4.1 Aspectual use

The aspectual auxiliaries uɨ and ufu are assumed to be derived from main verbs. Aspectual uɨ is grammaticalized from a verb of existence with animate subjects, which is attested in the current Karimata dialect and inflects in the same way as the auxiliary. The source of ufu is not attested as it is, but is assumed to be preserved in the verb utsɨfu (<* utsi+ufu ‘hit?+put’), which means ‘to put’⁹ and whose conjugation is the same as the auxiliary’s as in Table 1.¹⁰

The grammaticalization from a verb of existence and a verb of putting into aspectual morphemes is widely attested in Japonic languages and therefore is not surprising.¹¹ We now look at the properties of the relevant aspectual morphemes. Table 2 summarizes the interpretations of uɨ and ufu according to the classification of verbs based on Vendler (1967: Ch. 4).

As Table 2 shows, the primary aspectual interpretation of the verb (phrase) to which uɨ attaches is progressive, whereas that of the verb (phrase) with ufu is resultative.

When uɨ takes achievement verbs, however, it bears a resultative interpretation like the Japanese (-te) iru form, which is also grammaticalized from an existential verb with animate subjects. Let us examine the details of the above interpretations in turn.

Accomplishment verbs, which have both agentive subjects and thematic objects, make clear the distinction between uɨ-attached and ufu-attached patterns. While the sentence with uɨ receives an ongoing state interpretation of the predicate as in (10a), that with ufu entails the completion of the event as in (10b). Since the thematic object marks the telicity of this verb class, these patterns give rise to different connotations as to the existence of the object: While the latter entails the existence of the doughnut, the former does not when the tense is present because the event time of the former extends beyond the utterance time (Bennett & Partee 1972; Dowty 1979).

The existence of the thematic object is entailed even if the action is itself done in the remote past. For example, sentence (11) is interpreted as the speaker having the relevant house at the time of speech.

To cancel this entailment, it is required to add a past tense marker -daɨ (here devoiced) as in (12a) or an explicit experiential marker mii as in (12b).

(12a) cancels the existence of the product by shifting the reference time to the past. Without the past tense marker, ufu refers to the utterance time and relates the past event to the present as a resultative perfect as in (11) or as an experiential perfect as in (12b) (in the terminology of Comrie 1976: Ch. 3). Thus, ufu indicates the relevance of a preceding event denoted by the proposition to a situation at the reference time.

Activity verbs, without thematic objects which would exhibit the telicity of events, make it less clear than accomplishment verbs what the target state is. Still, the progressive and resultative contrast is obtained by whether the aspectual slot is filled by uɨ or ufu. When uɨ saturates it, the sentence obtains an ongoing interpretation as with accomplishment verbs, whereas the completion of an event is entailed when it is filled by ufu. In the latter case, the resultative state indicates that the agent acquired ability by doing the relevant activity, that is, the speaker’s ability to dance well at today’s dance competition in (13b).

Some activity verbs do not have an agent as a subject. While the lack of an agentive subject does not affect the use of uɨ as in (14a), it hinders the use of ufu because no subject can be attributed the “ability” evoked by the predicated event. In order for ufu to be used with non-agentive activity verbs, there must be some effect caused by the event denoted by the proposition.

In (14b), raining leaves the ground wet, which licenses the use of ufu. When it is not possible for the event to leave some effect, however, the sentence cannot have an interpretation that makes it appropriate.

The sentence (15b) is anomalous even after daytime.

Due to the absence of the thematic object, the use of ufu seems to extend to an experiential perfect. It is not clear in (16a) whether the agentive subject acquired any ability by reading the book three times and (16b) is appropriate even when the ground is not wet.

The use of uɨ in these examples is acceptable for some speakers, but ufu is always preferred.

Achievement verbs are exceptional in that they receive a resultative interpretation by combining with uɨ rather than ufu.

