Tense-evidentiality: a typological study

Tense-evidentiality: a typological study

Eline Visser

June 29, 2014

Abstract

To gain more insight into evidentiality, the grammatical marking of information source, it is worth to study the category in its relation to other grammatical categories. This study explores tense-evidentiality, the simultaneous marking of tense and evidentiality, in 36 languages from four macro-areas. The study reveals several interactions between tense values and evidential values. Firsthand future evidentials seem to exist (as opposed to what has been claimed); and visual and non-visual evidentials occur equally with present and past, whereas mental activity evidentials and reportative evidentials are much more limited to the past. These ndings conform to the logic of tense: every type of evidentiality can be gathered in the past, whereas some are less suited for present or future. As a result, we nd exceptionless implicational hierarchies of the form future ⊂ present ⊂ past for several evidential values. In general, it is argued that evidentiality is a dynamic grammatical category, among other reasons because it easily conforms to other grammatical categories such as tense. This dynamic nature of evidentiality should be taken into account when analysing the category in the future.

1 Introduction

Evidentiality, the grammatical marking of information source (Aikhenvald, 2004, p.1), has received much attention in the last few decades. Although we have a general idea of how languages mark information source, within evidentiality there is much more to explore, especially with respect to the interaction of evidentiality with other grammatical categories. One of those is tense, the gram-matical expression of location in time (Comrie, 1985, p.9). Tense and evidentiality are conceptually closely related, because the evidence for an utterance is always gathered at some point in time, and often this point in time is vital for the hearer to understand how the speaker got his/her information. Does he/she see it happening now? Does he/she infer it based on the result of a previous action? A fair number of languages mark tense and evidentiality on the same morpheme, conrming the close relationship between the two categories. For an illustration of so-called tense-evidentiality, consider the minimal pair in example 1 from Tsez below.

(1) a. obiy- a

father-ergmadina-rMadina-latk'icustrawberry y-is-si.ii-buy-firsth.pst

`Father bought strawberries for Madina.' (the speaker saw this) b. obiy- a

father-ergmadina-rMadina-latk'icustrawberry y-is-no.ii-buy-nonf.pst

`Father bought strawberries for Madina.' (the speaker did not see this) (Comrie & Polinsky, 2007, p.338)

The dierence between 1a and 1b is evidential: in 1a, the speaker saw or heard the event happening, whereas in 1b, the speaker learned about the event through some other means (perhaps someone told him/her, or he/she saw strawberries on Madina's kitchen table). What characterises the markers -si and -no is that they do not only express evidentiality, but also (past) tense. In Tsez, it is impossible to place an event in the past without referring to the source of information. Example 1 shows all the tense-evidentiality options available in Tsez. In other words, the Tsez

tense-evidentiality system consists of two members, rsthand past and non-rsthand past, which are in paradigmatic opposition to each other.

Although languages with tense-evidentiality, not seldom with much more elaborate systems than that of Tsez, have been mentioned in the literature on evidentiality and in individual language descriptions, only one typological study has been carried out (Dall'Aglio Hattnher, 2013, comparing two evidential values in 11 native Brazilian languages). San Roque and Loughnane (2012) mention several Papuan Highland languages with tense-evidentials, but do not analyse them in-depth. This typological study aims at nding commonalities of, dierences between and constraints on tense-evidentiality systems. A sample of 36 languages with tense-evidentials is used to make an initial step towards a typology of the form and function of tense-evidentials.

The following questions will be answered:

Q1 Which tense values are combined with evidentiality?

Q2 Is there a correlation between certain tense values and certain evidential values? Q3 Can we identify implicational hierarchies for tense-evidentiality?

Q4 What is the relationship between tense-evidentials and other evidentials within the same language?

Based upon earlier studies, the following hypotheses with regard to these questions are formu-lated.

H1 Past tense is most commonly fused with evidentiality, followed by present and future. H2a Firsthand evidentials (including participatory, visual and non-visual) are not fused with future

or a past tense referring to before the speaker's lifetime.

H2b With present inferentials, the tense refers to the moment of inferring (= moment of speaking), whereas the result on which the inference is based is prior to that.

H3 The expected implicational hierarchy within tense-evidentiality is firsthand ⊂ inferential ⊂ assumption.

H4 If tense-evidentials and other evidentials have semantic overlap within the same language, this might point at the former being in need of reanalysis.

We will see that only the rst hypothesis is fully conrmed. The hypotheses relating to the correlation between tense values and evidentiality values are modied. The implication hierarchy sketched in H3 does not hold for tense-evidentiality, however, we nd a strong tendency for the order past > present (> future) if we sort the data per evidential value. H4 is conrmed for some cases, but not for others. The answers to the questions posed do not only give us a cross-linguistic insight into tense-evidentiality, but can also tell us more about the nature of evidentiality.

The rest of the paper is structured as follows. Section 2 introduces the grammatical category evidentiality. In section 3 the relation between tense and evidentiality is investigated. Section 4 moves on with the topic under study here: tense-evidentiality. It discusses denitions and criteria, examples, and the hypotheses summarized above. Section 5 treats the method and the sample for this typological study. In section 6 the results are presented and discussed, divided in subsections per question/hypothesis. Finally, section 7 concludes the paper and gives an outlook on further research.

2 Evidentiality

This section introduces evidentiality as a grammatical category, and discusses some examples of evidentiality from dierent languages to illustrate the meanings that evidential markers can ex-press. It also discusses the basic concepts having to do with evidentiality that are needed for the investigation of tense-evidentiality.

We will use Aikhenvald's (2004) denition of evidentiality here, which states that it is the grammatical marking of information source. About a quarter of the world's languages are claimed

to do this, that is, to exhibit evidentiality as a grammatical category (Dixon, 2010).1 _{For an}

illustration of evidentiality, consider the minimal pair from Shipibo-Konibo in example 2 below. (2) a. jawen

poss.3jema-ravillage.abs-firsthanilargeiki.cop `Her village is large.'

b. jawen

poss.3jema-ronkivillage.abs-repanilargeiki.cop `Her village is large.'

(Valenzuela, 2003, p.34)

In English, both sentences translate as `Her village is large'. The dierence in meaning, however, is that in 2a the speaker has personally seen the village, whereas in 2b he or she has learned that the village is large by other means. In the latter case, it could for example be that the speaker was told by another person about the village, or that he or she inferred it is large because there is a train station at the village. Evidentials are organized in an evidentiality system with contrasting members, where the markers often form one morphological paradigm. In English, references to information source cannot be made grammatically, only lexically, as in example 3.

(3) a. `I saw that her village is large.' b. `They told me her village is large.'

Lexical evidentiality is usually not included when evidentiality is discussed, and plays no role in the current study. Grammatical evidentiality deals with closed class systems of evidentials.

Languages exhibiting evidentiality, such as Shipibo-Konibo, are found worldwide, although only few cases of evidentiality in Africa and Australia are known. Evidentiality systems dier in the number of evidential meanings that are distinguished. A frequent opposition in the evidential-ity paradigm is between rsthand information and non-rsthand or reported information, where some languages only mark the latter category. More elaborate systems are found as well, where a distinction is made between several types of non-rsthand information, such as hearsay, assump-tion or inferring from the situaassump-tion. Firsthand informaassump-tion can be further divided in visual and non-visual information, the latter including, for example, smelling and hearing. Aikhenvald (2004) identies six recurrent meanings of evidentials worldwide: visual, non-visual (sensory), inferential, assumption, hearsay, and quotative (overt reference to quoted source). Visual and non-visual can be grouped as rsthand, and the other values as non-rsthand. Hearsay and quotative group together as reportative in many languages.

