We present a new phenomenon in inflectional morphology, ‘repartitioning’, based on data from Soq (Trans New Guinea). In repartitioning, the semantic boundary between two sets of morphological forms is redrawn in a single domain; one feature value takes over part, but not all, of the meaning of the other. In Soq the boundary is redrawn between the yesterday past tense and the hodiernal; the domain is the lexeme s-‘stay’. For this one verb, the yesterday past takes over most of the range of the hodiernal, while the morphological forms remain regular. In clause chains the repartitioned verb surprisingly shows no syntactic effects. We demonstrate key differences from known phenomena, notably syncretism and overdifferentiation. Repartitioning is indeed new. It can be modeled in a theory based on default inheritance, but poses problems for other approaches. Finally, we present a typology of featural mismatches that situates Soq relative to known phenomena.*
inflectional morphology, morphosyntactic features, repartitioning, clause chaining, tense, Trans New Guinea languages, Soq, typology
Morphologists have established a typology of inflectional phenomena, specifying the various ways that lexemes deviate from the canonical ‘one form, one meaning’ principle (Stump 2001, 2016, Baerman et al. 2005, Baerman 2015, Corbett 2015:149–58, Bobaljik 2017:10–15). In this article we add a new phenomenon to that typology, based on data from Soq, a language spoken in Papua New Guinea. We call this phenomenon repartitioning. In repartitioning, part, but not all, of the meaning of one feature value is reassigned to another feature value under certain circumstances. It involves regular inflectional morphology having an irregular semantic interpretation.
We introduce Soq in the remainder of this introduction. Then we present the basic morphological facts about the language in §2, focusing on tense, since this is where we find repartitioning. In order to establish that repartitioning is indeed new, we next provide a variety of comparisons to potentially similar phenomena in §3. Having established that repartitioning is unlike any of the patterns we consider, we move on from the repartitioned morphology to its syntactic behavior (§4). This examination reveals some of the featural properties of the forms in question, which enables us to make further crosslinguistic comparisons in §5. These comparisons form the basis for a typology that sets repartitioning within the range of known morphological phenomena. It is interesting that this typology is, at least at our present state of knowledge, relatively empty, which highlights the unusual nature of Soq repartitioning. We conclude in §6. [End Page 711]
Soq [mdc] is a Papuan language spoken by some 700 people in Madang Province, Papua New Guinea. Its speakers live in three villages, two located along the coast of Astrolabe Bay and a third, Buram, farther inland in the foothills of the Finisterre Range (see the map in Figure 1).
The language belongs to the large Trans New Guinea family (Pawley 2005, Ross 2005, Pawley & Hammarström 2018). It is a member of the Madang branch of Trans New Guinea (Z’graggen 1975a, Pawley 1995), which contains some 108 languages, and within that it belongs to the thirty-member Rai Coast group, and then the four- or five-member Minjim group (Z’graggen 1980, Ross 2000).
Soq was first encountered by Europeans when the Russian anthropologist and explorer Nikolai Miklouho-Maclay (1846–1888)1 landed at Garagassi, near Gorendu village, on 20 September 1871 (Mikloucho-Maclay 1990:76–82). Although Maclay would spend nearly two years there, he was not focused on linguistic research. He recorded some 300 Bongu words, as well as a shorter word list comparing Bongu, Anjam, and four more distant languages (Mikloucho-Maclay 1882). He recorded some of the words of ‘the Male dialect’ during his stay (Mikloucho-Maclay 1990:250), and lists are available in Mikloucho-Maclay 1993:145–51.
For the next couple of decades, the linguistic literature on the languages of Astrolabe Bay features Bongu and Anjam (e.g. Zöller 1890, 1891, Hagen 1899, Schmidt 1900, Hanke 1905), but omits mention of Soq, except to say that the residents of Soq-speaking villages ‘speak a different idiom than the residents’ of the Bongu-speaking villages (Zöller 1891:68).2 Then in his 1909 grammar of Bongu, the missionary August Hanke (1909:115–18) gives a word list of about fifty items for each of the three Soq-speaking villages. These data were then incorporated into subsequent work (e.g. Ray 1919). As far as we know, some fifty years elapsed before there was more fieldwork on Soq. Oren Claassen collected a Bongu word list and may have collected ‘brief vocabulary lists from four coastal villages near Bongu’ (Claassen & McElhanon 1970:61), and John Z’graggen conducted a large survey of the whole Rai Coast area from 1971 to 1973 (Z’graggen 1980:v). He also made the first statement that the linguistic varieties of [End Page 712] Buram, Kaliku, and Male form a single language, which he called Male (Z’graggen 1975b:18). He grouped Soq with Anjam, Bongu, and Sam in a Mindjim branch of Rai Coast (Z’graggen 1975a:577).
Classificatory work on the Rai Coast languages has also been quite limited: the only proposed revision of Z’graggen’s classification has been Ross’s (2000), who added a fifth language, Pulabu, to Z’graggen’s Mindjim subgroup. He also shortened the spelling of Mindjim to ‘Minjim’, and we continue his usage here.
The languages of the area often did not have indigenous names, and were not usually given names by Europeans until Z’graggen, following a practice of Arthur Capell’s, started ‘naming languages after well-known villages or areas’ (Z’graggen 1968:422). Z’graggen named Soq ‘Male’ after the Male village, but this label is not used as a language name by the language’s speakers. They have instead decided to use the word soq ‘speech, talk’ as a label for the language spoken in Male, Buram, and Kaliku, and we follow that decision here.
Since the 1970s there has been no work on Soq apart from the fieldwork for this study. Daniels has conducted three field trips to research Soq, totaling about five weeks: one trip in February 2016, one in September 2016, and one in June 2018. The data for this analysis comprise 133 minutes of transcribed naturalistic speech, plus ten and a half hours of recorded elicitation and many more hours of informal, unrecorded elicitation. The first two trips yielded the basic morphological analysis of Soq and the discovery of repartitioning; on the third Daniels conducted more targeted elicitation about the phenomenon. All data are archived and publicly available at his PARADISEC collection (Daniels 2018).
2. Soq morphology
Here we present the relevant facts about Soq verb morphology. We begin with the key information on tense, and we include the basics on person and number marking, which is cumulated with tense. Soq possesses four tenses, which are marked by suffixes on the verb. We understand tense to be (a set of) morphological forms used for the expression of ‘location in time’ (Comrie 1985:9). The normal situation is illustrated by the verb mar- (and its allomorph mr-) ‘talk’ in 1.3
The yesterday past, illustrated in 1a, refers to events occurring yesterday. The tense in 1b, which we call the hodiernal, refers to the period starting at the beginning of the day of the speech act and running up to, and including, the time of the speech act. This is the regular, overwhelmingly most common pattern for verbs, irrespective of their lexical semantics. It is tricky giving translations for such examples, since English is a poor metalanguage here: there is no English equivalent tense and no single temporal adverb [End Page 713] to give an adequate rendering. The important point for the reader to bear in mind is that each of the four tenses in Soq covers a continuous period of time.
Just one verb, s- ‘stay’, takes the normal suffixes but uses them to carve up the semantic space differently—in other words, it repartitions it. Its pattern is shown in 2.
With this one verb, the yesterday past marks all events with past time reference from the day before, and also the day of, the speech act (2a), while the hodiernal marks events as having present time reference (2b). The forms are unremarkable, but their interpretations are modified, when contrasted with every other Soq verb. This is a hallmark of repartitioning. While s- ‘stay’ has various uses, as we shall see, it is important to stress that there are many examples, like 2, where it is a straightforward lexical verb. It is worth taking a moment to address the fact that repartitioning in Soq occurs with only a single verb. Readers may wonder—and referees did wonder—how much a single verb in a single language can teach us about morphological typology. We wish to stress that typology is more than simply describing the predominant linguistic patterns. It is equally important to give an account of rare phenomena and also of why they are rare, since that gives us clues about the functional pressures that delimit the space of possible linguistic systems. We return to this point in the conclusion.
In the rest of this section we outline the essential data on Soq verb morphology, including morphologically defined verb classes and the allomorphs they take. The discussion focuses on tense suffixes, since that is where the repartitioning occurs; aspect is marked periphrastically and does not concern us here. Soq possesses four tenses: the remote past, the yesterday past, the hodiernal, and the future. Each tense is marked by a paradigm of suffixes that also index subject agreement. The paradigm for the yesterday past is given in Table 1.
The suffixes for the yesterday past are heavily syncretic, so that there are only two forms: -kimbi (or -kembi for some speakers) marks second-person singular or first-person subjects, glossed ‘2sg/1’, while -kibi (or -kebi) marks second-person plural or third-person subjects, glossed ‘2pl/3’. These suffixes also exhibit some phonologically conditioned allomorphy: the initial k becomes a g when the stem ends in a voiceless consonant. The time reference of this tense is to the day before the speech act, as illustrated in 3.
[End Page 714]
The time reference of this tense can also extend back further if the speaker wishes to construe an event as having happened ‘just the other day’ or if an event is comparatively recent in the context in which it occurs. Example 4, for instance, describes an event that took place some twenty years before the speech act. But since it occurs as the coda to a story about events from over 100 years ago, the speaker uses the adverb yabi ‘yesterday’ and marks the events in the yesterday past. After this he switches back to the far past in describing what happened after 4.
The suffixes of the hodiernal tense are given in Table 2.
The pattern of syncretism in this set of suffixes is similar to that found in the yesterday past, but adds one distinction. The syncretism between second-person singular and first person remains, and this category is marked by -ekum or -ikum, depending on verb class. But unlike in the yesterday past, third-person singular subjects (marked by -ken) are distinguished from second- or third-person plural subjects (marked by -eku or -iku, again depending on verb class).
The time period covered by the hodiernal tense encompasses events with present time reference and stretches back to events that took place earlier on the day of the speech act. This period is illustrated in examples 5 and 6, the first of which was uttered as the speaker was watching a movie and described an event taking place at the time of the speech act, and the second of which describes events that took place earlier on the day of the speech act.
The boundary between the yesterday past and the hodiernal is nightfall yesterday, as illustrated in 7.
This example is taken from a story about what the speaker did the day before; when the story transitions from afternoon to night, she changes the tense marking from the yesterday past to the hodiernal.
The forms of the far past are shown in Table 3. The pattern of syncretism found in the far past mirrors that found in the hodiernal, distinguishing three categories: 2sg/1, 3sg, and 2pl/3pl. These suffixes exhibit some allomorphy, which is conditioned by verb [End Page 715] class. The time reference of the far past is fairly straightforward. It marks events as having happened before yesterday, as in 8.
Finally, the future-tense suffixes are given in Table 4. This paradigm adds yet another distinction to the pattern of syncretism seen in the far past: 1sg is distinguished from 2sg/1pl. The former is marked by -ari, the latter by -umsi.
The future’s time reference is also straightforward: it marks events as occurring in the future—that is, as not having happened yet, as in 9.