This seems to mean that uɨ is ambiguous between a progressive and resultative interpretation, but some attempts have been made to derive these ambiguous readings from a unified semantics of one aspectual morpheme in the case of Japanese (-te) iru (Shirai 2000; Ogihara 2020, etc.). Despite some differences in their approaches, it has been agreed among researchers since Okuda (1978a;b) that verbs with (-te) iru describe the situation of the subject. Therefore, the verb that describes the action of the subject receives a progressive interpretation whereas the verb that describes the change of state of the subject receives a resultative interpretation. I simply assume here that the same kind of idea can be applied to the case of uɨ in the Karimata dialect.

“Perfect” interpretations of achievement verbs are also obtained by attaching uɨ to the predicate, which can be considered an extension of the resultative use in (17a).

State verbs (predicates) have no state change and thus no completable event in the internal temporal structure of the lexical entry so that ufu is not allowed to apply to it. On the other hand, state predicates with uɨ bear an interpretation that the relevant state holds temporarily like some sentences of English progressives: the details of this use are discussed extensively in Kinuhata (2018).

Uɨ in (19) seems to take the whole state as the argument and make the duration short, which would suggest that ufu should be able to take the whole state as its argument and make it completive. But this evokes a modal, not simply aspectual, meaning that is treated separately in this paper and will be addressed in the next section.

4.2 Conjectural use

In the last section, we saw that aspectual ufu selects accomplishment and activity verbs as its main verb and rejects achievement and state verbs. This does not mean, however, that ufu is never used with the latter. For example, while the sentence in (20), which contains the state predicate maa ‘tasty,’ is ungrammatical if the speaker has personally tasted the water of Miyako in former days, it becomes appropriate if the speaker does not know (or has forgotten) the taste and conjectures it.

The same contrast is observed with achievement verbs. While the sentence in (17b), repeated here as (21a), is ungrammatical if uttered when the speaker sees the fallen tree, (21b) is acceptable when she has not seen any fallen trees directly and conjectures the situation.

In (21b), an epistemic modal marker padzɨ makes the context clearer and my cosultants prefer that it be added, but the sentence is still acceptable without it. I reserve the term “conjectural use” to refer to this use descriptively.

Given that there is a conjectural use of ufu in Karimata, a question now arises as to whether the conjectural use of ufu is compatible with verbs other than those denoting achievements and states. If ufu is grammaticalized to mark the proposition as conjectured, which would make it fall into the category of modality, then there is no reason for it to select verb types classified based on their aspectual properties. It is, however, difficult to ask speakers whether the use of ufu with accomplishment and activity verbs is conjectural or not. The reason for this difficulty stems from their preference to use other modal clitics such as padzɨ to make the context clear. To illustrate this point, let’s look at the examples in (22). The use of ufu in (22a) is aspectual as we saw in the previous section. What we expect as an example of conjectural ufu is something like the sentence in (22b), possibly without the modal clitic padzɨ. However, if it lacks padzɨ, the sentence sounds like the speaker knows that the subject has read the book, with ufu being interpreted as aspectual. Then, my consultants reported preferring to add padzɨ at the end.

But if we have padzɨ as in (22b), it cannot be decided whether the conjectural meaning of this sentence comes from ufu itself or the conjectural marker padzɨ, which leaves the possibility that ufu is an aspectual marker indicating the result of the event while the conjectural meaning is conveyed exclusively by padzɨ. (So, we glossed ufu as ??? here.)

A key to determining whether a particular use of ufu is conjectural or not is whether it is able to inflect for tense: while aspectual ufu inflects for tense, the conjectural use does not. (23) and (24) are examples of aspectual ufu used with accomplishment and activity verbs respectively. Depending on the reference time of each sentence, ufu inflects as ‘ufu-taɨ’ (past), ‘ufu’ (present), and ‘uka-di’ (volitional future).

In (24), we asked the consultants to utter the sentences with the reference time set to the time of a dance competition and the event denoted by the predicate completed before it.

