penutian family), values of person provide a context for a similar morphological alternation. The major inflectional patterns are shown in Table 4.20. Intransitive verbs fall into two main inflection classes based on their subject-marking suffixes, characterized by Sapir (Reference Sapir and Boas1922) as active (class I) and stative (class II). Transitive verbs (class III) have a distinct subject-marking paradigm, whose suffixes are either different from either class I or II, or correspond to the active suffixes. (First and second person objects are also overtly marked, by additional suffixes.) In addition, there is a fourth class. These are intransitive verbs that follow the stative pattern in the second and third person, but in the first person they additionally take the causative suffix -na-, and then inflect as transitives. If we attribute meaning to the first three classes (active, stative, transitive), the pattern in class IV makes little sense – particularly as it involves an alternation between the least active and most active types, skipping the intermediate class.

Such paradigm-internal irregularities are relatively unsurprising if we think of these paradigms as representing inflection classes: they are then instances of heteroclisis (Stump Reference Stump2006), which are hardly problematic from an analytical point of view. But given the apparently close alignment of the morphological paradigms with semantically and syntactically identifiable categories elsewhere in the system, we should not be too quick to dismiss the oddness of the situation.

4.5 Paradigm Shape and Morphosyntactic Function

The fluid boundary between morphological arbitrariness and morphosyntactic motivatedness that deponency represents has an analogue in the way individual paradigms are configured. Consider first the paradigm in Table 4.21, from Chipaya, an Uru-Chipaya language of Bolivia. This shows the declarative subject agreement enclitic, which docks onto various hosts. The paradigm distinguishes values of person, number, and gender, but the forms il and izh can only be described in terms of a disjunction of values: il is used for first-person (exclusive) and third-person singular feminine, whereas izh is used for third-person masculine and all the plural values other than first-person exclusive.Footnote ⁴ Because disjunctive patterns such as this cannot be described as a natural class of feature values, it looks to be morphologically arbitrary.

But not all disjunctive patterns can be quite so readily ascribed to morphological stipulation. Precisly here, at the intersection of person, gender, and number, we find a whole family of examples where apparent disjunction may be motivated at another level. Consider the verbal paradigm in Table 4.22, from Tucano, a Tucanoan language of Columbia. Verbs mark subject agreement for person, number, and gender; for the third person, masculine, feminine, and neuter are distinguished. First and second person subjects, however, do not distinguish gender, and, surprisingly, the form used for them is the same as the third person neuter – that is, precisely the gender value that would not apply to first- and second-person arguments. Thus, this form involves the disjunction of first and second person, by implication non-neuter, with the third-person neuter.

One could, however, understand this as reflecting the intersection of morphological and morphosyntactic asymmetries. The morphological asymmetry is manifested in the treatment of gender marking with the third person: only masculine and feminine are overtly represented, whereas the form used with neuter subjects is simply a default form, compatible in principle with anything, but effectively limited to neuter arguments. The morphosyntactic asymmetry is manifested in the syntactic properties of different persons: third-person arguments control gender and number agreement, but first- and second-person arguments do not. On this interpretation, first- and second-person subjects take the default form, which also happens to be the one used for third person neuter subjects.

On the face of it, this might seem to be a baroque explanation for a minor paradigmatic quirk, but the pattern in its essentials is repeated across various unrelated languages, increasing its plausibility. For example, Krongo, a Kadugli language of Sudan (Table 4.23), has essentially the same pattern as Tucano, with the difference that plural agreement applies to all persons, and not just third-person masculine and feminine.

Although the assumption that the first-/second-/third-person neuter form is a morphological default requires a certain suspension of disbelief (there is no evidence for it other than the contingent demands of analyzing these particular paradigms), the restriction of gender agreement to third person does mirror a cross-linguistically familiar restriction of pronominal gender distinctions to third person (Siewierska Reference Siewierska2004: 104–105). That said, this paradigmatic type cannot necessarily be construed as a translation of pronominal regularities, because further peculiarities may be found. For example, in some Nakh-Daghestanian languages, the person-based restriction on gender agreement occurs only in the plural. In most of the languages of this family there is no person agreement as such, only gender and number agreement (with the absolutive argument). Consider the Archi copula in Table 4.24. In the singular, four genders are manifested: genders I and II for humans, and genders III and IV for non-humans. In the plural, humans are distinguished from non-humans, but only for third-person arguments; first and second person take the same agreement marking as non-humans. A similar pattern is found in Ingush (Nichols Reference Nichols2011: 143, 431); this is striking, because although both languages are from the same family their relationship is remote, Archi being from the Daghestanian branch and Ingush from the Nakh branch, and thus of a degree of separation comparable to the different branches of Indo-European (Nichols Reference Nichols, De Busser and LaPolla2015: 270).