We now turn to a discussion of verb classes in Soq. Because repartitioning involves the irregular behavior of regular morphology, it is important to establish that s- ‘stay’ really is a regular verb, taking normal Soq verb suffixes. And to establish this it is necessary to discuss verb classes.5
Verb classes are distinguished by the suffixes they take. Certain suffixes exhibit allomorphic variation that is lexically conditioned, such as the 3sg.fpst suffix, which can be -albi, -elbi, or -ulbi, depending on the verb it attaches to. More precisely, it is not the verb that conditions the allomorphy, but the stem to which the suffix attaches. This is usually identical to the verb root, but it can also be one of the object suffixes.
In order to describe Soq verb classes, four patterns of suffix allomorphy need to be accounted for. The seven verb classes and the allomorphs of key suffixes are shown in Table 5. The verb s- ‘stay’ belongs to class V, a small class that, at our present state of knowledge, contains three members: i- ‘give’, s- ‘stay’, and the causative auxiliary t-.
To summarize, then, s- ‘stay’ is a regular class V verb, taking the allomorphs of suffixes that are expected for that class. There is nothing surprising about its inventory of forms. Two sets of its forms, though, show different semantic behavior. With s- ‘stay’: [End Page 716] (i) the yesterday past, which normally only refers to events from yesterday, covers the period from yesterday up to but not including the present; and (ii) the hodiernal, which normally covers the period from earlier today up to and including the present, refers only to present events. In effect, the partition between these two categories is moved: usually it is situated at the boundary between yesterday and today (which, as mentioned above, in the Soq conception is located at nightfall yesterday). But with this verb the partition is moved to the boundary between events earlier on the day of the speech act and now. This repartitioning is schematized in Figure 2, where the class III verb mar-/mr- ‘talk’ illustrates the predominant pattern.
As Fig. 2 shows, a subset of the functions performed by one form (the hodiernal) is reassigned to another form (the yesterday past) in a particular domain (the verb s- ‘stay’). Repartitioning is a robust feature of the grammar of Soq. We confirmed its existence with over a dozen speakers, from the villages of Kaliku and Male, in both natural speech and elicitation contexts. And although s- ‘stay’ must repartition, no other Soq verbs may do so, including other class V verbs and other semantically stative verbs like dan/daniŋ ‘sit’, un/unun ‘sleep’, and ayiq ‘not want’.
We claim that this is a new phenomenon. While there are related and similar behaviors in other languages, which we present in §3, closer examination always reveals clear differences. As a result of these comparisons we shall establish that repartitioning is to be understood in morphosemantic terms.
Note that we have described repartitioning in terms of an inferential-realizational approach to morphology (Stump 2001:1–30), since we believe that the phenomenon is seen most clearly in terms of the paradigm. For instance, there has been recent attention to the paradigm cell filling problem (Ackerman et al. 2009:54, Malouf 2017:431), the issue of how speakers of languages with complex morphology, given exposure to an inflectional form of a new lexeme, can infer the rest of its paradigm. In Soq the issue is [End Page 717] a more serious one, in that there is a verb where the problem is not just to fill the paradigm but rather to establish its unpredictable use. However, the situation is perhaps even more challenging if one takes a view based on the morpheme; in such an account, one would need to allow a set of person-number-tense morphemes to change their meaning in an identical way, just in case they attach to the verb s- ‘stay’.
3. Comparisons and definition
There are several well-established phenomena that are comparable to repartitioning, to various degrees. We review them in turn. There are certain resemblances that can shed light on repartitioning, but for each phenomenon we demonstrate that repartitioning is in fact distinct (and hence is new). This leads to our definition of repartitioning in §3.6. For Soq it is relatively straightforward to establish the relevant features and their values, since the forms and their grammatical meanings are consistent across the verbal lexicon (with the single exception that is our focus). And since the relevant forms in Soq are formally regular, and the issue is with their interpretation, this immediately excludes from consideration one set of inflectional phenomena; we need to look rather at those phenomena that relate to distribution of forms within paradigms.
Syncretism is the situation in which a single form realizes more than one morphosyntactic or morphosemantic specification. Thus English has a distinction between the present and past tenses, but I cut can be present or past; that is, cut is syncretic between present and past. There is a distinction, respected elsewhere, that is not found here. There can be different reasons for this situation, and hence different analyses (see Baerman et al. 2005 for extensive illustration and discussion).
We have already seen some interesting examples of syncretism. Take, for example, the yesterday past forms (which we saw in Table 1, reproduced below) and the future forms (Table 4, also repeated here). Let us focus on the third person.
The yesterday past tense has the same form for the third singular as for the third plural. However, in the future tense they are distinct. So we have clear morphological evidence that Soq distinguishes number in the third person, and we therefore conclude that there is syncretism of third singular and third plural in the yesterday past tense.
Syncretism does not cover repartitioning, however. Syncretism involves two or more paradigm cells having an identical form. In repartitioning, we might say that one set of forms (those comprising the yesterday past tense) encroaches on the semantic domain of another set of forms (the hodiernal), but the first set does not take over from the other. There are still two sets of forms available; it is rather the relation between them that is different.
There exists another possibility, which might tempt some readers: analyzing Soq repartitioning as two separate instances of syncretism. One, occurring on most verbs, between a ‘today past’ tense and a ‘present’, and another occurring on s-, between the yesterday past and this putative ‘today past’. This analysis proves inadequate, but we must defer a full discussion until §4.2, when we have given sufficient preparatory data to present the relevant syntactic facts. [End Page 718]
Overdifferentiation (Bloomfield 1933:223–24, Thornton 2010–2011:438–43) refers to the situation where a lexeme (or lexemes) is distinguished by having an additional cell in its paradigm, motivated by an additional form. Consider the East Norwegian adjective paradigms in Table 6 (Hans-Olav Enger, p.c.; for examples, see Corbett 2011:116–17).
East Norwegian has three gender values, as found in the articles. However, a normal adjective like tjukk ‘thick, fat’ makes only the distinction between masculine and feminine together versus neuter (Enger & Kristoffersen 2000:104). Compare now liten ‘small’. This adjective exceptionally distinguishes all three gender values. It is overdifferentiated in having distinct masculine and feminine forms.6 We may also note that liten ‘small’ shows an additional distinction, that between plural and definite singular. As a result, the paradigm of liten distinguishes five forms as opposed to the regular three; it is overdifferentiated in two respects.
The types of syncretism and overdifferentiation discussed above can be seen as sides of the same coin. We have lexemes that might be expected to be equivalent; syncretism can arise when some lexemes have ‘too few’ forms, and overdifferentiation is the situation where some lexemes have ‘too many’. Overdifferentiation merits discussion in that the verb s- ‘stay’ makes a distinction that other verbs do not. But unlike overdifferentiation, the distinction is not an additional one, preserving all of the regular distinctions. Rather, s- ‘stay’ makes a different distinction from all other verbs: it draws the dividing line in a different place, rather than adding a distinction.
Deponency offers a less obvious comparison, yet it shows some similarity to repartitioning. The best-known examples of deponency are found in Latin and are illustrated by the forms in Table 7.
Consider first the regular verb amāre ‘to love’. It has active and passive forms that are clearly distinct. A minority of verbs are like hortārī ‘to exhort’. They have forms that [End Page 719] formally resemble the passive, but are syntactically active. If the verb is intransitive there is no additional effect, but if a verb is transitive, like hortārī ‘to exhort’, then it will be defective. (For further examples and discussion see Baerman et al. 2007.)7
How is this relevant? The Soq verb s- ‘stay’ has forms that ‘ought’ to indicate yesterday past but that are used where every other verb has the hodiernal, that is, for earlier today. This is the analogy to deponency. And like deponency, a set of forms is involved (the relevant set of person-number forms), rather than a single form. However, deponency involves a straightforward substitution: in the Latin example, passive forms take the place of active. Repartitioning differs in that it is not the full substitution of one set of forms for another. Rather, just one part of the semantic range of one set of forms is reallocated to another. The two extremes of the grammatical meanings in question (yesterday and now) are not affected. Moreover, there is a new distinction in grammatical meaning, namely the present, made available by the Soq repartitioning; in contrast, deponency involves realization of the same grammatical meaning by different forms.
3.4. Metaconjugation and other possible analogies
We discuss here three interesting phenomena which, at first glance, appear similar to repartitioning but which, on careful examination, prove different in important respects. We treat them together here, since Stump (2016:197–227) gives insightful discussion of them, along with deponency.
Metaconjugation, introduced by Stump (2016:202–17, 227) with Sanskrit data, involves the use of the same inflection class material in different functions in different lexemes. Specifically, the same inflectional material is used for different tenses in different verbs. However, the inventory of tenses, and their grammatical meanings, are not affected. Since metaconjugation depends crucially on the distribution of different inflection classes, it is not directly relevant to repartitioning.
Closer to repartitioning are the interesting data from Kashmiri (Stump 2016:217–24). Here again we find the same forms being used with different grammatical meanings. If we concentrate on the distinction between recent past, indefinite past, and remote past, we find that there are identities of the indefinite past (conjugation II verbs) and recent past (conjugation III), and of the remote past (conjugation II) and indefinite past (conjugation III). While this ‘recycling’ of inflectional material is remarkable, the outcome is unremarkable: each verb has three sets of forms, which realize three different past-tense distinctions. There is no redrawing of the tense boundary comparable to what we find in Soq repartitioning.
Third, it is tempting to look for an analogy in the preterite presents of Germanic languages; we find there a small class of verbs whose present has forms with the appearance of strong preterite-tense forms, and whose preterite has the forms expected for a weak preterite tense. These may be seen as exhibiting an interaction of deponency and heteroclisis (Baerman 2007:16–17; cf. further discussion in Stump 2016:224–27). The complexities of this group of irregular verbs are laid out by Randall and Jones (2015), who also give a helpful review of the literature. While Grimm’s earlier theory of preterite presents involved ‘tense-shunting’, which might seem similar to repartitioning, more recent accounts avoid this (Randall & Jones 2015:171), and the difficulties of understanding the preterite presents then turn out to cast no light on repartitioning.8 [End Page 720]
In all three cases, the analogy to repartitioning breaks down for essentially the same reasons that it does with deponency. A form that is usually used for one meaning is used for another meaning in some context. This is similar to Soq. But in these cases, the inventory of grammatical meanings is not affected: for example, every Kashmiri conjugation class inflects for the same three past tenses. In Soq this is not so, since with s- ‘stay’ one distinction that is available to all other verbs is lost (today vs. yesterday) while another is gained (now vs. earlier today).9
3.5. The interaction of overdifferentiation and syncretism
Having failed to find a recognized phenomenon in inflectional morphology that covers the Soq phenomenon we have termed repartitioning, we should check whether there is some interaction of phenomena that could provide an analogue. We consider the interaction of overdifferentiation and syncretism. Since they are opposites in a sense, an interaction seems unlikely. We might expect them to cancel each other out. Yet this need not be the result, as data from the Russian ‘second genitive’ show (we follow Corbett 2011:118–20 here). Russian has six primary case values, plus additional candidates including the second genitive, which have generated substantial research (see, for instance, Zaliznjak 1973, Worth 1984, Comrie 1986, and Corbett 2012:203–6). Consider the forms of the noun kisel´ ‘kissel’ (a thickened fruit drink), which is a regular noun, and čaj ‘tea’. They are similar in the normal genitive.