Aikhenvald's (2004, p.1) classication will be used here, because it is the most theory-neutral and the most generally used. Note, however, that other classications exist (e.g. Plungian, 2010; San Roque & Loughnane, 2012; Willett, 1988, and see Hengeveld and Dall'Aglio Hattnher (n.d.) for yet another approach and an overview of earlier classications). To be able to discuss some of the salient features of some languages in the sample, the category of participatory evidentiality is taken into account here as well (discussed in Plungian (2010); San Roque and Loughnane (2012)). Participatory evidentials mark that the speaker has performed the event he/she reports about, and is therefore even stronger evidence than visual. In our sample, the languages Foe and Oksapmin (spoken in the Papuan Highlands) exhibit this evidential meaning.

Table 1 shows the evidential meanings that are found in the tense-evidentials analysed in the current study. It shows that rsthand and non-rsthand are meta-categories, comprising partici-patory, visual and non-visual and inferential, assumption and reportative evidentials respectively. Some languages only have an opposition between these meta-categories, others use some or all of the

1_{Sometimes evidentiality is described as epistemic modality (e.g. Chafe & Nichols, 1986). In fact, it is not always} easy to draw the line between evidentiality and epistemic modality. Evidentiality is claimed to be a grammatical category on its own (e.g. Aikhenvald, 2004, p.4), because in many languages we nd markers that have source of information as their primary meaning. These markers express the way in which information was acquired, whereby degree of certainty is not a necessary component of the meaning. In several languages, such as Tariana, markers of modality and evidentials can co-occur (Aikhenvald, 2003a, p.154). Consider De Haan (1999) for a discussion on the boundaries between evidentiality and epistemic modality and e.g. Peterson (2010) for semantic tests to distinguish between the two.

subcategories. A mix of a meta-category and subcategories belonging to the other meta-category (say, rsthand and inferential, as for Sabanê in the current sample) is also possible. However, no language will have e.g. both a rsthand and a visual evidential because one is a meta-category and the other a subcategory within this meta-category. Because we nd these mixes of meta-categories and subcategories in the tense-evidentiality systems of the current sample, all evidential values (either meta-category or subcategory) are treated equally.

Table 1: Evidential meanings relevant for the current study.

rsthand participatory visual non-visual non-rsthand inferential assumption reportative

No language has been found thus far that has all six recurrent evidentiality meanings. A number of languages have ve terms, many of them in the Vaupés area in Amazonia (Epps, 2005; Aikhenvald, 2004, p.60). Hup is one of these languages, and exhibits visual (default, unmarked), non-visual, inferential, assumption and reportative evidentials. The system is exemplied in example 4 below.

(4) a. manga

Margaritah1d-an3pl-objt@w-n1h=ø=kah.yell.at-neg=vis=disj

`Margarita didn't yell at them, actually.' (speaker was there) (Epps, 2005, p.623)

b. k'@h

sweetnaw=hO!good-nonvis

`It's nice and sweet!' (tasting something) (Epps, 2005, p.626)

c. t1h=doP

3sg=childs'Om-s1w1y=sud.bathe-already=inf

`The child already took a bath.' (we see he's wet) (Epps, 2005, p.631)

d. yup

thathOtd'ah=mahother.side=reph1d3pl ye-ni-ip=b'ay-ah.enter-ass-sub=again-decl

`There on the other side of it (they say), they got in again.' (assumption based on fact that sh has disappeared from house)

(Epps, 2005, p.638) e. t1h

3sgham-teg=mah.go-fut=rep

`He'll go.' (he or another said so) (Epps, 2005, p.634)

Because we will try to establish implicational hierarchies for tense-evidentiality, it is worthwile to look at implicational hierarchies for evidentiality. Based on 34 native Brazilian languages, Hengeveld and Dall'Aglio Hattnher (n.d.) propose a hierarchy for evidential meanings present in a language2_:

(5) firsthand ⊂ inferential ⊂ assumption

This means for example that if a language has an inferential evidential, it will also have a rsthand evidential, but not necessarily an assumption evidential.

2_{Earlier proposed hierarchies (Faller, 2002; Willett, 1988) include more evidential values, but are contradicted by} the sample in Hengeveld and Dall'Aglio Hattnher (to appear).

3 The relation between tense and evidentiality

This section deals with the relation between tense and evidentiality. Tense is taken to be the gram-matical expression of location in time (Comrie, 1985, p.9). We will see that there exist dependency relations between tense and evidentiality. These show that the two categories are semantically closely related.

Tense and evidentiality are conceptually closely related, because the evidence for an utterance is necessarily gathered at some point in time. As the past covers a much bigger time span than the present, and the future is unknown, we would expect most evidential distinctions in the past tense. Moreover, there is often some time needed between the occurrence of the event upon which evidence is based and its detection (the gathering of the evidence). As it is the occurrence of the event that is usually reported, past tense is more likely to be used (see also the discussion of `double tense' in 6.3.3). In line with this, Bybee, Perkins, and Pagliuca (1994) have shown that, historically, perfects which develop resultative uses may take on non-rsthand evidential functions.

The grammatical system of individual languages also shows that tense and evidentiality are often interrelated. Aikhenvald and Dixon (1998) claim that although in some languages all evidential values can be used in all tenses, this is rather rare. In some languages, the choices available in the evidentiality system are dependent on which tense is used. An example is Jarawara, a language with an evidentiality system that only applies in the past tenses (see also example 7 and table 2 below). The other way around the choice made in the evidentiality system can inuence the choices available in the tense/aspect system. Bulgarian, with a two-term evidentiality system distinguishing between reported information and everything else, has nine tense/aspect choices available when using the reportative. But as present and imperfect have only one non-reported counterpart, and this goes the same for the other tense/aspect values, only ve tense/aspect choices are left when using non-reported evidentiality. Macuna normally has two degrees in the past tense, but when an evidential is used (inferential, assumption or reported) this is limited to one past tense.

It is thus not unexpected that a number of languages, among which Jarawara, Bulgarian and Macuna, mark tense and evidentiality on the same morpheme. The section below elaborates on these so-called tense-evidentials.

4 Tense-evidentiality

4.1 Denition and criteria

Tense-evidentials are dened here as any grammatical marker or set of markers that express tense and evidentiality simultaneously. In the majority of the languages in the sample for the current study, these markers are part of a tense paradigm in which all items express tense and all or some of them express evidentiality as well. Only in the Cree dialects (Cree, Montagnais and Naskapi) the tense-evidentials seem not to be part of a tense paradigm, which distinguishes between past and non-past. They are described as evidentials that express past vs. present tense as well Blain and Déchaine (2007); James, Clarke, and MacKenzie (2001), however, they seem to be in paradigmatic opposition to the tenses (James et al., 2001, James p.c.). Whether we can regard them as part of the tense paradigm remains unclear. The Cree dialects are part of the current study because tense and evidentiality are expressed simultaneously on the same morpheme, and hence they have tense-evidentiality markers. All tense-evidentials in the individual languages in the sample of the current study form a so-called tense-evidentiality system. This refers to paradigms of tense-evidentials, the members of which are in contrast to each other.

To be able to dene tense-evidentiality, a set of criteria was developed for the current study. The rst and main criterion for a tense-evidential is that it expresses both tense and evidentiality, as illustrated in example 1. Many sources mention this explicitly. For Jarawara, for example, Dixon and Vogel (2004, p.589) explicitly claim: "All evidential markers are portmanteau verbal inectional suxes that simultaneously mark evidentiality and tense." The second criterion states that tense and evidentiality must be the primary meaning of the marker. This means that neither tense nor evidentiality can be analysed as an extension or alternative reading of the marker. An example of such an extension comes from Estonian, where the past tense of the auxiliary `must' can sometimes get a non-rsthand reading. However, this is dependent on the context and the

primary reading of the marker is a past tense auxiliary (Erelt, Metslang, & Pajusalu, 2006, p.133). It can therefore not be counted as a tense-evidential. The third criterion is that if the marker is separable into dierent parts, it can only count as a tense-evidential if these parts obligatorily occur together. For some languages in the sample, the tense-evidential can be analysed as being made up of two morphemes, one expressing tense and the other evidentiality. Only if these morphemes cannot occur independently, they can be marked as tense-evidentials. This can be illustrated with an example from Tariana, where tense-evidentials can be segmented into two morphemes. The rst one expresses the evidential value and the second expresses tense. Because tense cannot be expressed without evidentiality and vice versa in Tariana, their markers are usually presented as a portmanteau morpheme, as in example 6 below.