On the other hand, conjectural ufu never inflects for tense.¹² First of all, it has a strong past-orientedness: consultants have a firm intuition that the event situated by conjectural ufu has already finished. Thus, sentences referring to the future are always judged as deviant as in (25a) and (26a): note that (25) and (26) have achievement and state verbs respectively, with which only the conjectural use is compatible. As for the present, it takes the form of ‘ufu’ as in (25b) when the verb is an achievement. On the other hand, state predicates do not allow a present-tense interpretation corresponding to (25b) because of the aspectual-temporal interpretation retained in conjectural ufu (see the discussion on the interpretation of (32) in Section 5.2). What is crucial are examples (25c) and (26b), which obtain their evidence in the past but are incompatible with the past tense marker -daɨ.

(25c) is about an event that occurred yesterday, i.e., that someone got lost, and the sentence says that the street lamp might have gone out beforehand. Also, (26b) is about a past event, which is recalled by the speaker, and the speaker conjectures the existence of graves at that time. In such a case, we may expect the verb to take a past tense morpheme -daɨ, but this is in fact ungrammatical. This means that, though the reference time and, as a result, the time of the event are situated in the past semantically, the modal auxiliary is fixed as a “non-past” form.

Given the fact that conjectural ufu is frozen as a “non-past” form, let’s now look at the examples in (27), which have activity verbs, as with (22), but differ from it in the reference time, i.e., (27) is the past tense version of (22).

Now that we set the reference time to the past in (27), the past tense morpheme -daɨ is expected to be suffixed to aspectual ufu, which is born out in (27a) with the present form unacceptable. This predicts that if the morpheme ufu is only an aspectual even in the context where the relevant proposition is conjectured, e.g., (27b), then the past tense morpheme -daɨ must also be attached. In this context, however, the sentence is accepted irrespective of the attachment of the past tense marker. This shows that ufu can appear not only with its aspectual use but also with a conjectural use, frozen as the “non-past” form in the latter. This further indicates that conjectural ufu can combine with verb types other than achievement and state verbs (in this case, activity verbs).

The above consideration leads us to assume that if ufu is used in a non-past tense form even in a context that refers to the past, it is not aspectual but conjectural ufu. This further enables us to extract examples of conjectural ufu from spontaneous discourses. In spontaneous discourse, it is not easy to detect the use of ufu when it is used with accomplishment and activity verbs as indicated in (22), but when it does not inflect for past tense in a past context, we can identify it as conjectural. Some of those examples are given in (28) below.

In (28a), the speaker tells a story occurring after WW2, which refers to the past, as indicated by the past tense form at end of this example, i.e., ‘azzi=djaa.’¹³ Since the American soldier is assumed to have talked before the old man did, the tense of the first sentence should be the past. Nevertheless, the non-past tense form ‘ufu’ is used. (28b) also refers to an episode after WW2. The utterance of A, which is spoken in Japanese, indicates the tense to be the past as shown by the past tense form of to ‘die,’ i.e., ‘nakunat-ta.’ In spite of the fact that the boy had removed the screw before he died, the non-past tense form ‘ufu’ is used in B’s utterance. The same pattern is observed in (28c). A boy had been playing at ‘Miyaguniyaa’ before he died. The time he died is expressed by the past tense form in the first sentence, i.e., ‘sɨni,’ but the sentence describing his playing at ‘Miyaguniyaa,’ which occurred prior to his death, is marked by ufu and is lacking the past tense morpheme -daɨ. This reasoning leads us to conclude that the relevant occurrences of ufu are conjectural uses.¹⁴

Based on the considerations so far, let us lastly refer to the results gained from the survey of ufu in spontaneous discourse. My recordings consist of four independent dialogues, annotated with ELAN (2020), and the sum of the utterance durations of the speakers in (2) amounts to 110.20 minutes. I collected 34 instances of ufu that are pronounced clearly. 6 out of 34 examples are used in subordinate clauses, in which tense marking is not obligatorily synchronized with that of the matrix clause. The remaining 28 examples are broken down as in Table 3.

The instances of ufu used with achievement and state verbs are classified as conjectural ufu as we have discussed. The remaining examples (accomplishments and activities) are judged by the lack of the past tense morpheme (despite its clear past time reference) as we saw in the examples of (28).