These nouns contrast, however, in certain partitive expressions.
Čaj ‘tea’ is one of a restricted subclass of nouns with a separate form, the second genitive, as in 13. Some nouns in this subclass normally use the second genitive in the appropriate context (partitive expressions), while with others it competes with the ordinary genitive (Paus 1994, Comrie et al. 1996:124–25, Uspenskij 2004:11–26, Brehmer 2009, Corbett 2012:203–6, and references there). [End Page 721]
So far we have evidence for overdifferentiation. Nouns like čaj ‘tea’, a minority of the noun inventory, have an additional form. Now this additional form, čaju, is itself interesting. Consider the relevant part of the noun paradigms, shown in Table 8.
We see that the form of čaj ‘tea’ which gives rise to overdifferentiation is syncretic with the dative. We cannot simply claim that this form is the dative, because of the agreement evidence. This evidence is not straightforward: if there is an agreeing modifier the use of the second genitive is strongly disfavored, and the ordinary genitive is more likely.
The presence of the modifier zelenogo ‘green-m.sg.gen’ means that the ordinary genitive čaja is much more likely. But when the noun stands in the less likely second genitive, in expressions similar to 14, genitive agreement is still required. Thus zelenogo čaju ‘green tea’ is possible—if rare—as a second genitive. We may try putting the attributive modifier in the dative.
Example 15 is acceptable, but only in syntactic positions where a dative is required. It is not possible as a second genitive: it cannot be used in 14. Thus second genitives are not syntactic datives. The additional form, the second genitive, is syncretic only with the dative.
This interaction of overdifferentiation and syncretism comes closer to the Soq situation, but still does not match it. A key point about Soq is that the tenses involved, namely remote past, yesterday past, hodiernal, and future, have just one logical arrangement. They denote a progression from past to future, along the timeline, and there is no alternative arrangement. Repartitioning involves moving a delimiting point along a single scale.
Since the analysis for Russian does not carry over into Soq, we should ask whether, conversely, we could reanalyze Russian as a second instance of repartitioning. This seems dubious. There are various suggestions for a case hierarchy, and genitive is often, but not always, treated as contiguous to dative (see, for instance, Blake 1994:89–91, 157–62). But it does not seem that, if a partitive section of the genitive (the range of the second genitive) were to be reallocated, that part of the genitive would fall to the dative. There have been discussions of a case hierarchy specifically for Russian, in Chvany 1982 and Caha 2008. Neither of these includes the second genitive, and in any event neither of them has genitive contiguous with the dative, so there are no grounds to expect syncretism of genitive with dative. However, Jakobson (1971 :175) did include the second genitive in his cube representation of the Russian case system, and he positioned it maximally distant from the dative. Hence there is no reason to argue that there is a motivated repartitioning within the Russian case system, aligning the second [End Page 722] genitive and the dative, which would be at all comparable to the repartitioning of the Soq tense system.
3.6. Defining repartitioning and establishing that it is new
Given the perspective provided by phenomena that are in some respects comparable, we now give a definition.
(16) In repartitioning, some subset of the functions performed by form A is reassigned to form B in a particular domain. The functions performed by form B are semantically adjacent to the form A functions that have been reassigned to form B.
In this definition, a ‘form’ can be conceived of as a single form or as a set of forms—in the Soq case, for example, we are actually dealing with whole sets of agreement suffixes, and their patterns of syncretism. The ‘particular domain’ in which the form is repartitioned is just the verb s- ‘stay’; we say more about possible domains in §5 below. And we mean ‘semantically adjacent’ in the sense suggested by the timeline in Fig. 2, that is, adjacent in terms of temporal progression.10 We recognize that defining the term in this way essentially limits repartitioning to semantic domains in which the set of possible meanings is a one-dimensional progression, such as tense and number. (Others, such as respect or animacy, might be conceived in this way, but this would seem to be less justifiable.) It would be possible to operationalize a concept of multidimensional semantic space using the technique of semantic maps (Haspelmath 2003) and to incorporate that into the definition. (The timeline in Fig. 2 is, essentially, a one-dimensional semantic map.) However, since the Soq case does not motivate the use of multidimensional semantic maps, we do not employ them, and simply leave the possibility of repartitioning in multiple dimensions open for now.
It is worth keeping track of the characteristics of the Soq phenomenon that we have labeled ‘repartitioning’. The Soq verb s- ‘stay’ distinguishes four tenses, like all other verbs. These tenses are realized by a set of forms, not a single form. All of them are regular. However, the semantic boundary between two of the tenses is shifted, just for this single verb. This is not syncretism, since what is ‘gained’ by one tense is ‘lost’ by another. There is no extra set of forms, only the expected regular forms. We also stress that repartitioning is seen in different uses of the verb. For example, s- ‘stay’ is employed as an auxiliary verb in the imperfective construction, which marks progressive or habitual aspect.
We also find repartitioning when it is used as a lexical verb.
[End Page 723]
We have gone to some lengths to show that repartitioning differs from various analogous phenomena. Possible analogies were both those that we took to be contenders, and those that have been suggested to us, for instance by the lists in Baerman 2015, Corbett 2015:150–58, Bobaljik 2017:10–15, and Stump 2017a, 2017b:69–79. Having checked some obvious possibilities, we should now take a step back. Rather than going through more and more increasingly less likely phenomena, we rather give a higher-level review of the types of inflectional phenomena that are attested, and this again confirms that repartitioning is indeed new.11
There is a whole typology of morphological phenomena based on what Corbett (2015:149–53) calls ‘requirements of form’. Irregularities of form may involve lexical material (various types of alternation, right up to suppletion) and irregularities of inflectional material (syncretism and uninflectability). We have discussed the most likely of these. However, as we pointed out at the beginning of §3, the forms in question in Soq are fully regular, and so all of the phenomena based on irregularity of form can be ruled out as not being possible analogues for repartitioning.
We must therefore turn to the next possibility, namely ‘requirements of structure’ (Corbett 2015:153–58)—how the inflectional paradigm interacts with the syntax. The contrast between requirements of form and requirements of structure has also been articulated by Stump (2017a:2) as ‘the twin tasks of defining a language’s inflected forms and of identifying the grammatical dimensions of the inflectional paradigms that house them’. The most promising place to start is with phenomena that involve irregularities in the way that inflectional features (such as case or tense) interact with the syntax. Our next step must therefore be to survey such phenomena to see whether the phenomenon we claim to be new has in fact already been discussed. In order to do this we first need to determine whether repartitioning has any role in the syntax of Soq, which we do in the next section, on clause chaining. That will provide information on the feature values relevant to verbs that are required for an analysis of Soq syntax. Then, in §5, we shall have the necessary background to return to the issue of whether repartitioning is new.
4. Soq repartitioning and clause chaining
Many Papuan languages, particularly those belonging to the Trans New Guinea family, exhibit frequent clause chaining (Roberts 1997, Foley 2018). Soq is no exception. In a typical Papuan clause chain, a number of medial verbs are chained to a final verb. Final verbs are fully finite and are marked for the full range of verbal categories, such as subject agreement and tense. [End Page 724] Medial verbs are typically marked for a more restricted set of categories, and certain categories that are marked on the final verb will then have scope over the preceding medial verbs. For example, in Mand, a Madang language distantly related to Soq (Daniels 2015), medial verbs do not make any tense distinctions, but final verbs distinguish several tenses. The tense information from the final verb has scope over preceding medial verbs. So in 19, for example, the medial verbs akre ‘fish’ and krɨc ‘throw’ are interpreted as having immediate-past time reference because the final verb of their chain, akwid ‘go up’, is marked for immediate past.
This example illustrates another common feature of Papuan clause chaining: switch reference (Roberts 1997). In many languages medial verbs bear a suffix that indicates whether their own subject is the same as, or different from, the subject of the following verb in the chain. Thus in 19 the same-subject suffix on akre ‘fish’ indicates that the subject of that verb is coreferential with the subject of krɨc ‘throw’, and the different-subject suffix on krɨc indicates that the subject of ‘throw’ is different from the subject of akwid ‘go up’.
A final feature of many Papuan clause-chaining systems is a modality distinction (Roberts 1990). Many languages have more than one set of medial suffixes, and the choice of medial-suffix set depends on the tense-aspect-mood category of the final verb. In most languages the division between different sets of medial suffixes falls along modal lines, and one set can be labeled ‘realis’ and the other ‘irrealis’. But this is not the case for all languages, as we will see in Soq.
4.1. Clause chaining in soq
Soq medial verbs mark switch reference, but not in the typical way. When two verbs in a chain have the same subject, the first is marked with a same-subject suffix.
Here the subject of pumun ‘we took’ is the same as the subject of waŋumbi ‘we went up-river’, and this identity is marked by the same-subject suffix on pumun. Note also that the tense marking on the final verb matches the interpretation of the medial verb: pumun has far-past time reference.
When two verbs in a chain have different subjects, though, the matter is somewhat more complicated. Essentially, the change in subject is marked by inflecting the first verb as a final verb. So in 21, the first verb nainjalbi ‘he fed them’ is marked with final far past morphology owing to the fact that the next clause has a different subject. Ebes ‘they ate and’, however, bears medial morphology because it has the same subject as the following verb, marubi ‘they talked’.
[End Page 725]
An obvious question is whether this construction is properly analyzed as (i) a clause chain in which a verb bearing final morphology, nainjalbi ‘he fed them’, serves as a medial verb marking different-subject switch reference, or (ii) a series of two independent sentences, nainjalbi ‘he fed them’ and ando ebes marubi ‘they ate (it) and spoke’, which are simply juxtaposed. We leave this question for future discussion.
Rather, we concern ourselves here with the clear cases of clause chaining in Soq, shown by same-subject marking. Soq has two paradigms of same-subject suffixes. The first is shown in Table 9.
This set, the hodiernal same-subject set, is used only when the final verb is inflected for hodiernal tense, as in 22.
This paradigm is frequently suffixed with the linking form -qo, in which case the -kem and -ke allomorphs are used. Verbs with -qo often indicate discontinuity between their clause and the following clause. This discontinuity can be, for example, a temporal delay between the two events or the beginning of a new phase of activity, as in 23.
Analyzing -qo as a simple marker of discontinuity, however, does not account for its full range of functions, and this is an area that requires more research. For our purposes, though, we observe that -qo does not affect the scope of tense information from the final verb.
When the final verb belongs to any tense or mood category apart from the hodiernal, medial verbs employ the other paradigm of medial suffixes. This paradigm, given in Table 10, we call the nonhodiernal same-subject (nhod.ss).
These forms are used with both of the tenses that are semantically adjacent to the hodiernal, the yesterday past (24) and the future (25), as well as with the far past and the imperative mood. This means that these forms cover two periods (the past up to last night, and the future from now) that are not contiguous on the timeline.
[End Page 726]
We have established that medial same-subject verbs make a two-way distinction between hodiernal and nonhodiernal, whereas final verbs make a four-way distinction. This pattern is illustrated with all four tenses in 26.