(6) Ceci

CecíliatSino-nukudog-top.non.sbjdu-kwisa-pidaka.3sg.f-scold-rec.pst.rep

`Cecília scolded the dog.' (I have learned it from someone else.) (Aikhenvald, 2003a, p.135)

So even though we know that in -pidaka it is -pida expressing reported evidence and -ka past tense, they count as a tense-evidential because they invariably occur together Aikhenvald (2003b). In other words: it is not possible to express either tense or evidentiality without also expressing the other in Tariana, and hence the markers are counted as tense-evidentials.

The next section illustrates some of the forms these tense-evidentials or tense-evidentiality sys-tems can take.

4.2 Tense-evidentiality as separate morpho-syntactic system

This section illustrates the above criteria, as well as raising some issues in identifying tense-evidentials. The examples show how tense-evidentiality behaves as a separate morpho-syntactic sys-tem. It is shown that tense-evidentials have dierent types of markers and that tense-evidentiality systems show dierent interactions between tense and evidentiality within individual languages. After showing a typical evidentiality system, this section discusses some subtypes of tense-evidentials and systems. Some languages have neutral tenses, which are unmarked for evidentiality and stand in contrast with tense-evidentials. Others have what can be called a scattered system: some evidential values are only available in some of the tenses. Finally, examples of tense-evidentials that are expressed on more than one morpheme are discussed.

The following sentence from Jarawara, an Arauan language spoken in Amazonia, contains two typical tense-evidentials.

(7) katoso

cartridges(f)ka-foja-niappl-be.inside-rec.pst.nonf.f(...)(...) o-wa-kiti1sg.sbj-appl-take.out o-na-hara

1sg-list-rec.pst.firsh.fo-ke.1sg-dec.f

`The cartridges were inside (a bag) (...) I took them out.' (Dixon & Vogel, 2004, p.204)

In this example, the sux -ni expresses both recent past and the fact that the speaker did not see the cartridges in the bag, but thought they were (based on some non-rsthand evidence). The sux -hara, on the other hand, marks recent past and a rsthand experienced event: the speaker saw the cartridges when she took them out of the bag. These suxes also happen to mark gender, which is a topic that will not be investigated here. Table 2 shows the entire feminine tense-evidentiality paradigm for Jarawara (Dixon & Vogel, 2004, p.197).

Table 2: The tense-evidentiality paradigm of Jarawara (feminine forms only).

rsthand non-rsthand recent past -(ha)ra -(ha)ni remote past -(ha)ro -(he)te very remote past -(ha)maro -(he)mete

In Jarawara, there is no way of expressing past tense without also expressing the source of information. Present and future are unmarked for evidentiality.

In Sabanê, on the other hand, tenses unmarked for evidentiality exist next to tense-evidentials (de Araujo, 2004). The so-called `neutral' tenses are used when evidence is lacking. In table 3, the neutral and evidential forms are listed.

Table 3: The neutral tenses and tense-evidentials of Sabanê.

neutral rsthand inferential future -tapanal -telon

present -al -dana

past -ntal -datinan -tika

Hence, there are three possible answers to the question `Did she cry?', depending on the (kind of) evidence. In the answer in 8a, the speaker has seen or heard the subject crying, whereas in 8b, no evidence is available for the statement. An example with the past inferential is not given by de Araujo (2004), but could for example be used when the speaker has seen the person's swollen eyes.

(8) a. nan-i-datinan. cry-vs-pst.firsth `S/he cried.' b. nan-i-ntal-i. cry-vs-pst.neut-asrt `S/he cried.' (de Araujo, 2004, p.140)

As can be seen in table 3, not every language lls up all the possible cells in a tense-evidentiality paradigm. Sabanê does not have a grammaticalised future or present inferential. In Macuna, the system is even more `scattered': the present has a rsthand and a reported evidential, and the past a reported, an inferential and an assumption evidential. In addition, there is a non-past assumption evidential.

Estonian is an example of a language where a combination of more than one marker expresses tense-evidentiality. Past reported evidence is expressed by a dedicated evidential marker -vat, which is axed to the rst verbal form of the predicate, the auxiliary `be' in example 9. Combined with a past participle, this indicates general past tense.3

(9) ma

I ole-vatbe-rep kirjuta-nud.write-pst.part `I am said to have written.' (Erelt et al., 2006, p.128)

The past reportative can also be shortened to -nuvat (from the past participle -nud and the reportative -vat), as in example 10.

(10) poiss

boy tul-nuvatcome-pst.repkoju.home

`The boy is said to have come home.' (Erelt et al., 2006, p.129)

Desano provides another example of a complex of suxes indicating a tense-evidential, in this case the inferred past. Miller (1999) states that the inferred past is a verb phrase consisting of a nominalized subordinate verb (with -d past) plus a sux indicating gender, number and animacy. This is followed by the verb ari- `be', optionally a distant or recent past marker, and the subject agreement marker. This is exemplied in 11.

3_{In Estonian, indicative forms are contrasted to their evidential counterpart, which expresses reported information.} The present indicative has its own evidential counterpart, but the imperfect, perfect and pluperfect have only one evidential counterpart, which expresses general past tense (Erelt et al., 2006; Kehayov, 2002).

(11) pisadã

cat wai-resh-spc ba-di-g1eat-pst-sg.mári-bi.be-3sg.m

`The cat must have eaten the sh.' (you can see his paw marks on the ground where he ate it)

(Miller, 1999, p.68)

4.3 Hypotheses based on earlier accounts of tense-evidentiality

Although tense-evidentials have been described for individual languages, little is known about the phenomenon cross-linguistically. Nevertheless, there is enough previous research on which we can base some hypotheses related to the questions posed in the introduction. As for the question which tenses are fused with evidentiality, and which are not, the observation by Aikhenvald (2004) that past tenses show most variety in evidential meanings, followed by present and then future serves as a hypothesis here. We would also expect the order past > present > future based on logical reasoning: a speaker has more possible sources of knowledge about past events than about present events (which one would expect to be directly experienced), let alone future events.

The rst typological study on tense-evidentiality Dall'Aglio Hattnher (2013) tells us something about the correlation between certain tense values and certain evidential values. The study focuses on native Brazilian languages, comparing eleven in total. It is written within the framework of Functional Discourse Grammar (Hengeveld & Mackenzie, 2008) and aims at proving a certain part of the theory, thereby focussing on tense-evidentials expressing rsthand information and inference only (Dall'Aglio Hattnher: event perception and deduction, respectively). The following ndings are relevant for the current study. First, in the native Brazilian languages rsthand evidentials do not co-occur with future or pasts referring to a point in time before the speaker's lifetime. This observation will be tested here with data from more languages. Participatory, visual and non-visual evidentials, which fall under rsthand evidence, will also be included. The second important nding is that if inferential evidentials co-occur with present tense, the tense refers to the moment of inferring, which is simultaneous to the moment of speaking, but the result on which the inference is based is prior to that.

As for implicational hierarchies, we might expect to nd the order firsthand ⊂ inferential ⊂ assumptionwithin tense-evidentiality, which is the one found for evidentiality in general (as dis-cussed in section 2 above).