4.3 Summary

As an aspectual auxiliary, ufu takes accomplishment and activity verbs and conveys the resultative interpretation. It cannot be used with achievement and state verbs because another auxiliary uɨ gives the resultative interpretation of the former,¹⁵ and state verbs do not lexically encode a change of state and thus no “result.” The conjectural use of ufu is thus more explicitly manifest in the latter verb types: ufu used with achievement and state verbs must involve conjectural meanings. Despite its superficial appearance, however, the conjectural use of ufu is concluded to be prevalent across verb types if we pay attention to its incompatibility with tense inflection.

5.1 Development of evidential/modal meaning

It is a widely acknowledged fact that resultative aspect markers grammaticalize to inferential evidential markers across languages (Comrie 1976: §5.2.2.1, Bybee et al. 1994 §3.15, Aikhenvald 2004: §9.1.3, etc). The development of the conjectural use of ufu relates to this grammaticalization. Bybee et al. (1994) make the following remark on the similarity between resultatives and inferential evidentials.

(T)he resultative indicates that a state exists due to a past action. This meaning is very close to the evidential meaning of an inference from results, which indicates that a past action is known or inferred on the basis of a current state. (p. 96)

To infer the past action from the current state, there needs to be a reasonable connection from the former to the latter. For example, if you infer the proposition that grandma fried doughnuts from the fact that there are doughnuts, you presuppose that if grandma fries doughnuts there are fried doughnuts.¹⁶

In most instances of conjectural ufu, a causal relation of this kind can be reconstructed as confirmed by the examples from (25) through (28), which have explicit contexts under which the sentences are uttered. There are, however, examples that are not characterized by such causal relations. Let us look at the examples taken from spontaneous discourse, i.e., (28).

We can find causal connections behind the use of ufu in (28b): the fact that the boy removed the grenade’s screw caused it to explode. The speaker infers the former from the fact that the boy died.

On the other hand, it is not straightforward to retrieve a causal relation in (28a) and (28c). The speaker surmises the background of the incident where the boy was shot under the tree in (28c), particularly, the reason why he was where an air raid took place. But it seems rather odd to consider his playing at Miyaguniyaa then as the cause of his being shot, i.e., it is far from possible to infer the former from the latter by the “causality” between the two. Rather, the speaker concludes the former from the habit of children playing at their parental home. This is a typical case of “assumed” evidentiality in the sense of Aikhenvald (2004), who considers it as mainly involving general knowledge or common sense to infer the proposition as the following quote shows.

The difference between the ‘assumed’ evidential … and the ‘inferred’ … lies in access to visual evidence of something happening and to degree of ‘reasoning’ involved. The less obvious the evidence and the more the speaker has to rely on reasoning based on knowledge or on common sense, the more chance there is that the assumed evidential will be used. (pp. 2–3)

The use of ufu in (28a) also relies on the general knowledge about conversational exchange to a great extent. Based on this knowledge, the speaker assumes that the American soldier said something before the old man spoke. It is also easily imaginable what the American soldier said from the situation where he was chasing a woman. Therefore, the proposition to which ufu attaches is not recovered by the evidence at hand, but by general knowledge about conversation and the relevant situation, which allows us to regard ufu in (28a) as an assumed evidential use rather than an inferential one.

The same point can be made with state verbs, which restrict the use of ufu to the conjectural. A causal relation can be retrieved from the examples in (29). In (29a), since the relevant woman was inherently ill-natured, she did terrible things to her daughter: the speaker infers her nature from what she did. In (29b), speaker B concludes the nature of the pig as foolish from the fact that it cannot wait for food: obviously it is the pig’s nature that triggers its action.

On the other hand, it is not obvious what causality or general knowledge is assumed behind the use of ufu in (30). Speaker B of (30a) states that the dialect card had been used before the time when speaker A thinks it to have been. The conjectural meaning contributed by ufu plus a question marker targets the exact year when this card started to be used, but this is neither an inference based on evidence nor general knowledge but merely a search based on her memory. The speaker in (30b) also tries to recall the year, and, if ufu is used as the result of this recalling, then it involves neither causally established reasoning nor general knowledge.