We see the hodiernal same-subject suffixes just in 26c, when the final verb is in the hodiernal, and in all other examples we find the nonhodiernal same-subject suffixes. There is no good translation of 26c; the Soq example covers the range of the two English versions.
4.2. Clause chains ending in s- ‘stay’
The crucial question is what happens when the final verb of the chain is s- ‘stay’. Given that its tense system is out of step with the rest of the verb inventory, we must ask whether the boundary between the hodiernal same-subject and the nonhodiernal same-subject of the medial verb is repartitioned along with that of the final verb, or whether the medial verb retains ‘its own’ system, that of all verbs except s- ‘stay’. Here are the key examples.
The hodiernal versus nonhodiernal distinction that medial verbs make, illustrated in 26, is not affected when the final verb is s- ‘stay’ (27). When s- exceptionally uses the yesterday past tense for events occurring earlier on the day of utterance (for which every other verb would stand in the hodiernal), any preceding medial verb takes the semantically appropriate hodiernal same-subject form (27b). In other words, the form of the medial verb is determined by semantic considerations and not directly by the tense form of the final verb. The special properties of s- ‘stay’ have no syntactic consequences. This fact creates a possibility that is not allowed with other verbs: a hodiernal same-subject medial verb followed by a (repartitioned) yesterday past verb, as in 27b. Thus in one corner of the grammar of Soq—specifically, in a same-subject clause chain [End Page 727] that ends in s- ‘stay’—there is an extra distinction in time reference that the language does not make anywhere else, between now (27c) and earlier today (27b).12
These facts pose a challenge to some formal accounts of clause chaining, and we need to discuss these before returning to the significance of the clause-chaining data for repartitioning. Some linguists analyze clause chains as a series of coordinated units that all receive their tense specification from a node which dominates them all. Thus Foley (2010:39), working in lexical-functional grammar, considers Papuan clause chains to consist of ‘a single I[nflectional] P[hrase] projected by a single I head dominating a string of coordinated S constituents’. Keine (2013), working in a minimalist framework, analyzes same-subject medial verbs as coordinated VPs and different-subject medial verbs as coordinated vPs. In both cases, for Keine, the coordinated elements receive their tense specification from a single dominating tense phrase node.
All of these analyses raise a question, though: What is the tense specification on the final verb sgimbi in 27b? Or put another way, what kind of feature might occupy the dominant node in 27b, to account for both the hodiernal form of tiŋ ekom ‘stand’ and the yesterday past form of sgimbi ‘stay’? We review two potential solutions below, and then propose our own analysis.
One obvious suggestion is to posit that Soq actually has five tenses, not four, and to say that the forms involved exhibit massive syncretism.13 On this analysis, the grammar of Soq makes a distinction between the yesterday past, the today past, and the present (in addition to the future and the remote past). Yet no Soq verb distinguishes five tense values. This would mean that all verbs apart from s- ‘stay’ show syncretism between the today past and the present forms, and s- alone shows a different pattern of syncretism, namely between the yesterday past and the today past. Recall that this would not be straightforward syncretism, since each tense involves a set of forms. Such an account captures the morphological facts about Soq verbs by brute force. It suggests why medial verbs are unaffected by repartitioning in final verbs, because in this view there actually is no repartitioning—only syncretism. Each of the clause chains in 27 has a different tense value, and every verb in these chains is inflected for that value; the forms involved just happen to be syncretic.
This analysis can account for the data, and it is compatible with the theoretical notion that morphosyntactic features reside at nodes in a tree and dominate lower nodes directly, imparting their features to them. But it comes at a high cost. We must allow that every single verb in Soq shows syncretism between the present tense and the today [End Page 728] past—every verb, that is, except one. And that verb, as it happens, exhibits syncretism between the today past and the yesterday past. This strikes us as highly implausible, and, as we now demonstrate, the analysis is not supported by the evidence.
Consider first the majority of Soq verbs. If they showed syncretism between today past and present, we should expect to find evidence for two discrete values, but there is no evidence for this in Soq. For example, it has been observed that, when faced with syncretic morphology, speakers often compensate for it in some way. Thus some Russian nouns show syncretism between nominative and accusative case. When both core arguments are expressed by such nouns, the usually free word order ‘freezes’ to SVO order as a way to disambiguate grammatical function, according to Bloom (1999).
The evidence of Soq points in quite the other direction. The indicators of syncretic values are lacking, and on the contrary speakers move seamlessly between present and past. If the Soq hodiernal were truly two separate tense values that only happen to be syncretic, we would expect to see evidence for switching between the two, and compensation for the lack of specificity, perhaps with time adverbs or in some other way. But this is not what we find. Soq has no adverb meaning ‘now’ that could serve such a disambiguating function, since the closest thing, the adverb pala, refers to a time span that includes the present moment but is unspecified for how much other time it includes: it can range from ‘right now’ to ‘today’ and on to ‘these days’. Moreover, actual usage demonstrates that there is no linguistic boundary between two syncretic values here. The following exchange is between Tun Lynn Yanam and Tiga Julita Yum. Lynn has seen the Pear film (Chafe 1980) and told Julita about it, and now they are watching it together. They move between present and past time reference effortlessly, as indicated by the English translations, without any overt marking to signal the transitions.
So the indicators we might expect to find if syncretism were involved are conspicuously absent. We now turn to positive evidence for a single value, which we find in speakers’ behavior in breaking clause chains. When relating a series of events, speakers must decide when to chain verbs into one another, and when to end a chain and begin a new one. When relating events with the same tense value, speakers tend very strongly to package them into the same clause chain; in elicitation this was consistently done. Conversely, speakers never chain from one tense value into another. This provides a test for the status of the hodiernal. We find that sequences of events that begin earlier today and end now, such as 29 and 30, are consistently chained together, again suggesting that ‘earlier [End Page 729] today’ and ‘now’ are not two different tense values that are syncretically merged, but are in fact a single value.
The negative evidence (the lack of linguistic material to signal transitions between now and earlier today) and the positive evidence (the treatment of clause chains) both point to the conclusion that the ‘massive syncretism’ analysis must be rejected.
Another possible analysis would be to posit some kind of proto-feature at the dominating node (the inflectional phrase for Foley and the tense phrase for Keine). This proto-feature would be specified for time reference (that is, the semantic content of when the events in the clause chain took place) but not for tense. In this model the proto-feature would achieve full realization only when it combines with a lexeme. This model would then require the lexical entries for verbs to contain instructions for how the proto-feature should be realized. In a case like 27b, the proto-feature would contain the time reference ‘earlier today’. Most verbs would instruct the model to realize that proto-feature with hodiernal morphology, but s- ‘stay’ would instruct it to realize the proto-feature with yesterday past morphology.
This approach is not significantly different from the syncretism approach that we have just rejected. If anything, it is more costly from a theoretical perspective. Instead of each lexeme specifying a pattern of syncretism, it specifies a spellout procedure for a proto-feature. But the generalization that features reside at nodes in a tree and percolate down to be spelled out by affixes is severely weakened. We must resort to a notion of a ‘proto-feature’ that contains semantic properties, but not morphosyntactic ones, until it combines with a lexeme. But even so, this approach is essentially equivalent to the syncretism approach. To account for the examples in 27, we still need the dominating node to be able to specify three values of the ‘proto-feature’: present, today past, and yesterday past. After that, the mechanism of achieving morphological spellout via lexically specified spellout rules is analogous to lexically specified syncretism. One of the few merits of this approach is that it salvages the notion—which we consider correct—that Soq has only four tense values. These solutions, though they strike us as awkward, could be made to work, albeit at great cost.
For the vast majority of the grammar of Soq, four tense values are sufficient: remote past, yesterday past, hodiernal, and future. Even with s- ‘stay’, four values are enough, if we accept that the semantic boundary between two of the values has been repartitioned. It is only in one very specific context—a same-subject clause chain that ends in s- ‘stay’—that Soq can express a five-way distinction between remote past, yesterday past, today past, present, and future. The ideal theoretical account would capture all of these facts: the near-complete adequacy of four feature values, the presence of some idiosyncratic repartitioning with one verb, the emergence of a five-way distinction in one particular construction, and also the rarity of that five-way distinction. We now provide such an account, using the mechanism of default inheritance (Gazdar 1987, Goldberg & van der Auwera 2012) and the distinction between morphosyntactic and morphosemantic features (Corbett 2012:49–50, Spencer 2013:219–32). [End Page 730]
Default inheritance is a way of modeling the relations between facts about language, at different levels. We state a general rule, a default, and this applies in every instance except when it is overridden by a more specific rule or an exception. We might give the default properties of English verbs, including word-order constraints, and the fact that their past tense is formed with -ed. We need override statements for the past tense of bring, go, and other irregular verbs, but the regularities about word order, present tense, and so on all hold. These verbs are irregular, but in a tightly constrained way.
Default inheritance can neatly capture the facts about Soq s- ‘stay’, but in order to show how this is so we first need to do a little formalization with normal Soq verb morphology. We use the formalism of construction morphology (Booij 2013, 2016), called a second-order schema, but the point could be made equally well using any morphological theory where default inheritance plays a central role, such as network morphology (Corbett & Fraser 1993, Fraser & Corbett 1997, Brown & Hippisley 2012).14 Soq verbs have many inflectional possibilities, each of which expresses particular morphosyntactic or morphosemantic features. These constructions are interrelated in that the same word (in principle, any Soq verb) can enter into all of them, and the set of these constructions expresses various paradigmatic oppositions. Focusing for now only on the difference between a 2sg/1 yesterday past verb and a 2sg/1 hodiernal verb, the relation between two constructions can be expressed as in 31.
(31) <[xi-kimbi]ω,j ↔ [Vi, +2sg/1] ↔ [SEMj happened yesterday]> ≈
<[xi-ekum]ω,k ↔ [Vi, +2sg/1] ↔ [SEMk happened during the period from earlier today to now]>
Here we have two constructions in a paradigmatic relationship (expressed by ≈). Each construction is set off by <angle brackets> and is itself composed of formal information, featural information, and semantic information. Each kind of information is set off in [square brackets], and the fact that these three levels of information are all part of the same construction is indicated by the double arrows (↔). The formal specification of the first construction indicates that some form x combines with the suffix -kimbi and that this combination forms a morphological word (ω). Featurally, this combination has the person/number feature +2sg/1 (this is actually a syncretic combination of three person/number values, but we ignore that fact for the sake of simplicity). Semantically the construction indicates that something happened yesterday. Relationships between the different levels of information are expressed by coindexation. So the person/number feature 2sg/1 is, in the grammar of Soq, a feature of verbs, so the featural information specifies that this construction involves +2sg/1 on a verb and not some other part of speech. The verb in the featural part of the construction is coindexed with the x in the formal part of the construction, indicating that x has to be a verb—in other words, that it is only verbs that can enter into this construction. Similarly, in the representation of semantic information, SEM is coindexed with the formal morphological word, indicating that it is the meaning of [x-kimbi], whatever that is, that happened yesterday.