This interpretation assumes a certain separation between morphological forms and the values they represent: the forms realize certain morphosyntactic values, but the way they are deployed in the paradigm interacts with restrictions imposed by other feature values. The fluid nature of this relationship is strikingly illustrated by Orejón, like Tucano a Tucanoan language, but from the Western rather than the Eastern branch. Before highlighting the relevant points, it should be noted that Orejón differs from the languages presented so far, in that there are only two genders, masculine and feminine, and nouns that denote inanimates take masculine agreement. With that in mind, consider first the indicative present-future paradigm in Table 4.25. This is, in effect, the two-gender analogue of the Tucano paradigm, with gender agreement in the third-person singular, and one form for the rest of the singular. Contrast this with the corresponding interrogative paradigm. Each paradigm comprises four suffixes that, although not identical (two of the four differ slightly), are no doubt morphologically related. But the striking fact is that their distribution is different: whereas in the indicative the gender-agreeing suffixes are restricted to the third-person singular, in the interrogative their range is extended to the second-person singular. This can be seen even more clearly in the past tense paradigms in Table 4.26, which have only three forms each: two gender-agreeing forms, and a single form for the rest. The indicative and interrogative paradigms thus are characterized by different distributions (and partly different forms, it must be admitted) of the gender markers based on which person values are licit agreement controllers in these contructions.

On the other hand, we should not get too carried away with an account driven by twists in the rules of agreement. Consider Tucano again. Many verbal constructions involve a nominal form, termed gerundive in the description. The nominal form marks gender-number using suffixes identical to those found on nouns, as in Table 4.27, part (b).Footnote ⁵ This gerundive forms a periphrastic construction together with an auxiliary verb (the verb ‘do’ shown earlier in Table 4.22). But whereas the auxiliary displays the apparent person-based restrictions on gender agreement, the gerundive does not. The result is a periphrastic construction, such as that shown in Table 4.28 (what we interpret as the default form of the auxiliary is shown underlined), whose individual members display different gender agreement patterns. If we treat this construction as a single agreement target, then clearly the gender restriction is morphological and not morphosyntactic.

In most other Eastern Tucanoan languages the auxiliary element is suffixed to the nominal form; this means that the morphological unity of the construction is even more apparent, as in the non-past conjectural paradigm of Carapana in Table 4.29. Note here that the auxiliary – that is, the equivalent of wee in Tucano – is simply zero.

The paradigmatic configurations surveyed in this section show us again that the distinction between morphosyntactically motivated and morphologically arbitrary is not always clear-cut. The disjunctive distribution of forms within these paradigms, which ought to be a sign of arbitrariness, may well have a morphosyntactic motivation, in the sense that morphosyntactic features may license agreement. At the same time though, we should not allow ourselves to be completely seduced by syntactic explanations, as the dual agreement patterns in auxiliary constructions in Tucano and Carapana show. This suggests that, sometimes at least, we face not a choice between competing accounts, but rather the simultaneous validity of both.

4.6 Conclusion

In this work, we understand morphological complexity as arising where inflectional morphology strays out of bounds, those bounds being defined somewhere other than in morphology proper, such as in syntax or semantics. We can treat this as an all or nothing affair: forms either follow the terms defined by syntax and semantics or they do not. On this simplifying assumption, calling something an inflection class means it is wholly arbitrary and must be explicitly stipulated. By the same token, the distribution of forms in a paradigm should be definable in terms of natural classes of morphosyntactic feature values – otherwise it is arbitrary and morphologically stipulated. The facts, as so often is the case, do not always cooperate with this simple dichotomy. There are seeming inflection classes that nevertheless have some relation to meaning or function, and there are paradigms whose distribution of forms is certainly related to morphosyntactic values but where this relationship is indirect. One could take this as evidence that a motivated versus arbitrary distinction is artificial, and we need a more gradient and nuanced view of how morphological paradigms come about. Or one could maintain the distinction, and take these data as a call for a more precise mapping between the two, say by weighting the differing contributions coming from either side of the grammar. From a descriptive and typological point of view, we suggest maintaining the latter approach, because the terms and analytic concepts we currently have at our disposal presuppose it.