Research on Cuzco Quechua by Martina Faller is relevant for our understanding of the rela-tionship between tense-evidentials and other evidentials. In Faller (2002, 2004), she shows that in Cuzco Quechua, the marker usually analysed as past tense and rsthand evidential -rqa is actually a normal past tense with a rsthand evidential extension (or, in Faller's words, implicature). She concludes this from, among other things, the combinability of -rqa with evidentials such as infer-ential -chá. It does namely combine with reportative and inference evidinfer-entials. An example of the latter is 12.

(12) Para-sha-rqa-n-chá. rain-prog-rqa-3-inf

`It is raining.' (Speaker infers so.) (Faller, 2002, p.149)

Hence, the combinability of evidentials and tense-evidentials could hint at the latter coding primarily tense, with evidentiality being an extension of the meaning of the marker, which means it is not a tense-evidential. Note, however, that Hengeveld and Dall'Aglio Hattnher (n.d.) found that in some cases the cooccurrence of evidentials is not problematic. Especially the reportative is prone to cooccur with other evidentials. Combinations of assumption and visual or non-visual were also found, as well as a dubious instance of assumption and inference. Whether or not cooccurring is problematic, depends thus on the type of evidentials that cooccur. Moreover, as Faller (2002) shows, combinability is only one way of gaining insight into the semantics of (tense-)evidentials.

Summarizing, the hypotheses to be tested in the current study are as follows:

H2a Firsthand evidentials (including participatory, visual and non-visual) are not fused with future or a past tense referring to before the speaker's lifetime.

H2b With present inferentials, the tense refers to the moment of inferring (= moment of speaking), whereas the result on which the inference is based is prior to that.

H3 The expected implicational hierarchy within tense-evidentiality is firsthand ⊂ inferential ⊂ assumption.

H4 If tense-evidentials and other evidentials have semantic overlap within the same language, this might point at the former being in need of reanalysis.

5 Method and sample

In order to give an answer to the questions posed in the introduction, a sample of as many languages with tense-evidentials as possible was established. The convenience sample consists of 36 languages. The languages were identied primarily with help of volumes on evidentiality (Aikhenvald, 2004; Aikhenvald & Dixon, 2003; Chafe & Nichols, 1986; Guentchéva, 1996; Guentchéva & Landaburu, 2007; Johanson & Utas, 2000), and references therein. Other important sources were San Roque and Loughnane (2012) and Dall'Aglio Hattnher (2013). Sources for the individual languages are listed in Appendix B. These are all the languages with suciently described tense-evidentiality that have come to my attention.

There are four criteria for a language to be included in the sample, relating to what was discussed in section 4.1.

1. The marker(s) must express tense and evidentiality simultaneously.

2. It leads from criterion 1 that constructions which do not have both tense and evidentiality as primary meaning are not counted as tense-evidentials (cf. Estonian `must' and Quechuan -rqa discussed above, which only have a secondary tense-evidential reading).

3. If the marker for tense-evidentiality can be analysed as two separate morphemes (one for tense, one for evidentiality) it is only included in the sample if these cannot occur separately. 4. There has to be sucient information available to answer the research questions. Sucient is here dened as an overview of the complete tense-evidentiality system, plus information on which other tenses and which other evidentials the language has. If no information about combinability of tense-evidentials and other evidentials was available, the language was nev-ertheless included in the sample.

The languages in the sample are listed in table 4 below. Information on macro areas, families, and naming and spelling of language names from the World Atlas of Language Structures (Dryer & Haspelmath, 2014). Figure 1 shows the locations where these languages are spoken.

For each language in the sample, a questionnaire with the following items was lled out.4 _The

answers are to be found in Appendix A. They lead to a quantitative analysis for questions 1-3, and a more qualitative one for question 4.

1. Which tense-evidentials can be identied?

2. Are there tenses that do not express evidentiality? 3. Are there evidentials that do not express tense?

4. If the language has evidentials: can these be combined with tense-evidentials? If so, how?

4_{The marking of tense-evidentiality was also investigated. It was found that all languages mark tense-evidentiality} as a verbal sux, except for Sanuma, where tense-evidentials are separate particles. As these ndings are not relevant for the main questions raised in the current study, we do not take them into account in the results and discussion section.

Table 4: Sample of 36 languages with tense-evidentiality

macro area family language

North America Algic Cree

Iroquoian Cherokee

South America Arauan Jarawara

Arawakan Tariana Aymaran Aymara Nambikuaran Mamaindê Nambikuára Sabanê Panoan Matses

Quechuan Pariarca Quechua South Conchucos Quechua

Tucanoan Barasano Desano Macuna Secoya Siona Tucano Tuyuca Yanomam Sanuma

Eurasia Altaic Kazakh

Turkish Salar Uzbek Indo-European Bulgarian Kalasha Nakh-Dagestanian Chechen Hinukh Khwarshi Tsez Uralic Estonian Livonian Northern Khanty

Papunesia Oksapmin Oksapmin

Trans-New Guinea Angal Foe Huli

The terminology for past tenses and the dierent evidential values found in the literature was standardized if there was a sucient description of the used term in the source, which was the case for all terms. The labels used here are presented in the left-hand column of table 5 below. The right-hand column shows the terminology used in the original sources. When two terms correspond, no variance exists for that particular term in our sample. In a language with two distinctions in the past, these are labelled `recent past' and `remote past'; a language with three distinctions gets the labels `recent past', `remote past' and `very remote past'. This does not mean that every past refers to the same time span.

Figure 1: Approximate location of the languages in the sample

Table 5: Terminology used for tense and evidentiality

terminology used

in current study gloss divergent terminology from literature

tense fut future future

present prs present

past pst past

recent past rec.pst immediate past, near past

remote past rem.pst recent past, intermediate past, distant past, far past very remote past v.rem.pst distant past, far past

evidentiality participatory pcp participatory

rsthand firsth (direct) experience(d), direct, sensory, witnessed, eyewit-ness, personal knowledge, experiential, visual/sensory, rsthand/speaker's perspective, conrmative

visual vis seen

non-visual nonvis sensory, unseen, heard

non-rsthand nonf indirect experience, experienced, unwitnessed, non-eyewitness, non-personal knowledge, indirect (evi-dence), inferred, second hand/reported perspective, non-conrmative, verication

inferential inf visible/previous evidence, apparent, deduced assumption ass conjecture, supposition, customary, mental deduction reported rep second-hand, hearsay, oblique/evidential

6 Results and discussion

6.1 Summary

The most obvious nding of the current study is that tense-evidentiality is found in many dierent parts of the world, and in all of the areas in which evidentiality tends to occur: North America, South America, Eurasia and Papunesia. Table 6 shows the 36 languages represented in the current study. For the sake of clarity of presentation, the dierent past tenses are merged as `past'. Barasano and Macuna, which have a non-past assumption, are displayed under both future and present. Figure 1 shows where the sample languages are spoken. More languages with tense-evidentiality will most certainly be found when more descriptive material becomes available (e.g. from the Papuan Highlands) or when existing datasets can be reanalysed with help of native speakers.