The second line in 31 contains essentially the same construction, but for the hodiernal tense instead of the yesterday past. All that is different is that x now combines with -ekum instead of -kimbi, and that it refers to an event from now or earlier today, not one from yesterday. Note that x is coindexed across both constructions; this captures the fact that the same stem can enter into both. This is the general idea with these kinds of representations: any verb stem in the language can occupy the x slot in the formal representation and contribute its semantics (SEM) to the semantic representation, and such constructions are well-formed. [End Page 731]
All of this is simply a rather technical way of expressing some very plain facts about Soq verb morphology, which we described in §2 above. The payoff for this formalism comes when we examine the behavior of s-, and especially its behavior in clause chaining. First, here is a second-order schema for s- that corresponds to the generic one given in 31 above.
(32) <[si-gimbi]ω,j ↔ [Vi, +2sg/1] ↔ [STAYj happened during the period from yesterday to earlier today]> ≈
<[si-ikum]ω,k ↔ [Vi, +2sg/1] ↔ [STAYk is happening now]>
There are a few differences between 32 and 31. First and foremost, 32 contains an actual lexeme, s-, instead of the variable x in 31. For this reason 32 also specifies particular allomorphs of the 2sg/1.ypst and 2sg/1.hod suffixes—the phonologically conditioned -gimbi and the lexically conditioned -ikum—rather than the more unmarked allomorphs used in 31. These are the ordinary results of plugging any lexical verb into the schema represented in 31. But there is also another, more significant difference: the semantic information conveyed by the schema in 32 is different from that conveyed by 31. Instead of distinguishing ‘yesterday’ from ‘earlier today to now’, the two schemas in 32 distinguish ‘yesterday to earlier today’ from ‘now’. This way of representing Soq morphology captures all of the relevant facts—that s- is a perfectly ordinary verb in terms of its form and (as we will see below) its morphosyntactic features, and that it exhibits peculiarities in its semantic behavior. But not only does this representation capture the relevant facts; it also accounts for them in terms of a common crosslinguistic feature of morphological systems, which might be termed the ‘idiosyncratic subschema’ (Masini & Audring 2019).
So far we have shown that the notion of default inheritance captures the relevant facts about s- in a crosslinguistically plausible way. When we examine the syntactic behavior of this verb, we see why the distinction between morphosyntactic and morphosemantic features becomes important. Simply put, the difference between these two kinds of features is that the syntax of a language is sensitive to morphosyntactic features—they might be involved in agreement, or government, or other similar phenomena—while morphosemantic features are invisible to the syntax (Corbett 2012:49–50). In our schematic representation, morphosyntactic features are given in the second set of square brackets, while morphosemantic features are given in the third. (Even though the terminological distinction might suggest that morphosemantic features have meaning while morphosyntactic features do not,15 this is not the case. Both kinds of features typically have meaning, but only one kind is referenced in the syntax.)
This distinction is necessary to account for the clause-chaining data. In order for a Soq same-subject clause chain to be grammatical, all of its verbs must have compatible morphosyntactic features. In our analysis, only person/number features are morphosyntactic, because only they must match between a medial verb and the following verb. Tense features, by contrast, are morphosemantic: they convey certain grammatical meaning, but they do not interact with the syntax above the level of the word. Each verb in a Soq clause chain thus receives its own tense marking on the basis of its own semantics, [End Page 732] and tense ‘agreement’ in clause chains like 26c and 27b is not actually syntactic agreement. Rather, it is the natural result of verbs in clause chains being semantically coherent with each other.
This account, combining the notions of default inheritance and morphosemantic features, satisfies all of our analytic desiderata. It preserves the observation that Soq has four tenses. It explains why even the one verb that behaves irregularly with regard to tense also only has four tenses, and why it uses the same morphology as regular verbs: it has inherited the general four-way tense schema but specified its own idiosyncratic exceptions. Our analysis also explains how a five-way distinction can arise in one Soq construction without having to posit that the five-way distinction is a general feature of Soq grammar. The distinction simply emerges when a same-subject medial verb, making the normal tense distinction that these verbs make, is combined with s- ‘stay’ as its final verb, which has repartitioned the boundary between two of the four tenses that final verbs inflect for.
Note that our analysis accomplishes all of this without proposing any new theoretical machinery. All that we needed were analytic and theoretical concepts that have already been found necessary to account for other languages.
4.3. S- ‘stay’ as a medial verb
We now ask what happens when s- ‘stay’ occurs as a medial verb. There are three areas to investigate. Recall that in the different-subject condition, the medial verb uses the same forms as when final, and it is unclear whether this should be counted as clause chaining. Nevertheless, we should ask whether repartitioning is found in this context, as it is when s- ‘stay’ is actually final. If repartitioning is a matter of the inflectional paradigm rather than of the syntax we expect that it will. Next, and most importantly, we must ask what happens in the same-subject condition; recall that here there is a different subparadigm, contrasting hodiernal with all other tenses.
First we observe that when used as a different-subject verb, if that is what the construction really is, s- ‘stay’ also exhibits repartitioning. The clause chain in 33 has earlier-today time reference, as shown by the final verb glken ‘s/he went, is going’, but s- ‘stay’ is in the yesterday past (repartitioned).
Moving on to same-subject clause chains—the key point—we observe that s- ‘stay’ is not repartitioned when it is a medial verb. Recall that medial same-subject verbs distinguish hodiernal from nonhodiernal, which means that they also mark the boundary between yesterday and earlier today. Unlike when s- ‘stay’ is final, this boundary is not moved when s- is medial: the chain in 34a illustrates the nonhodiernal form with yesterday time reference, and that form is not used with earlier-today time reference (34b).
[End Page 733]
A natural question to ask is how this fact affects clause chaining. Recall that verbs with the same tense value can be chained together, but verbs with different values cannot. Thanks to repartitioning, the tense systems of the two verbs may not be in sync. While this is an area that requires more research, the rule appears to be that when a medial verb is produced, the time reference of the following verb is checked against the medial paradigm of the first verb. If the values are the same, the first verb is inflected medially; if not, it is given final inflection. This can produce some unexpected chains, as shown below.
In 35a both verbs are marked hodiernal, but one might not expect the first verb to be chained into the second, given that for the second verb, the hodiernal is exclusively a present tense. But since the first verb checks its own medial paradigm, and medial s- ‘stay’ does not repartition, we see clause chaining. Similarly, 35b shows a hodiernal verb chained into a yesterday past one. This is made possible by the fact that jaqli ‘laugh’ does not check the following verb’s paradigm, but its own; and its own paradigm assigns hodiernal tense to earlier-today time reference. Note that the observations made above, that the syntax does not have access to the repartitioning behavior of s- ‘stay’, are not affected by this analysis, since no verb’s medial paradigm shows repartitioning.
These surprising data show that we need to specify repartitioning more narrowly: s- ‘stay’ is not repartitioned when it serves as a true medial verb, with same-subject morphology.
4.4. Clause chaining with s- ‘stay’: summing up
We see now that repartitioning is a phenomenon restricted to one verb in Soq. This verb can serve as a full lexical verb and it can also occur in more auxiliary-like guises, as in 17 and 18. In each guise it can be repartitioned. It is also repartitioned when it is used with final morphology as a different-subject medial verb. However, when it serves as a true medial verb, with same-subject morphology, it is not repartitioned. Repartitioning thus affects only one subparadigm of one word: the final inflections of s- ‘stay’. And it affects just one boundary: that between yesterday past and hodiernal.
We are now ready to make another set of crosslinguistic comparisons, bearing in mind the featural properties of Soq—in particular that tense is a feature that has only four values. In the next section we examine various languages in which apparently similar phenomena turn out to be different from Soq precisely because of their featural properties.
5. The nature of repartitioning
The clause-chaining data have proved highly significant for understanding repartitioning. It turns out that the syntax does not have access to the unique tense behavior of s- ‘stay’. Hence we are not dealing with an additional morphosyntactic feature value. We know that the repartitioned semantic values of the hodiernal and the yesterday past need to be dealt with in terms of the feature system, [End Page 734] however, since the semantic change affects two whole sets of forms. In other words, every final, morphologically hodiernal form of s- ‘stay’ behaves alike in repartitioning the semantic value of ‘hodiernal’. We noted that the forms of s- ‘stay’ are fully regular, which means that repartitioning is not covered by any of the recognized morphological phenomena concerned with form.
Because there are no morphological analogues to be found in terms of the form of the inflected lexeme, we must look for analogues in terms of the interaction of the lexeme with the feature system. These are what we referred to earlier as ‘requirements of structure’ and in particular ‘the grammatical dimensions of the inflectional paradigms’ (§3.6). As we conduct our search, we should bear in mind three key characteristics: (i) this is a featural issue; (ii) the syntax is not affected, hence this is not a question of a morphosyntactic feature, but a morphosemantic one; (iii) the different behavior is found in just one lexical item versus all others of the same part of speech. The degree of lexical idiosyncrasy is extreme in Soq, since just one verb is involved; the situation would still have been very interesting if a small set of verbs had been involved, but the limitation to a single verb is particularly noteworthy.
We begin our investigation by looking at other featural mismatches, which might be analogous to Soq (§5.1). This leads us to a typology of such mismatches, in which Soq occupies a unique position (§5.2).
5.1. Featural mismatches as potential analogues to soq
To structure our comparisons, it is helpful to abstract away from the particular Soq facts and to create a vocabulary for comparing Soq repartitioning with other phenomena. We observe that, in the simplest and most straightforward case—the canonical case, in fact—feature values and forms map onto each other in a one-to-one way. Using the Latin forms from Table 7 as an example, we see in Table 11 that the person/number values 1sg, 2sg, and 3sg are each marked by a single form in both of the relevant contexts (in this case, active and passive voice).
This kind of situation is illustrated abstractly in Table 12. A feature is marked in two contexts, I and II; forms a, b, and c, which mark the feature in context I, correspond in a one-to-one way to forms x, y, and z, which mark it in context II.
It is a well-known fact of language, though, that the straightforward mapping in Table 12 is not always how things work. Sometimes we see one-to-many or many-to-one correspondences, as with overdifferentiation and syncretism, respectively. For example, the person/number values 1sg, 2sg, and 3sg are distinguished by verb suffixes in the Soq future tense, but 1sg and 2sg are syncretically merged in the far past, as in Table 13. Situations of this type are represented abstractly in Table 14. [End Page 735]
But it is also possible to have many-to-many mappings, as diagrammed in Table 15.
This might provide an analogue to Soq tense, but before discussing whether it does, we present a well-established example. We consider the notion of a nonautonomous feature value, using data from the system of core cases in Classical Armenian. The key forms of the Classical Armenian noun paradigm are given in Table 16 (Klein 2007: 1053); we omit further values. We find two singular forms, am and ami, and three plural forms, amkʿ, ams, and amacʿ. Their case values map onto each other in the following way.
Given that, within a number, there are at most three different forms in this part of the noun paradigm, we might expect to find evidence for three case values. Yet when we examine the syntactic environments in which they occur, we see that there are four case values to be distinguished from these forms (Table 17).