The examples in (30) show that the semantic change of ufu proceeds beyond inferential and assumed evidentials defined above. According to Aikhenvald (2004: §4.2, 5.1), inferential evidentials change to assumed evidentials with the loss of obvious evidence and further extend to epistemic modals gaining the meaning of probability and doubt. Since the speakers in (30) do not commit to the truth of the proposition, conjectural ufu is shown to have arrived in the domain of epistemic modality in the sense discussed here.¹⁷

5.2 Development of temporal meanings

As we saw in Section 4.2, conjectural ufu has a past-orientedness and is incompatible with a past tense marker despite its interpretation. This can be regarded as a “de-categorialization” from verbs, one of the typical processes observed in grammaticalization (Hopper & Traugott 2003: §5.3). This subsection closely looks at the temporal properties of the conjectural use, comparing it with that of the aspectual use.

In the aspectual use of ufu, the resultative state follows the event which triggers that state in time. This temporal relation is expected to hold even in the conjectural use if the inference is based on the result of an event (cf. the quote from Bybee et al. (1994) in the previous section). For example, a boy had removed the screw before the grenade exploded in (28b), where ufu marks the preceding event.

It is, however, logically possible for the event causing the result not to be completed before the utterance time. This is what is observed in inferential evidential forms in other languages. The following is an example of an inferential evidential in Bulgarian developed from “present perfect” of this language.

Since the evidential form in (31) is compatible with adverbials like right now, it loses the original aspectual meaning characterized as “perfect,” as argued by Izvorski (1997), and, as a consequence, the letter-writing event is understood to be taking place at the time of utterance. On the other hand, the interpretation in question is precisely the one that is excluded from the interpretation of conjectural ufu. The following example, whose context resembles (26b), is a case in point.

(32) differs from (26b) in the time when the road is trimmed: the speaker sees the evidence at present in (32). This would enable the speaker to conjecture the background presently held as with (31). However, the only interpretation allowed with sentence (32) is, according to the speakers, the one where the graves existed in the past and no longer exist at the utterance time. This fact convincingly shows that conjectural ufu preserves the aspectual-temporal relation of resultatives, which situates the causing event precisely before its result.¹⁸

While ufu is more conservative than the Bulgarian “perfect of evidential” in (31) with respect to the above aspectual property, the former develops a tense-like meaning that is not attested in the latter. Izvorski (1997) connects the evidential in (31) to its source, i.e., present perfect, arguing that the evidence must hold at the utterance time contributed by the “present” tense. Karimata ufu, on the other hand, loses this property, freely recruiting past events as evidence for inference, as witnessed in (25c) and (26b), which is repeated here as (33).

Considering the time of evidence as the reference time of the sentence (Izvorski 1997: §4), we can conclude that ufu in (33) has a tense-like meaning that shifts the reference time to the past and thus prevents the attachment of the past tense marker -daɨ.

Comparing the case of ufu with that of the Bulgarian evidential, thus, reveals two different ways to change the temporal meaning in the transition from the resultative/perfect to evidentials. It is also in this transition that one can naturally account for the acquisition of the past tense meaning by ufu. According to Izvorski (1997), in using evidentials, the speaker can access the possible worlds where indirect evidence, e.g., a road into a grove being trimmed in (33), is true. Acquiring a past tense orientation means, according to this view, to come to establish the speaker’s epistemically accessible worlds based on the available evidence in the past. This process is easily realized because indirect evidence available in the past is also available in the present. It is, on the other hand, not determined what evidence is available in the future. So, the shift to the past is more likely to occur than the shift to the future in the transition to evidentials. This is indirectly supported by the fact that more distinctions of evidentials are made in the past, or in non-future, than in the future typologically (See Aikhenvald 2004: §8.4).

It should be lastly noted that the emergence of the past tense meaning in ufu is epiphenomenal as the ongoing discussion indicates. While the past tense meaning can be suitably placed in the rise of evidential meanings, it is not straightforward to induce the evidential/modal meanings in ufu if its primary change is to come to designate the past. This is evident in the typological fact that the change from the resultative/perfect is more readily attested than that from the simple past in the development of inferential/assumed evidentials.¹⁹

5.3 Discrepancy of syntax and semantics

Given the development of modal and temporal meanings discussed above and the structure of the verb complex of this dialect, repeated here as (34), it is expected for conjectural ufu to appear not in the Aspect slot but in the Tense or Modality slot.