Table 6: Tense-evidentials for all languages in the sample, past tenses merged.

future present past

rsthand Sabanê Angal, Aymara,

Ma-cuna, Nambikuára, Sabanê, Sanuma, Siona

Angal, Aymara, Chechen, Cherokee, Hinuq, Jarawara, Kalasha, Kazakh, Khwarshi, Matses, Nambikuára, Oksapmin, Pariarca Quechua, Sabanê, Salar, Sanuma, Secoya, Siona, South Conchucos Quechua, Tsez, Turkish, Uzbek

participatory Foe Foe Foe, Oksapmin

visual Foe Barasano, Desano,

Foe, Mamaindê, Tariana, Tucano, Tuyuca

Barasano, Desano, Foe, Mamaindê, Tariana, Tucano, Tuyuca

non-visual - Barasano, Foe, Huli,

Mamaindê, Tariana, Tucano, Tuyuca

Foe, Huli, Mamaindê, Tariana, Tucano, Tuyuca

non-rsthand Aymara, Bulgarian Bulgarian, Cree,

Sanuma Angal, Aymara, Bulgarian, Chechen, Chero-kee, Cree, Hinuq, Jarawara, Kalasha, Kazakh, Khwarshi, Northern Khanty, Pariarca Quechua, Sabanê, Salar, Sanuma, Secoya, South Conchucos Quechua, Tsez, Turkish, Uzbek

inferential Foe Foe, Mamaindê,

Nambikuára, Tu-cano, Tuyuca

Desano, Foe, Macuna, Matses, Mamaindê, Nambikuára, Sabanê, Tariana, Tucano, Tuyuca

assumption Barasano (nonpst), Macuna (nonpst), Sanuma

Barasano (nonpst), Desano, Foe, Ma-cuna (nonpst), Tuyuca

Barasano, Desano, Foe, Macuna, Matses, Nambikuára, Tariana, Tuyuca

reportative - Desano, Estonian,

Livonian, Macuna, Northern Khanty, Tariana

Desano, Estonian, Livonian, Macuna, Nam-bikuára, Tariana, Tucano, Tuyuca

The following sections discuss the main questions of this study in turn.

6.2 Tenses used to express evidentiality

As predicted by Aikhenvald (2004), past tenses are most frequently fused with evidentials, followed by present and nally future. All 36 languages in the sample have at least one past evidential, 20 languages have a present evidential and 5 languages have a future evidential. Two languages have a non-past evidential. The future evidentials are dubious in some cases, and deserve a more detailed analysis. These cases are elaborated on in section 6.3.2. The explanation for the found order is related to the logic of tense. While all information can be gathered in the past or be based on some past result, some information is less likely to be gathered at the moment of speaking (i.e. in the present) or to be based on present result.

The dierences between how heavily the tenses are used for evidentiality cannot be explained by dierences in the number of languages that have grammaticalised certain tenses (in which case one would expect proportionally less grammaticalised present and future). Only Jarawara Dixon

and Vogel (2004) and arguably Cherokee (Foster, 1985) do not have grammaticalised present, and the majority of the languages in the sample have a grammaticalised future.

6.3 The interrelation between tense and evidentiality

6.3.1 Summary

The hypothesis that rsthand evidence is never fused with a past referring to a point in time before the speaker's lifetime is conrmed. However, two languages seem to contradict Dall'Aglio Hattnher's (2013) nding that rsthand evidence is not fused with future tense. Sabanê and Foe are reported to have a rsthand future and a visual and participatory future respectively. For a discussion of the questions these markers raise, see section 6.3.2, which discusses future evidentials. Future evidentiality is problematic in the sense that future has the modal meaning irrealis, which seems to clash sometimes with evidentiality. There seems to be no problem for future evidentials where past or present evidence is used to predict the future, but rsthand futures cause semantic problems: you cannot have seen something that has not happened (yet).

Another nding of Dall'Aglio Hattnher (2013) is nuanced by the present sample. In Dall'Aglio Hattnher's sample, the tense of inferential present tense-evidentials refers to the moment of inferring of the evidence, but the perceiving of this evidence was done prior to the moment of speaking. Foe actually has the possibility to add an extra dimension to this: the language has a present inferential for `visible results' (where result-seeing, inferring and utterance are at the same time) and one for `previous results' (where result-seeing is prior to the utterance). In section 6.3.3 this is related to the concept of `double tense' (Fleck 2007). That section is devoted to the interactions we see with present and past evidentials. We see a dierence between the visual and non-rsthand evidentials, which are about equally fused with present and past, and non-rsthand (including inference, assumption and reportative) evidentials, which are much more common with past tense. This can be explained as follows: visual and other sensory evidence is always possible to gather in the present tense, while inferring or assuming often requires some time, as well as being informed by someone else.

No instances were found of the combination of non-visual or reportative evidence and future tense.

6.3.2 Future tense-evidentials

The current sample shows no instances of non-visual future and reportative future. We did nd in-stances of rsthand, visual and participatory future, which seem to be in contradiction to Dall'Aglio Hattnher's (2013) ndings. After discussing these instances, we will discuss the other future evi-dentials. These are inferential future in Foe, non-rsthand future in Aymara and Bulgarian, and assumption future in Sanuma.

The languages seemingly contradicting Dall'Aglio Hattnher's (2013) claim about the non-compatibility of future and rsthand evidentiality are Sabanê and Foe. Sabanê is described as having a future rsthand, as illustrated in 13.

(13) amayl-i-telon. rain-vs-fut.firsth `It is going to rain.' (de Araujo, 2004, p.146)

Although described as an evidential by the author of the grammar, the rsthand future should probably rather be considered a marker of certainty. De Araujo (2004) seems to imply this himself by presenting a minimal pair of an evidentially unmarked future and a rsthand future, explaining that the rsthand future is used when the speaker is certain about the event, and the neutral future when evidence is lacking. See 14, where the speaker in example a is "certain about the event (because there is sensory evidence, for example)" (de Araujo, 2004, p.139).

(14) a. t-ilup-a-telon.

1.obj-vomit-vs-fut.firsth `I am going to vomit.'

b. t-ilup-a-tapanal-i.

1.obj-vomit-vs-fut.neut-ass `I will vomit.'

(de Araujo, 2004, p.139)

Perhaps the marker is analysable as a prospectual aspect marker (Comrie, 1976, p.64), which relates a present state to a subsequent situation. It indicates that something is about to happen, e.g. you see someone tripping and know he is about to fall. Prospectual markers, like resultatives, might then be good candidates to develop into evidentials. In any case, more research is needed to dene the status of the Sabanê future rsthand.

Foe is analysed by Rule (1977) as having a participatory and a visual future. The examples he gives are presented in examples 15 and 16. Glosses are not available.

(15) na davi wa'agerege. (or: na davi wa'anebege) `I will come in 2 days' time.'

(Rule, 1977, p.75)

(16) gavemane bare wiki me samo wa'anege. `The government plane will come next week.' (Rule, 1977, p.75)

The participatory future in 15 is also described as a factual marker, without elaborating on its exact semantics.5 _{Leaving the analysis of the participatory future as a factual marker aside, this}

form does not seem to be problematic, although its meaning remains vague. On the one hand, the speaker him/herself is the authority when it comes to future actions he/she is involved in, which makes the speaker the source of information. On the other hand, a participatory future seems to be a kind of inferring, but in this case based on a personal `sensation'.

Based on example 16, which is the only example available of a visual future in Foe, we could cast some doubt on the analysis of this marker (-'anege) as a visual future. We do not know the context of the utterance, but it might be that the speaker knows the plane comes biweekly and hence can predict its next coming. In that case we should rather analyse -'anege as an assumption future, since the speaker assumes based on what he or she knows about the world. Foe has tense-evidentials in the assumption category, but they are all in past and present tense, so in principle there is an empty cell in the paradigm for an assumption future. This is a mere suggestion for an alternative analysis which can only be conrmed by gathering more native-speaker data.

Besides the visual and participatory futures discussed above, Foe is argued to have two inferential (visible results and previous results) futures. See 17 and 18.

(17) me gaboba'ae; kagi wa'aiba'ae.

`The sky is dark; it will rain.' (going on the evidence of the dark clouds) (Rule, 1977, p.75)

(18) ba'a ko'oso gamagehamo wa'abege.