Given just the singular, there is no evidence for an accusative (the form is syncretic with the nominative). Equally in the plural there is no unique accusative form. However, we still must recognize the existence of an accusative case, given the requirements of the syntax. The accusative is required by transitive verbs, certain prepositions, and in adverbial expressions (Tumanjan 1971:188, Schmitt 1981:91). The alternative analysis, which denies the existence of an accusative case in Classical Armenian, would require special requirements in each of the three different syntactic environments (specifying the nominative case of singular nouns and the locative of plural nouns). The point of case values is precisely that they generalize over different syntactic environments. [End Page 736] Since the accusative must be acknowledged as a genuine feature value in the Classical Armenian case system, but it has no unique form, it is a ‘non-autonomous value’ (Zaliznjak 1973:69–74).17
It is worth drawing out the key characteristics of this paradigm. There is a nonautonomous value, the accusative, and this value arises from the mapping across another feature, namely number. The evidence is easy to find: we need only look at the number values within the paradigms of individual nouns. That is, we compare the same noun, in the same syntactic environments (including as object of a transitive verb), and we find four different values. Thus contexts I and II of our abstract mapping are here singular and plural. A comparable situation is found in Romanian, where the mapping of gender, again across number, shows the need for three gender values, even though agreeing targets distinguish only two values in each number (Corbett 2012:82–83, 168–69, Loporcaro 2018:92–109). This generalization involves all items that inflect fully for gender and number.
Importantly, Classical Armenian case and Romanian gender are both rather different from the situation with Soq tenses. The many-to-many mappings in Classical Armenian and in Romanian are general characteristics of the system of features; they can be observed by examining the various forms of a single noun, taken at random. By contrast, the phenomenon of repartitioning is evident only when we compare across lexemes, and the comparison must involve s- ‘stay’ and any other verb.18 Moreover, and most importantly, the standard of comparison is not the syntactic context but the timeline.
A second phenomenon that bears comparison with Soq is the case system of various languages of Australia; in these, according to the traditional analysis, pronouns operate on a nominative/accusative axis, while common nouns operate on an ergative/absolutive axis. A particularly clear instance is found in Guugu Yimidhirr, a Pama-Nyungan language of Queensland; the data are from Haviland (1979:47–51, 66–67), discussed in Baerman et al. 2005:42–45. To illustrate this pattern, the forms for the first-person singular pronoun and the noun gabiirr ‘girl’ are given in Table 18. As with Classical Armenian, although there are only two forms in each context, they exhibit a many-to-many mapping with each other, which raises the possibility that they are marking more than two values.
The erg column shows the form a word takes when it is the subject of a transitive clause, the nom_abs column when it is the subject of an intransitive clause, and the acc column when it is the object of a transitive clause. Pronouns have a unique accusative form that is distinguished from what might be called a nominative form, while nouns have a unique ergative form, distinct from what might be called the absolutive. There are two main analyses here: (i) the ‘split-ergativity’ analysis, which proposes two different systems, one for pronouns and one for nouns; and (ii) the syncretism analysis, which proposes three case values here, with two different patterns of syncretism. It is the first analysis that might offer some analogy to repartitioning.
The first analysis, following the forms, has two case values here, for pronouns and nouns, but these two values overlap only partially. It has been argued that such analyses, in general, are inadequate from the syntactic perspective (Goddard 1982). The problem is describing the case requirements of the various types of verb. Thus an intransitive verb would need different case frames according to whether its subject is a pronoun (assign nominative) or a noun (assign absolutive). This argument is normally considered sufficient to say that we should adopt the second analysis, with three morphosyntactic case values for this part of the system, and two patterns of syncretism in the morphology.
Guugu Yimidhirr provides an additional, rather telling, piece of evidence, namely case relations within the noun phrase (Round & Corbett 2017:27–29). Several elements in the Guugu Yimidhirr noun phrase take case marking. In 36, for example, both constituents of the NP ‘that girl’, gabirrinh and nhaamuun, are marked for ergative case.
This fact becomes relevant when we observe that noun phrases with animate referents often include a pronoun as the first element, as in 37.
In examples like this, a pronoun, functioning on a putatively accusative axis, agrees with a noun, functioning on a putatively ergative axis. (This structure undermines any attempt to handle the difference by some sort of differential case marking.) Such examples demonstrate that what initially seem like nominative pronouns are in fact two distinct things—not in form, but in their featural specification. This is because these pronouns can also match absolutive nouns, as in 38.
The same can be said of absolutive nouns, like dyiiraal-gaar above. These can match nominative pronouns, as in 38, or accusative pronouns, revealing that what initially appears to be a single value really is two.
So although Guugu Yimidhirr has, formally speaking, two distinctions within the pronoun (in this part of the paradigm) and two distinctions within the noun, from the [End Page 738] syntactic perspective we see that it has three kinds of noun phrase: ergative, nominative-absolutive, and accusative. And because of the ways that nouns and pronouns participate in these noun phrases, it is clear that they are syncretic: pronouns show syncretism between ergative and nominative-absolutive, and nouns show syncretism between nominative-absolutive and accusative. Thus, to give only one example, 38 above would be more accurately glossed as in 39 below. Here, instead of positing that a nominative feature agrees with an absolutive feature, we simply posit a case value called nominative-absolutive (as in Table 17). Similar reglossing could be done to 37, labeling the noun phrase ergative.
So in Guugu Yimidhirr we have what at first glance appears to be a situation with some similarity to repartitioning: two case values in pronouns and two case values in nouns, with the boundary between them drawn in different places. But closer inspection reveals that both pronouns and nouns actually have three case values, and moreover that they have the same values—they only have different patterns of syncretism. A vital difference is the syntactic argument: the syntax of Guugu Yimidhirr requires reference to three different morphosyntactic case values to account for the syntax of the core arguments. We have described Guugu Yimidhirr at some length; a similar situation, but concerning number rather than case, is found in Mele Fila (see Corbett 2012:180–81). The key difference of these two systems when compared with Classical Armenian (and Romanian) is that we are now comparing across parts of speech (pronoun vs. noun) and not across a feature within the paradigm. The feature values under discussion in Classical Armenian and in Guugu Yimidhirr are nonautonomous, albeit within different types of system.
There are two important distinctions to draw between Soq, on the one hand, and the nonautonomous case values of both Classical Armenian and Guugu Yimidhirr on the other. First, in both Classical Armenian and Guugu Yimidhirr the nonautonomous feature value must be postulated, analytically, as a separate morphosyntactic value. So the Classical Armenian accusative case is distinguishable, in its syntactic behavior, from both nominative and locative, and likewise the Guugu Yimidhirr nominative-absolutive is distinguishable from both ergative and accusative case. In contrast, the Soq ‘earlier today’ tense does not exist at the featural level as a tense value that is distinguishable from the yesterday past and the present. In terms of the feature system, Soq only has four tense values.
The second distinction, which is logically independent from the first, is the domain where we see a nonautonomous or a repartitioned value. Nonautonomous feature values and repartitioning are similar in that both can only be seen through a comparison of two sets of forms, as illustrated abstractly in Table 20. In each case, three apparent values (A, B, and C) are marked by two apparent forms (x and y), such that value B can only be distinguished from A and C by comparing the two rows, which represent two different contexts of marking.
In Classical Armenian, for example, the fact that the case value accusative is nonautonomous is apparent only after a comparison of the forms that mark singular case with [End Page 739] those that mark plural case. And in this respect, each of Soq, Classical Armenian, and Guugu Yimidhirr is distinct from the others. In Classical Armenian we see a nonautonomous value (accusative case) when we compare across another feature (number); in Guugu Yimidhirr we see a nonautonomous value when we compare across parts of speech; and in Soq we see repartitioning across lexemes (in fact by comparing a single verb with every other verb in the language).
5.2. A typology of mismatched features
In order to describe the Soq data adequately we have invoked three parameters, each of which can vary. These can therefore serve as the basis for a typology of mismatched features. We shall see that in this typology there are several unattested systems. This is perhaps surprising, but it is not at all problematic. We take one of the roles of typology to be the establishment of what is conceivable, as a logical prior to what is possible (Bond 2019, Round & Corbett 2020). Given the existence of Soq, other related systems are conceivable, as we shall see. The fact that they are not attested, while Soq is, highlights the particular interest of the Soq system.
The three parameters that underpin a typology of mismatched features are:
(i) Using the abstract language of Table 20, is B a genuine feature value or not? If it is, as with Classical Armenian and Guugu Yimidhirr, then the phenomenon is not a case of repartitioning but of syncretism. But if B is not a genuine feature value, then we have two feature values, A and C, with repartitioning of the boundary between them.
(ii) What is the domain where we observe the mismatch? So far we have observed mismatches across other features (Classical Armenian), across parts of speech (Guugu Yimidhirr), and across lexemes (Soq).
(iii) What kind of feature is involved: morphosyntactic or morphosemantic? Both Classical Armenian and Guugu Yimidhirr case are morphosyntactic; Soq tense, as we established in §4.2, is morphosemantic.
The typology created by these parameters is shown in Table 21.
In Table 21 there is one column for mismatches involving morphosyntactic features that do not qualify as repartitioning—this is where Classical Armenian, Guugu Yimidhirr, and their typological kin are placed—and two columns for genuine repartitioning. We also distinguish three rows: one for mismatches across other features (as in Classical Armenian), another for mismatches across parts of speech (as in Guugu Yimidhirr), and the last for mismatches across lexemes. This represents, in a sense, a progression from a more general domain (since features can range across multiple parts of speech) through less general domains. [End Page 740]
This table clarifies several points. First, since column Y and column Z are a matching pair—one morphosyntactic, the other morphosemantic—one may wonder why there is not a morphosemantic column to match column X. The answer is that in order for a mismatched feature to not qualify as repartitioning, it must be morphosyntactic. Going back to the abstract labels of Table 20, in order to tell that B is a genuine feature value, it is necessary to look at syntactic processes like government and agreement. This means that, by definition, these features must be morphosyntactic, and therefore column X has no morphosemantic equivalent.
A salient point about Table 21 is that six of its nine cells are empty. This emptiness may be an indication of just how remarkable the Soq phenomenon is, or it may just indicate a gap in the research record. Either way, it will be instructive to describe the kinds of morphological phenomena that would fit into each of the empty cells. This account provides directions for future research, whether to find the missing systems or to explain why they do not occur. The system we have given the arbitrary label Q has a nonautonomous value, based on the mismatches across individual lexemes. Such a system might, for instance, be like Guugu Yimidhirr except that the ergative pattern is manifested in only some nouns. In this way, for the same reasons we describe for real Guugu Yimidhirr, the nonautonomous nominative-absolutive case value would need to be posited, but its existence would arise from a comparison of lexemes within a single word class.19
System R involves morphosyntactic repartitioning across another feature. It can be envisaged for a language with case and number, where number has three values: singular, dual (for exactly two), and plural. In the genitive case only, dual marking is used to refer to two or three entities, and plural marking is interpreted as ‘four or more’ instead of the usual ‘three or more’. Verbs in this language show number agreement with arguments, including, in some contexts, genitives. The dual agreement marker on the verb follows the repartitioning that takes place with nouns in the genitive; that is, if the dual is used of three entities, verbal agreement will also be dual, not plural.