This is not, however, true of ufu in the Karimata dialect: ufu stays in the Aspect slot even after acquiring those non-aspectual meanings.

Before proceeding to show this syntactic property, it is in order here to mention the evidential meanings discussed in Section 5.1 again, since they have no place in (34). Some authors consider inferential/assumed evidentials basically as an epistemic modal (Izvorski 1997; McCready & Ogata 2007; Matthewson et al. 2007 etc.), and others don’t (de Haan 1999; Aikhenvald 2004; Davis & Hara 2014, etc.). If they are a kind of epistemic modality, it occupies the position outside of Tense and this is exactly where the reported evidential in Karimata appears as shown in (35).

However, it is frequently observed that reported evidentials behave differently from inferential/assumed evidentials. Moreover, heterogeneous morphosyntactic characteristics of evidentials in one single language have been reported cross-linguistically (Aikhenvald 2004: §3.3–4), which makes it difficult to predict the exact syntactic position of evidentials in Karimata’s verb complex. Therefore, I will avoid specifying the syntactic position through language-internal evidence and demonstrate the underdevelopment of syntax by comparing Karimata with another Ryukyuan language: Okinawan.

The development from a resultative to an inferential evidential marker is more abundantly observed in North Ryukyuan than in South Ryukyuan. Arakaki (2010: 82) gives the following example to show that Shuri Okinawan (see Figure 1) has three homonyms originating in the resultative marker tee: the number subscriptions are added by the current author.

The first -tee₁, with the consonant /t/ deleted due to the preceding consonant, is the resultative. The second -tee₂, glossed here as ‘m,’ is analyzed as a ‘mood’ marker conveying ‘irrealis’ meaning by Arakaki (2010), but has been viewed as an inferential evidential by other researchers (Shinzato 1991; Kudo et al. 2007; Davis 2017). The third tee₃ is the one claimed by Arakaki (2010) to be the inferential (inf) evidential marker in this dialect. Irrespective of the analysis adopted, sentence (36) is enough to show the development from a resultative morpheme to other grammatical categories, each occupying different syntactic positions. We can consider the syntactic slot occupied by -tee₁ as Aspect in a structure analogous to (34) because of its function. In spite of different analyses above, researchers agree that -tee₂ has a past tense function occurring in the same position as the simple past tense morpheme -ta. Since tee₃ appears outside the Tense slot, it resides in the Modality slot if we apply the structure (34). Those syntactic manifestations are considered as a natural consequence of the meaning extension of tee.

Karimata exhibits a striking difference from the pattern in Okinawan. Since the syntactic position of the relevant morpheme would proceed in the order Aspect > Tense > Modality, I focus on the difference between the Aspect and Tense slot here. If it occupies the position outside Aspect, it should be able to coexist with aspectual morphemes in a single verb complex, like Okinawan tee. However, ufu never appears with aspectual morphemes such as uɨ and ufu, as the following ungrammatical example shows:

This is most naturally accounted for by the assumption that conjectural ufu stays in the Aspect slot.

Another piece of evidence for ufu remaining in the Aspect slot can be given by its relative order with honorifics. According to Davis (2017), honorifics in Okinawan sit in the boundary between Aspect and Tense and differentiates the inferential -tee₂ from the aspectual -tee₁, with only the former appearing after honorific morphemes. On the other hand, conjectural ufu, as well as aspectual, must precede the honorific morpheme. Note that the following has a state predicate so that ufu is conjectural.

The first and the second options in (38) convey almost the same meaning with the addition of a conjectural marker =padzɨ. The first pattern shows that a past tense morpheme, -daɨ, appears after the honorific -ama. The second and third contrast shows that conjectural ufu occupies the position before the honorifics, the position occupied not by the tense morpheme but by the aspect morpheme in the structure (34). The second option is less preferred to the first but still acceptable according to the consultant.