`Master Ko'oso will come later on.' (he always comes later on, so he surely will this time) (Rule, 1977, p.75)

Example 17 contains the inferential future -'aiba'ae. In this example, current evidence (dark clouds) is used to do a prediction for the future (it will rain). The other inferential, 18, is the one used when the inferring is done based on previous evidence. These examples do not seem to pose any problems for usage with future tense.

For Aymara, it is claimed that the future expresses non-rsthand evidence (Coler, n.d.). This seems logical: one cannot possibly have rsthand evidence of future events. Indeed, local jatiri (future-tellers, hence people who claim to have rsthand knowledge of the future) use the tense to

5_{San Roque and Loughnane (2012) do elaborate on markers of factuality, analysing them as part of evidential} systems. Their meaning relates to knowledge that the speaker obtained from experience, and seems to be close to common knowledge (which is often included with assumption evidentials). This is not the place to discuss whether factuality should be regarded an instance of evidentiality. The interested reader is referred to Oswalt (1986) and Mushin (2001).

lessen their level of certainty when predicting the future. But earlier analyses of Aymara speak of the future as having also a rsthand knowledge reading (Hardman, 1986, p.116), this being "the result of looking back over one's shoulder to see what is coming from behind. The future in these [Jaqi] languages begins immediately beyond the present moment; if I am standing in the door with one foot out and coat on, I will still use a future [to say that I'm going]."6 _{This suggests that any}

evidential meaning of the future tense is an extension. An example is given in 19. (19) Kuwintt'amamawa kuwint(a)-t'a-mama-wa tell-m-1>2fut-top Tarukat. taruka-t(a) deer-abl `I will tell you of Deer.'

(Coler, n.d., p.382)

In Bulgarian, the evidentially unmarked future and future-in-past share their evidential coun-terpart: the non-rsthand future. Kehayov (2002) reports that the future evidential is a recent innovation, and is therefore less clearly semantically dened. About the non-rsthand past future perfect (not under consideration here because aspect) Kehayov (2002) writes that even mother-tongue speakers have a hard time using it. No examples of the non-rsthand future are given. In this case, either time or further analysis will tell use more about the semantics of the Bulgarian non-rsthand future.

Sanuma has one evidential value in the future, namely assumption, and it does not have coun-terparts in other tenses in the language. The marker kite is analysed by Borgman (1990) as future, but with description or emotion verbs it is ambiguous between present and future, as in 20.

(20) ulu

childte3sgohihungryipöaug kite.ass.prs/fut `I suppose that the child is/will be hungry.' (Borgman, 1990, p.172)

When exactly this form is used, is unclear. All examples given in the Sanuma grammar (Borgman, 1990) lack context, but it seems to be based on either general knowledge, which would make it a true assumption evidential, or deduced from current or previous evidence (the kid is/has been crying), which would make it an inferential.

In sum, the hypothesis that visual evidence does not fuse with future was not borne out, but the cases we found are problematic and further investigation is needed to dene the exact meaning and use of these markers. When past or present evidence is used to predict the future (examples 17, 18 and maybe 20), there seem to be no semantic problems.

A characteristic all future evidentials share, whether problematic or not, is that the gathering of the evidence and the event about which something is predicted are detached from each other: the evidence is gathered before or while speaking, and the event is in the future. We should therefore interpret a future evidential not as a marker of source of information gathered in the future, but as a marker of data source which helps the speaker predict something about the future. In that case, even rsthand futures should be possible. The order of events for future evidence is dierent from that for past evidentials, where both the event and the gathering of the evidence are before the moment of speaking. Variations to this scheme are presented in the next section with help of the concept of `double tense'.

6.3.3 Present, past and `double tense'

This section treats evidence based on past results (inference) and how this relates to `double tense' (a term coined by Fleck (2007)), the connection between rsthand evidence and present tense, and the dierent degrees of past tense we see in some languages of the sample.

As mentioned above, data from the current study conrm the hypothesis that rsthand evidence is never fused with a past referring to a point in time before the speaker's lifetime. Mamaindê is the only language in the sample that has a very remote past visual, which refers to events from the

6_{See also Núñez and Sweetser (2006) on the perception of space and tense by Aymara speakers, who picture future} events as behind themselves (because what is behind you is invisible, as is the future) and the past as in front of them (because it is known hence visible).

speaker's early childhood or before his/her lifetime (Eberhard, 2012, p.132). However, Eberhard (p.c.) claims that when a speaker uses the visual distant past and talks about events from before his/her lifetime, this is an extension of the meaning of the tense-evidential, which stresses truth of the preposition.

The current study conrms that more evidential distinctions are made in the past than in the present. This goes especially for non-rsthand evidentials (including inference, assumption and reportative), because in these cases evidence is often based on some past result or event. There are languages, however, that can also express these values in the present tense. Example 21 from Mamaindê illustrates how present tense and inference can be combined: the speaker infers something at the moment of speaking.

(21) ta-tukwinPni-tu

poss.1-father.in.law-fns Paik-tu

eld-fnstau-ø-sihna-wa.chop-sbj.3-prs.inf-dec

`My father-in-law is clearing his eld.' (and I know this because both he and his axe are gone)

(Eberhard, 2012, p.146)

Fleck (2007) analyses the relation between tense and inferentials in Matses as having several components: the event itself, the detection of the result of the event on which the inference is based, and the reporting of the evidence (Fleck, 2007). These take place at subsequent points in time. This is illustrated in 22.

(22) |||> event detection result report

The Matses speaker, reporting information he infers, has to use two evidentials: an inferential to indicate the ultimate source of information, plus a rsthand to indicate when the evidence was gathered (relative to the moment of utterance). This is illustrated in 23, and is referred to as `double tense' by Fleck (2007).

(23) a. mayu-n

non.matses.indian-ergbeste-wa-ak-onda-²h.hut-make-rec.pst.inf-rem.pst.firsth-3

`Non-Matses Indians (had) made a hut.' (a recently made hut was discovered by the speaker a long time ago)

b. mayu-n

non.matses.indian-ergbeste-wa-nedak-o-²h.hut-make-rem.pst.inf-rec.pst.firsth-3

`Non-Matses Indians (had) made a hut.' (an old hut was discovered by the speaker a short time ago)

(Fleck, 2007, p.598-590)

In these examples, the inferential marker (-ak or -nedak) indicates how old the evidence was at the time of discovery, and hence refers to the period between the event and the detection of the result. The rsthand marker (-onda or -o) refers to how long ago from the moment of utterance the inference was done, and hence refers to the period between the detection of the result and the report. Dall'Aglio Hattnher (2013) has claimed that double tense is always implicitly there when using an inferential, because there is always an event, the detection of its result, and the reporting of it, even though this is not overtly marked. It is the inferred event that is usually reported; the detection is implicit. Matses is an exception also reporting the detection of the evidence. When detection and reporting coincide, we get a present inferential as in 21 above. The distribution of events looks as in example 24.

(24) ||> event detection result

report

Dall'Aglio Hattnher (2013) has claimed that in present inferentials the inferring happens at the moment of utterance. This inferring is based on the result of some event, which was prior to the

time of speech. This involves another labelling of events than Fleck (2007) proposed. Detection and inferring are now teased apart as two separate events. Example 25 presents the distribution for present inferentials according to Dall'Aglio Hattnher (2013).

(25) ||> detection result inferring

report

The current sample shows another way of dealing with double tense. Foe has two inferentials: one for `visible results' and one for `previous results'. Fused with present, recent past or remote past, these markers tell us something about when the evidence was discovered, but also when the inferring was done relative to time of speaking. Contrary to what Dall'Aglio Hattnher (2013) claims about present inferentials, a Foe speaker has the possibility to express that the detection of the result was prior to the moment of speaking (when the inferring is done, is not clear, corresponding to example 22), but also to make detection and inferring coincide with the time of speaking:

(26) |> detection result inferring report

The following examples are illustrative, with 27a including the visual results present inferential (-boba'ae) and 27b the previous results present inferential (-bubege).