System S is defined as morphosyntactic repartitioning across parts of speech. It could be like the hypothetical system R language except that nouns distinguish singular, paucal (two or three), and plural (four or more) in all case values. Pronouns, by contrast, distinguish singular, dual, and plural. Verbs still agree with nominals, and the dual agreement marker on verbs has dual meaning when agreeing with a pronoun and paucal meaning when agreeing with a noun.
System T is morphosyntactic repartitioning across lexemes. Imagine a language with three number values: singular, dual, and plural, and in which verbs agree in number with their arguments. On a subset of nouns, dual marking is interpreted as referring to two or three entities, while plural is interpreted as referring to four or more. On a verb agreeing with such a noun, the dual agreement form participates in this repartitioning. [End Page 741]
The nonexistence of the systems in column Y (or it may be wiser to say their extreme rarity) can be attributed to a single root cause. Morphosyntactic feature values can be used inherently or contextually (morphosemantic features are always used inherently; Corbett 2012:67). We can see this opposition in agreement: in the Russian example Marina prišla ‘Marina arrived’, Marina is inherently singular and feminine (it denotes a single female), while prišla ‘arrived’ is singular and feminine only contextually (it shows agreement with a feminine singular controller). Importantly, in most cases, these contextually marked values—that is, values marked for morphosyntactic reasons—can occur on words that are separated from their controller, or even without an overt controller. And by default, the grammatical meaning of these values is compositionally combined with the lexical meaning of the host: prišla ‘arrived’ is interpreted as ‘arrived’ with a feminine singular subject. This default behavior of the contextual use of feature values serves to maintain regularity of grammatical meaning. Thus any pressure to repartition particular feature values in a morphosyntactic system is likely to be met by strong pressure to maintain regular rules of semantic interpretation by the contextual use of those feature values. To take the example of system T, a verb with plural inflection has the default interpretation of having a subject denoting three or more entities, which would strongly inhibit the repartitioning of a subset of nouns.
The remaining two unattested systems involve a morphosemantic feature. System U involves morphosemantic repartitioning across another feature. Such a system might be identical to that in the language described in system R, except that verbs do not agree with nouns. System V is defined as morphosemantic repartitioning across parts of speech. This is like system S, except that, again, verbs do not agree with nouns. This language also lacks third-person pronouns. (We posit this to avoid the question of whether anaphoric reference is a syntactic process.)
We suggest that the rarity or absence of systems U and V is in part due to pressures that can be seen as somewhat similar to those that work against systems R, S, and T. In U and V, the repartitioned feature value has to be interpreted either across another feature value (system U) or across parts of speech (system V). Again this interpretation would be, by default, compositional. We would expect, for instance, that computing a particular value across another feature or across parts of speech would lead to pressure for a uniform interpretation, since there is nothing in another feature or in different parts of speech that would be expected to induce repartitioning. In addition, system V would be even less expected because it requires ‘the same’ feature on two different parts of speech, but there must be no morphosyntactic behavior linking them (if there were, we would have system S). This line of reasoning could explain both why the Soq system is attested—unlike, to the best of our knowledge, any of our hypothetical systems in columns Y and Z—and why it is so rare. We recognize a strong functional pressure to regularize the interpretation of feature values; this militates against semantic irregularity in inflectional morphology and accounts for the rarity of repartitioning. But it would seem that the interpretation of feature values across lexemes is less subject to the pressure of compositional interpretation than in systems U and V. However, given the extreme rarity of the Soq system, we should not speculate further.
6. Discussion and conclusion
We have surveyed a wide variety of morphological phenomena and found that Soq tense marking does not match any of them. In other words, repartitioning does not fit into the existing typology of morphological phenomena. In part this is because many morphological phenomena involve idiosyncrasies of form, but the relevant affixes in Soq are, at least formally, fully regular. The irregularity [End Page 742] is found in the feature system. However, a careful comparison of repartitioning with featural phenomena in other languages also failed to turn up any analogues. We therefore conclude that the phenomenon we describe is, at least to date, unique.
To summarize, tense is marked on Soq verbs via sets of subject agreement suffixes. There are four such subparadigms, one for each of the four Soq tenses. When we look specifically at the verb s- ‘stay’, the normal semantic values for two of the subparadigms change: the boundary between the hodiernal and the yesterday past is redrawn. This repartitioning occurs in its final verb forms (and hence also in different-subject use), but when it occurs as a medial same-subject verb, it behaves like any other.
In many clause-chaining languages, tense marking on final verbs governs particular forms of medial verbs. Soq has such clause-chaining structures, yet repartitioning in the final verb does not affect the form of any preceding medial verbs. Both verbs take the ‘right’ tense independently, according to the timeline, not by any relation between them. In other words, clause chaining in Soq is semantically rather than syntactically constrained. These data are problematic for some accounts of clause chaining (Foley 2010, Keine 2013), but we propose an analysis that avoids those pitfalls.
We used our analysis of repartitioning as the basis for a typology that related it to certain other phenomena, such as the nonautonomous accusative in Classical Armenian and split ergativity in Guugu Yimidhirr. The typology proposed is mostly empty—it offers six types of repartitioning, only one of which is attested. It will be interesting to see, in the coming years, whether this is a true picture or whether it is simply an artefact of insufficient descriptive coverage. Nevertheless, we were bold enough to take the nonattestation of other kinds of repartitioning as a fact in need of explanation, and suggested some possible reasons for it. And in any case, repartitioning must at least be very rare, so these explanations account for its rarity.
It is worth stepping back to consider why Soq repartitioning is so striking. One way of appreciating its unusual behavior is through Stump’s distinction between content and form paradigms.20 A lexeme’s content paradigm constitutes its ‘interface with syntax and semantics’ (Stump 2016:104). It determines the distinctions available to the lexeme. In the canonical case, a lexeme’s content paradigm maps straightforwardly onto its form paradigm, which is the basis for defining its inflected forms. The content paradigms of lexemes belonging to the same syntactic category are also, in the canonical case, isomorphic (2016:113). That is, we expect members of the same part of speech to make the same featural distinctions. There is a ‘wide variety of ways in which the relation between words’ content and form may deviate from a relation of pure isomorphism’ (2016:115), several of which we discussed above (§3). Morphological research has tended to focus on irregularities in the mapping between the two paradigms. However, the content paradigm itself brings its own issues, though these are as yet rarely treated (but see Bonami & Boyé 2019:188–89 for a recent example). For example, it is becoming clear that the semantic specifications of a paradigm cell and its external morphosyntactic specification do not have to be identical. This is demonstrated (i) at the level of specific lexemes, notably by pluralia tantum nouns in various languages which are semantically singular but control plural agreement, and (ii) at a more general level in languages where plural nouns regularly control singular agreement (see Corbett 2019 [End Page 743] for examples of both types). The essential point here is that we are finding various instances where the content paradigm is not the simple construct it once appeared. It is in this context that we can make progress on understanding repartitioning. We were careful to point out that Soq tense is morphosemantic; it is not visible to the syntax. The problem with s- ‘stay’ is that its content paradigm, with four tense values, maps perfectly onto the form paradigm, but it maps irregularly to the semantics (as diagrammed in Fig. 2). In this way we highlight what is genuinely unusual and surprising about Soq, namely, that one verb is morphosemantically out of line: it maintains a regular set of tense forms, but maps two of them irregularly to the semantics.
This goes against our expectations of the means of irregularity available to individual lexemes, and the ways in which change can occur and bring about these irregularities. We are used to changes in lexical meaning that have no effect on grammatical meaning (the feature system). If island originally meant ‘mass of land surrounded by water’ but comes to mean ‘item that does not allow access to its parts as antecedents for anaphoric elements’, then we have no difficulty in knowing how it will behave with regard to number: it will have singular and plural with predictable meanings. There may be inflectional consequences of a change in lexical meaning (computer mouses or computer mice?), but the feature system is unaffected. Conversely, grammatical meaning can change, but without affecting lexical semantics. Thus, for example, the Gants present tense suffix -c comes etymologically from a serial verb construction that marked stative aspect (Daniels 2015:135–36). It is now a part of the tense system and is compatible with any verb, even the most stative of all (and its own etymological source) cɨ- ‘stay’: cɨ-cɨ-k [stay-prs-3sg] ‘s/he is staying, is there’. It goes across the verb inventory and does not affect the lexical semantics of any verb. Indeed, this orthogonality of lexical and grammatical meaning is a key motivation for the use of features. What we do not expect is that a change in grammatical meaning could lead to the situation where one lexical item is out of step in terms of the system, not the form, of featural meaning. This is why Soq is so surprising.
Department of Linguistics
1290 University of Oregon
Eugene, OR 97403
Surrey Morphology Group
Department of Literature and Languages
Faculty of Arts and Social Sciences
University of Surrey
Guildford, Surrey, UK GU2 7XH
revision invited 17 December 2018;
revision received 6 March 2019;
revision invited 29 May 2019;
revision received 15 July 2019;
accepted 24 July 2019]
Appendix. The diachrony of repartitioning
It is reasonable to wonder which set of tense meanings is archaic in Soq—the general one, with a distinction between a hodiernal and a yesterday past, or the one with a distinction between a present and a recent past—and which is innovative. At this point we are aware of no comparative evidence from other Minjim or Rai Coast languages that bears on the question, so if we are to resolve it we are limited to internal reconstruction. And here we also run into difficulty, since while it is plain that the repartitioned forms are cognate with the un-repartitioned forms—they are formally identical, after all—establishing the likely direction of change is challenging. There are several factors that are relevant, but they point in different directions. We discuss considerations of typology, areality, frequency, and semantics.
From a typological perspective, one might suppose that the un-repartitioned system, as found in most of the lexicon, is probably innovative because the hodiernal is typologically unusual in combining present time reference with past time reference. But in fact, tenses of this kind are an areal feature in the Madang area. Other languages that mark a ‘now and earlier today’ tense, in opposition to a more remote past tense, include Koromu (Priestley 2008:310), Pamosu (Tupper 2012:440), Kalam (Pawley & Bulmer 2011:39), Manat (Daniels 2015:588), and Sirva (Daniels 2015:706). In other languages the analogous tense extends only a few hours into the past, as opposed to a whole day; languages like this include Apalɨ (Wade 1989:166) and Mand (Daniels 2015:480). Still other languages, like Aisi (Daniels 2015:812) and Anamuxra (Ingram 2001:234), include now, earlier today, and yesterday in the time reference of this category. However, while tense categories of this type are common in the area, they are by no means universal: languages with simple present tenses include Amele (Roberts 1987:224), Kobon (Davies 1981:166), Nend (Harris 1990:128), and Siroi (Wells 1979:30). So the areal considerations confound the typological ones, but not conclusively so.