An interesting mismatch between syntax and semantics is manifest in the use of ufu with negation. The structure (34) predicts ufu to appear before negation if it occupies the Aspect slot and to appear after negation if it occupies the Tense slot. The result illustrated in (39) shows that the former hypothesis is correct (adaari includes the infinitive form of aɨ, i.e., ‘ari’).

Though occupying the position internal to negation, ufu is interpreted externally to it: (39) does not mean that the speaker is not inferring p but means that the speaker infers not p. Thus, the scope of ufu is wider than the negation semantically, but not in syntax.²⁰

The above syntactic evidence indicates that conjectural ufu does not leave the Aspect slot despite its semantics. Due to this syntactic property, it might be argued that the conjectural use of ufu does not constitute a distinct morpheme from aspectual ufu, but is just a “strategy” or “overtone” accompanying it (cf. Aikhenvald 2004: §4.2). However, as argued in previous sections, conjectural ufu differs from the aspectual one in its 1) property of unrestricted selection of preceding verbs, 2) deficiency of inflection for tense morphemes (Section 4.2), and 3) different temporal interpretation from resultatives (Section 5.2). These semantic properties, particularly 1) and 3), cannot be viewed as imposed by an “overtone” on the semantics of resultative aspects. Therefore, it is natural to consider conjectural ufu as a distinct morpheme diverging from aspectual ufu, but one that has not made any progress in its syntactic properties.

5.4 Summary and discussion

In the same way as other cross-linguistically attested cases, the resultative marker ufu has developed to carry an inferential evidential meaning. But, beyond this, it also gains assumed evidential and epistemic modal meanings, which leads us to use the label “conjectural” in this paper. Not only in the modal but also in the temporal domain does the meaning of ufu proceed: conjectural ufu can refer to evidence in the past as with the reference time of the past tense morpheme. Despite its progress in semantics, conjectural ufu still sits in the Aspect slot, as shown by the comparison with Okinawan -tee.²¹

Given this syntax/semantics discrepancy in the grammaticalization, why does semantics change before syntax in the verbal domain? Among many factors related to the difference between syntax and semantics, I suspect the fine-grained nature of meaning to be most relevant to answering this question: arguably, that the syntactic structure is highly restricted compared to the gradience of meaning in this domain. Let me conceptualize this as a “one-to-many” relation between syntax and semantics in the sense that many semantic features could share one syntactic node of the verb complex, e.g., past, present, and future in the Tense slot. If a morpheme changes its meaning, this relation enables us to predict its syntactic node by mapping the former to a unique node, which I would like to call semantic predictability of syntax. However, it is not the other way around from syntax to semantics because a single syntactic node corresponds to multiple semantic candidates. Thus, the emergent meaning cannot be predicted only by its syntactic position, i.e., syntactic non-predictability of semantics. This (non)predictability is significant, particularly in language change. Imagine a situation where some semantic change occurs in the verb complex and gets stabilized in a language community. The speakers in the next generation can predict its syntax, making it possible for the change to go smoothly. But even if a community were to accept a syntactic change without any semantic one, what would come next as its verbal meaning could not be predicted and might vary from speaker to speaker, which would prevent a fluent transition of language change. This is the reason, I propose, why semantic change can precede syntactic change in this domain but not vice versa.²²

The theoretical and empirical priority of semantics to syntax in the order of language change casts doubt on a purely syntactic explanation for grammaticalization (see also Vincent & Börjars 2010: §4). Roberts & Roussou (2003), for example, propose “structural simplification” as the reason for why grammaticalization is so common; van Gelderen (2004) considers the syntactic operation “Late Merge” as an explanation for “up the tree” grammaticalization. But at the stage where only semantic change occurs as seen in our case study, no syntactic factor is responsible for the change. This implies that the change could proceed in both directions, i.e., to the semantic category higher or lower in the structure (34). But given that changes such as an epistemic modal to a tense category, a tense to an aspectual category, etc. are not attested even semantically, let alone syntactically, grammaticalization is highly constrained by semantics. Thus, though beyond the scope of this paper, the account for the pervasive nature of grammaticalization should be sought not in syntax but in semantics (cf. Haspelmath 1999).