(27) a. agu amena wa-boba'ae.

`The Augu men are coming.' (can see the smoke rising on the Agu track) b. Kabe maduane minage wa-bubege.

`Mr. Maduane is still coming.' (both left together, but the speaker came faster than Maduane, and so he knows he's still on the way)

(Rule, 1977, p.74)

In 27a, the speaker detects a result (rising smoke) and infers at the moment of speaking that the A(u)gu men are coming. In 27b, the speaker infers from an event prior to the moment of utterance that Mr. Maduane is still on his way. It is not clear when the inferring is done: it could be at the moment of speaking, but the speaker could have also inferred earlier on. In that case, the present tense would refer to the event of Mr. Maduane being on his way. In any case, the Foe data add another `double tense' scheme (see 26) to the ones proposed by Fleck (2007) and Dall'Aglio Hattnher (2013).

Now, to continue with other issues dealing with present and past evidentials, we have seen there is a balance between present and past for visual and non-visual evidence: they are about equally often used with this kind of evidence. Barasano is the only language that only has a present non-visual but no past counterpart (the language does have visual present and past). This might be because the system is still developing. We have no information on the development of the Barasano tense-evidentiality markers, but for example for Tariana it has been argued that once a few tenses had fused with evidentials, other tense-evidentials developed to ll up gaps in the paradigm (Aikhenvald, 2003a).

We do not nd equal numbers for rsthand present and past (as one might expect based on the results for visual and non-visual evidence). This is because most languages make a rsthand versus non-rsthand distinction in the past. In these languages, the present is often unmarked for evidentiality, hence only a few languages are listed under rsthand present.

The reason the dierent degrees of past are all merged under `past' in table 6 is that splitting does not reveal any patterns. It is not the case, for example, that the further we go in the past, the less likely we are to nd rsthand evidentiality. If a language has two or three degrees of past and expresses a certain evidential meaning with one past tense, it usually also does with the other(s). As an example, see the tense-evidentiality system of Mamaindê in table 7, where all evidential values are expressed in all tenses.

Table 7: The tense-evidentiality paradigm of Mamaindê).

visual non-visual inference present -lath_{a -nha(P) -sihna} recent past -nãn -hin -ntin remote past let-nãn -le-hin -le-ntin very remote past -hinP -le-hin -sihinP

Apparently tense-evidential paradigms are driven by balance rather than frequency of use. Nam-bikuára is the only exception: it does not use the very remote past for its tense-evidentials (rsthand, inference, assumption). This can be explained by the fact that the Nambikuára very remote past refers to before the speaker's lifetime. Only the Nambikuára reportative can be fused with all past tenses, which is expected as information based on hearsay can carry over many generations. Jarawara is remarkable in the sense that although it has a gapless paradigm (rsthand and non-rsthand in all three pasts, see table 2), the non-non-rsthand remote past is rarely used, but the very remote past is common (Dixon & Vogel, 2004). The reason for this is unknown.

In sum, the evidential meanings seem to be in line with the meaning of the tenses they are fused with, obeying the rules of logic. Evidentials having to do with perception, be it visual or non-visual, occur in present and past alike, because there is nothing to hinder perceptual evidence-gathering in the present tense. Those having to do with mental activity (deducing, assuming) or reportativity, seem to be more limited to the past, because there often has to be a nished event in the past before one can deduce, assume or be reported about it. A number of future evidentials cause some problems and need further analysis. This section also shows that the tense of a tense-evidential does not necessarily refer to the moment of evidence gathering. For future evidentials, evidence gathering is done prior to or at the moment of speaking. For present and past evidentials, the tense does generally refer to the moment of evidence gathering, but Foe has a special inferential tense-evidential which makes clear when the evidence was gathered relative to the moment of speaking. The tense, in these cases, refers to the event about which something is said.

6.4 Implicational hierarchies within tense-evidentiality

Based on the implicational hierarchy found for evidentiality, it was hypothesized that the hierarchy f irsthand ⊂ inf erential ⊂ assumptionmight be found for tense-evidentiality as well. This could not be conrmed. When sorting the results per tense value, the picture is very scattered, see table 8 for an example of present tense-evidentials, which were tried to order as well as possible.

Table 8: Present tense-evidentials sorted per language and per evidential value.

rsthand particip. visual non-vis. non-rsth. inference assumption reported

Aymara + Nambikuára + + Siona + Sanuma + + Sabanê + Angal + Barasano + + Huli + Bulgarian + Cree + Northern Khanty + Foe + + + + + Mamaindê + + + Tucano + + + Tuyuca + + + + Desano + + + Tariana + + + Macuna + + Estonian + Livonian +

However, if we sort the results per evidential value instead of per tense value and look for patterns there, a clear past > present (> future) order appears. In other words, within every evidential value,

past is used more than present which is used more than future (if future is used at all). This is in line with what we see for tense-evidentiality in general (see section 6.2), and can be explained in a similar fashion: all evidential values combine well with past, some (rsthand, participatory, visual, non-visual) just as well with present. At least for visual and non-visual evidentiality, we see that the number of languages that fuse it with past and present is about the same. Future evidentiality is the least likely to be permitted, and for the languages in which it is used, the semantics are often not entirely clear (see section 6.3.2).

Some values show exceptionless implicational hierarchies. These are visual (future ⊂ present = past, see table 9), inferential (future ⊂ present ⊂ past, see table 10) and reportative (present ⊂ past, see table 11). Only two languages exhibit a participatory tense-evidential. Less convincingly, this evidential value also shows an implicational hierarchy (future = present ⊂ past, see table 12).

Table 9: Visual tense-evidentials sorted per language and per tense value.

future present past

Foe + + + Barasano + + Desano + + Mamaindê + + Tariana + + Tucano + + Tuyuca + +

Table 10: Inferential tense-evidentials sorted per language and per tense value.

future present past

Foe + + + Nambikuára + + Mamaindê + + Tucano + + Tuyuca + + Desano + Macuna + Matses + Sabanê + Tariana +

Table 11: Reportative tense-evidentials sorted per language and per tense value.

present past Desano + + Estonian + + Livonian + + Macuna + + Tariana + + Nambikuára + Tucano + Tuyuca +

For the other evidential values, there is always one language that does not behave according to the pattern. For rsthand evidence, this is Macuna, which only has a present rsthand (see table 13). For non-visual evidence, it is Barasano that does not behave like the other languages. The language only has a present non-visual, whereas all the others have their non-visual evidential in both present and past (see table 14). Aymara `lacks' a present rsthand marker. Its future non-rsthand, however, is dubious (see 6.3.2). If we leave this one out, the non-rsthand evidentials also show an exceptionless evidential paradigm (see 15). As for assumption tense-evidentials, Sanuma is the diverging language. It only has a future assumption. Barasano and Macuna have non-past assumption, and are therefore displayed in both future and present (see table 16).