Turning to frequency, we know that frequent items more readily retain older features (Bybee 2001, 2010). Since s- ‘stay’ is a frequent verb in Soq, as in many languages, we might suppose it has preserved an earlier situation. Conversely, since ‘stay’ is such an archetypically stative meaning, it may be that the verb’s aktions-art attracted it to present-tense meaning, and that it is therefore innovative while the rest of the verbal lexicon is archaic. In this case the argument for s- reflecting an archaic situation is somewhat less plausible, since that account would require positing that the entire verbal lexicon underwent a semantic change along with s- in medial contexts, but not s- in final contexts. It seems somewhat more likely that final s- underwent a semantic change to create the situation we describe in this article. Still, this assessment is too speculative to posit an actual reconstruction, and we lack conclusive reasons to prefer one directional hypothesis over the other. Since we also lack external cognates from which to argue, we must remain agnostic as to which partitioning of temporal meanings is innovative and which is archaic.
* We thank our main Soq fieldwork consultants, Mark Kamon, Kalex Kud, Pom Sangi, and Thomas, Lynn, and Andrew Yanam, along with the rest of the Kaliku and Male communities, for making our research possible. The support of the Australian Research Council Centre of Excellence for the Dynamics of Language, and of the AHRC, UK (grant: AH/N006887/1: Lexical splits: A novel perspective on the structure of words) is gratefully acknowledged. We are thankful for the helpful suggestions from audiences at the Australian National University (March 2018), the Surrey Morphology Group (March 2018), the 10th International Austronesian and Papuan Languages and Linguistics Conference (APLL10, Surrey, May 2018), the University of Oregon (March 2019), and the 12th Mediterranean Morphology Meeting (MMM12, Ljubljana, June 2019). Special thanks to Matthew Baerman, Dunstan Brown, Patricia Cabredo Hofherr, and Greg Stump for comments on draft versions, and to the referees for Language who, each in different ways, prompted significant improvements. We are greatly indebted to Megan J. Crowhurst (co-editor of Language) and Carmel O’Shannessy (associate editor). For help in preparing the manuscript we thank Penny Everson and Lisa Mack.
Daniels carried out the fieldwork and produced the basic description of Soq. The significance of the data emerged in discussion between Daniels and Corbett, the latter bringing expertise in the typology of features. The analysis of repartitioning and the writing up are our joint work.
1. This name is spelled in several different ways; we use the most common version, that used also in his biography (Webster 1984). In the references we use the form in the original publication, and a consistent transliteration for items published in Russian.
2. Our translation from the German ‘sprechen schon ein anderes Idiom als die Bewohner’.
3. Grammatical abbreviations here largely follow the Leipzig glossing rules. We use a slash ‘/’ to indicate syncretic forms, for example 2sg/1 = second person singular or first person (singular or plural). The following abbreviations are used: 1: first person, 2: second person, 3: third person, abs: absolutive, acc: accusative, aux: auxiliary, caus: causative, dat: dative, def: definite, dem: demonstrative, dist: distal, ds: different subject, du: dual, erg: ergative, f: feminine, foc: focus, fpst: far past, fut: future, gen: genitive, gen2: second genitive, hod: hodiernal, indf: indefinite, ipfv: imperfective, ipst: immediate past (Mand), loc: locative, lnk: linker, m: masculine, n: neuter, nhod: nonhodiernal, nom: nominative, nom_abs: nominative-absolutive, obj: object, pl: plural, poss: possessive, prs: present, pst: past, sg: singular, spec: specific, ss: same subject, temp: temporal, ypst: yesterday past (‘hesternal’).
4. Examples that are taken from fieldwork recordings are cited to the line in the recording they come from. Thus the example in 4 is from a story about Maclay—and is not from Maclay’s own materials!
5. Note also that some verbs in Soq, such as wit ‘throw’ and basa ‘finish’ in 6, do not themselves inflect, but always occur with an inflected auxiliary. This construction is common in the area (Foley 2018) and has sometimes been called a ‘verb adjunct construction’ in the Papuanist literature (Pawley 2012). In Soq there is a semantically inert auxiliary that serves this function, which is phonologically null in the presence of some suffixes, as seen in 3. We indicate phonologically null as <∅> for convenience, but this does not imply the use of zero morphemes in our morphological analysis.
6. We should also mention the remarkable behavior of English be, which has suppletive past-tense forms that uniquely differentiate singular and plural values (was vs. were). However, there is no additional value in this part of its paradigm: other verbs differentiate past vs. present, and singular vs. plural. The verb be realizes no extra value here; rather, it realizes the number opposition in an unexpected place. It does show overdifferentiation for person in the present.
7. Evans’s (2019) interesting extension of deponency to ‘distributed deponency’ also does not cover repartitioning; distributed deponency involves ‘non-compositional combination of inflectional material from different stem or affix positions’, while repartitioning involves a single set of inflectional forms.
8. While preterite presents are restricted to a subset of verbs, we may find a similar shifting of forms more generally. Thus in Ngkolmpu (Carroll 2016:285), we find the same prefix for recent past in the durative aspect, and for remote past in the imperfective aspect; this is a general regularity. However, the pattern of other affixal material makes clear what the tense-aspect of these forms is. And, most importantly, there is no suggestion of any impact on the grammatical meaning of these cells; hence there is no real analogy to repartitioning.
9. We must also mention the unusual system of Seri, analyzed by Baerman (2016). The verb has four paradigm cells, as is clear from those verbs that have four distinct forms. Many verbs, however, have fewer distinctions, and the possible mappings between inflections and cells are extensive and remarkable. These mappings require a good deal of lexical specification, but there is an overarching parallelism between the range of grammatical meanings and the inventory of available affixes. Even in this highly unusual system, however, there is no evidence of repartitioning. Each verb has four cells, with more or fewer distinct exponents, but no shifting of boundaries. Again, the inventory of grammatical meanings is not affected.
10. A referee suggests abandoning this portion of the definition, but we do not think this move would be justified. We suspect that repartitioning is cognitively manageable precisely because it involves an expansion of the semantic space occupied by one feature value and a corresponding contraction of another feature value. If part of a feature value were reassigned to another, noncontiguous value (say, if the yesterday past took on part, but not all, of the future tense), that would surely be more cognitively burdensome. We think it better for our definition to account for the observed Soq phenomenon and its potential analogues presented in §5 below before envisaging even stranger situations.
11. A referee puts forward the possibility that what we term repartitioning could also be analyzed as a two-step process of pragmatic inferencing. In the first step the lexical aspect of s- ‘stay’ predisposes it to be interpreted with present time reference, and in the second the time reference of the yesterday past is extended to compensate. While this would be a plausible (albeit speculative) way to account for the origin of repartitioning in Soq, it does not hold up as a synchronic analysis. First, repartitioning is grammatically required. If it were a pragmatic inference, we would expect it to be cancelable, but there is no evidence that it is. Second, repartitioning does not happen with all morphological forms of s-, as we will see in §4.2 below. If repartitioning were a pragmatic inference, we would expect it to occur with all forms of the verb. Third, we might expect pragmatic inferencing to be sensitive to the type of use (auxiliary vs. lexical) of the verb, but repartitioning is not affected by this distinction. And lastly, repartitioning does not take place with other verbs that have similar lexical aspect, as shown with uriŋ ‘shine’ below.
Thus there is no evidence to support pragmatic inferencing as a mechanism relevant to the synchronic phenomenon of repartitioning.
12. Note again that this is not overdifferentiation, which is an additional value within the lexeme, since the extra distinction in Soq arises by comparison across two lexemes in a single construction. It is more similar to distributed exponence, the phenomenon in which a morphosyntactic feature value is realized by a combination of forms at more than one morphological site, such that no individual affix can be said to contribute the featural meaning (Caballero & Harris 2012, Carroll 2016). Similarly, in Soq the meaning ‘earlier today’ is not ascribable to the tense inflection on either the medial verb or the final verb; it arises only from a particular combination of them. However, this is not distributed exponence, for three reasons. First, again, distributed exponence occurs within the paradigm of a single lexeme; the Soq phenomenon is observable only through the combination of two different lexemes, in a particular construction. This leads to the second distinction: obligatoriness. The medial verb is not an obligatory part of the Soq construction under discussion, but affixes in a pattern of distributed exponence are. And lastly, there is a featural distinction. In distributed exponence, a combination of affixes marks a feature value. But in Soq, as we discuss below, ‘today past’ is not a tense value that should be distinguished from both ‘yesterday past’ and ‘present’.
13. This suggestion comes easily to mind because we think it natural for a language to have a present tense of the familiar type. But Soq is clearly different (as are several of its neighbors, as noted in the appendix), and we need an analysis that respects this.
15. Indeed, a more accurate term would be semantico-syntactico-morphological, but this term has no future. There is much more to be done in order to understand which features can or must be morphosemantic as opposed to those which can or must be morphosyntactic. Kibort and Corbett (2008) provide a preliminary inventory of the two. The well-established morphosyntactic features are number, person, gender, and case; in addition to these, Corbett (2012:134–10) examines the possible claims of definiteness and respect in certain languages. A key source for the basic distinction is Spencer (2013:219–32).
16. This example is from the Male dialect, in which the 2sg/1.hod suffix is -igum instead of -ikum.
17. This analysis, assuming syncretism, appears to be universally accepted; see for instance, Jensen (1959:48–67), Tumanjan (1971:187–88), Thomson (1975:13–14, 16), Schmitt (1981:90–114), Halle and Vaux (1998), Klein (2007:1052–59), Caha (2013), and Olsen (2017). In line with the discussion of compensation for syncretism in §4.2 above, Daniel Kölligan points out (p.c., 27 December 2019) that Classical Armenian has a preposition z-, used to distinguish objects that are specific (examples can be found in Jensen 1959:146–50). This is an instance of differential object marking, but it does not make the accusative autonomous: it does not have a unique inflectional form. For pronouns the preposition is obligatory. Nonautonomous case values have been found in other languages, including the Estonian accusative (Zaliznjak 1973:70, discussed in Miljan 2009) and the famous case of the Latvian instrumental (Zaliznjak 1973:70–71, Holvoet 2010, and Corbett 2010 and numerous references there).
18. Since s- ‘stay’ does not repartition in medial contexts, repartitioning could be seen by comparing medial and final s-. But given the impoverished tense inventory of medial verbs, comparison across lexemes is necessary to establish the full system.
19. This type does not seem too unlikely, and indeed we have a ‘near miss’. The Dagestanian language Tsakhur has four genders, with agreement in gender on a wide range of targets, including a conjunction (Kibrik 1999:608–20, Corbett 2006:51–52). Owing to regular patterns of syncretism, a typical verb distinguishes only three gender values. So-called ‘strong’ verbs distinguish genders I+II vs. III vs. IV, while ‘weak’ verbs distinguish I+IV vs. II vs. III. If this were the full verbal system, and there were no other targets distinguishing more gender values, we would have an instance of system Q. However, there are also mixed verbs, which are strong for forms based on the perfective stem and weak elsewhere in the paradigm (Kibrik 1999:62, Corbett 2006:82–84). These mixed verbs therefore give evidence for all four gender values. Nevertheless, Tsakhur suggests that system Q is fully feasible.
20. This corresponds to the network morphology use of two default inheritance hierarchies, the lexemic hierarchy and the morphological hierarchy, for which see Fraser & Corbett 1995; the distinction is developed in Brown et al. 1996 and in Brown & Hippisley 2012:54–80. For comparative discussion see Stump 2002:149–53, 178 and Boyé & Schalchli 2016:218–21.