Inferential and reportative are the evidential values where one would most expect to nd im-plicational hierarchies, and indeed these values show exceptionless hierarchies. It would be hard to come up with restrictions for their past form. These values are very closely related to the past,

Table 12: Participatory tense-evidentials sorted per language and per tense value.

future present past

Foe + + +

Oksapmin +

Table 13: Firsthand tense-evidentials sorted per language and per tense value.

future present past

Sabanê + + + Angal + + Aymara + + Nambikuára + + Sanuma + + Siona + + Macuna + Chechen + Cherokee + Hinuq + Jarawara + Kalasha + Kazakh + Khwarshi + Matses + Oksapmin + Pariarca Quechua + Salar + Secoya +

South Conchucos Quechua +

Tsez +

Turkish +

Uzbek +

because inferring is usually done based upon previous evidence, and reported information is also usually from the past. For visual evidence, on the other hand, it could be expected that in some cultures this is only accepted when used with present ("if you do not see it now, it is not visible evidence") - and the same goes for rsthand and non-visual evidence. The results show that visual evidentiality has an exceptionless implicational hierarchy, whereas non-visual and rsthand eviden-tiality have each one exception to the hierarchy. We see that Macuna and Barasano do not accept a past rsthand and past non-visual, respectively. Although it is not unexpected that non-visual and rsthand have an exception, it is remarkable that Barasano has a present and past visual but only a present non-visual. The acceptability of the Barasano visual past and the Barasano tense-evidentiality system in general therefore deserve some further research. The last evidential value, assumption, does not display an implicational hierarchy because Sanuma only allows a future assumption. This could be explained culturally, if we suppose that for Sanuma speakers assuming is something that can only be done about future events. But as for Barasano, the reasons for the `lacking' values are unknown, as well as the exact semantics of the diverging values.

Another noteworthy observation is that although reportativity is the most commonly grammat-icalised evidential value in Hengeveld and Dall'Aglio Hattnher (n.d.), it is the evidential value that is least fused with tense in the current sample: 8 languages have reportative tense-evidentials, and 14 others have a reportative that is not fused with tense. This is a much smaller proportion than in the Hengeveld and Dall'Aglio Hattnher (n.d.) study. Apparently, there is something that prevents reportativity from fusing with tense. The evidential value seems to have a special status, at least for Hengeveld and Dall'Aglio Hattnher (n.d.). They set reportativity apart from the other evidentials, among other things because it combines with several basic illocutions (declarative, interrogative, imperative), and takes both absolute and relative tense within its scope. They argue that repor-tativity operates at a `level' in the grammar having to do with pragmatics, whereas the others operate at a semantic level (within the framework of Functional Discourse Grammar, (Hengeveld & Mackenzie, 2008)). Hence, it is expected that reportativity fuses less with tense than other eviden-tial values.7 _{This does not, however, explain why some languages do fuse reportativity with tense.}

Table 14: Non-visual tense-evidentials sorted per language and per tense value. present past Barasano + Foe + + Huli + + Mamaindê + + Tariana + + Tucano + + Tuyuca + +

Table 15: Non-rsthand tense-evidentials sorted per language and per tense value.

future present past

Bulgarian + + + Aymara + + Cree + + Northern Khanty + + Sanuma + + Angal + Chechen + Cherokee + Hinuq + Jarawara + Kalasha + Kazakh + Khwarshi + Pariarca Quechua + Salar + Secoya +

South Conchucos Quechua +

Tsez +

Turkish +

Uzbek +

Table 16: Assumption tense-evidentials sorted per language and per tense value.

future present past

Barasano + + + Macuna + + + Sanuma + Desano + + Foe + + Tuyuca + + Matses + Nambikuára + Tariana +

Only for one language in the sample, Tariana, it has been suggested how the reportative got fused with tense. The reported present -pida is cognate with reportatives in other languages from the same family (Arawakan). In Tariana, it was reanalysed as a zero-marked present form to t into the tense-evidentiality paradigm, and from there extended to the other tenses (Aikhenvald, 2003a). The main reason for the fusion of tense and reportativity in Tariana is thus to ll up gaps in the paradigm. Whether this can be extended to the other languages with reportative tense-evidentials, remains for further study.

This section has shown that tense-evidentials, when sorted per evidential value, show a past > present (> future) order. Visual, inferential and reportative tense-evidentials show an exceptionless implicational hierarchy.

(Aikhenvald, 2004, p.271). It might be that in the course of grammaticalisation this verbal origin somehow prevented the tense marker to fuse with the evidential, for example because the two markers were not adjacent. This argument probably does not hold, however, as visual and non-visual evidentials also may have verbal origins (in perception verbs, (Aikhenvald, 2004, p.273)), although they are nearly always fused with tense in the current sample. In any case, as we have only very limited information about the origin of the tense-evidentials in the current sample, we cannot check whether the reportative and visual and non-visual evidentials actually have verbal origins.

6.5 The relationship between tense-evidentials and other evidentials

If languages have both evidentials and tense-evidentials, the former are sometimes in complementary distribution with the latter, while other times there is an overlap in evidentiality values. Barasano, for example, has tense-evidentials for visual, non-visual and assumption evidence. In addition, there is an inference evidential, which also seems to be used as a reportative. Pariarca Quechua, on the other hand, has a rsthand/non-rsthand past tense-evidentiality system, in addition to a rsthand and a non-rsthand evidential. They are argued to reinforce each other's meaning. Of the non-rsthand evidential Howard-Malverde (1988, p.138) says that it reinforces "the objective, non-personalized, locutionary attitude of the narrator" when it is combined with a non-rsthand past. Thus, although there is semantic overlap in the Pariarca Quechua evidentiality and tense-evidentiality systems, this does not seem to cause any problems. In line with the ndings in Hengeveld and Dall'Aglio Hattnher (n.d.), several languages in the sample combine a reportative with all kinds of tense-evidentials (e.g. Chechen, Jarawara, South Conchucos Quechua and Tsez). The rest of this section is devoted to discussing two problematic cases, where evidentials seem to override the meaning of tense-evidentials. They were taken into account for the current study anyway, because what is discussed below is deduced from literature, but has not been researched thoroughly enough to draw denitive conclusions.

Oksapmin combines evidentials (reportative and inferential) with tense-evidentials (participa-tory and visual). These seem to be in complementary distribution, but apparently they can be com-bined. San Roque and Loughnane (2012) mention that the Oksapmin tense-evidentials lose their evidential meaning when combined with an inference evidential (or epistemic particle or clitic), as in 28. There, a participatory recent past is overridden by the inferential (note that the latter is not marked on the verb). The recent past meaning of -l remains. When combined with the reportative clitic, this does not happen.

(28) ap

houseixle=xe3pl.poss=focseinfde-l=oeat-ipfv.pcp.rec.pst=quotli-msay-seq nuxul1pe imd-il=xe

mother.and.child-pl=focaptevillageko-N.arrive-pnct

`We thought that they must have eaten theirs already so me and my children came to our house.'

(Loughnane, 2009, p.343)

This raises the question whether we should regard the Oksapmin tense-evidentials as such, or as tenses with an evidential extension. Faller (2004) has argued along similar lines for Cuzco Quechua, undressing the non-rsthand past -sqa as "a deictic element which locates the described eventuality outside the speaker's perceptual eld at topic time" (p.45). To arrive at such an analysis of tense-evidentials, however, requires thorough semantic analysis, of which checking for combinability with other evidentials is only part. This is not within the scope of the current study, but it would be interesting for further research.

Also for Aymara the existing literature raises doubt about the status of their tense-evidentials. In Aymara, where simple tense (present or immediate past) is argued to express rsthand knowledge (Coler, n.d.; Hardman, Vásquez, & Yapita, 2001), the evidential meaning seems to be overridden when the same phrase contains an inferential.

(29) ukat uka-t(a) that-abl jichax jicha-x(a) now-top khuyt'asijwapunchix. khuy(a)-t'a-si-jwa-pun(i)=ch(i)-i-x(a) whistle-mom-refl-bfr-em=inf-3sim-top `Then he really just started to whistle, it seems.'

(Coler, n.d., p.27)

As predicted, the cases where tense-evidentials and evidentials can be combined can point at problems with the analysis of the tense-evidentials, as for Oksapmin and Aymara. Pariarca Quechua, however, shows that this is not necessarily the case. What Oksapmin and Aymara show is that analysing tense markers as tense-evidentials should be done with great care, a topic that will taken up again in the next section.