On Evidentiality in Nuosu Yi

Abstract

This paper delineates the evidential system of Nuosu Yi, which is found to be comprised of a reported evidential and an inferred evidential. We first describe the semantics of the two evidentials in Nuosu Yi and the interaction between them in terms of double evidential marking. Then we analyze the reported evidential by examining its relation to the verb of speech and (in)direct speech, and demonstrate how reported speech expressions give rise to the reported evidential. Finally, two syntactic tests are used to draw a clear-cut line between the epistemic modal and the inferred evidential in Nuosu Yi.

Headnote

This paper delineates the evidential system of Nuosu Yi, which is found to be comprised of a reported evidential and an inferred evidential. We first describe the semantics of the two evidentials in Nuosu Yi and the interaction between them in terms of double evidential marking. Then we analyze the reported evidential by examining its relation to the verb of speech and (in)direct speech, and demonstrate how reported speech expressions give rise to the reported evidential. Finally, two syntactic tests are used to draw a clear-cut line between the epistemic modal and the inferred evidential in Nuosu Yi.

Key words: evidential, modal, Nuosu Yi

...

(ProQuest: ... denotes non-US-ASCII text omitted.)

1. Introduction

Evidentiality is generally accepted to be the category which is responsible for asserting the source of information of a statement. Every language has its own way of specifying this source of information. Some languages have dedicated grammatical markers of evidentiality, while others use lexical or periphrastic strategies to specify the source of information, achieving the same pragmatic effect as those dedicated markers.

Based on a study of more than 500 languages, Aikhenvald (2004:63) developed a typology of evidential marking, based on the following information sources:

(1) a. VISUAL: covers information acquired through seeing.

b. NON-VISUAL SENSORY: covers information acquired through hearing, and is typically extended to smell, taste, and touch.

c. INFERENCE: based on visible or tangible evidence, or result.

d. ASSUMPTION: based on evidence other than visible results; this may include logical reasoning, assumption, or general knowledge.

e. HEARSAY: for reported information with no reference to those by whom it was reported.

f. QUOTATIVE: for reported information with an overt reference to the quoted source.

Although evidentials take the various information sources listed in (1) as their primary meaning, they are usually found to acquire epistemic and mirative extensions. For example, an inferred evidential may express an extended meaning of a degree of probability. Therefore, it is sometimes difficult to decide whether a particular particle is modal or evidential in nature. Different hypotheses have been proposed. Palmer (1986) takes the position that evidentiality is included under modality. Chafe (1986) holds an alternative view that modality is included under evidentiality. Recently, more and more linguists (Aikhenvald 2004; De Haan 1999; Faller 2002; Lazard 1999; Speas 2008) have argued for the position that evidentiality is a category in its own right with source of information as its primary meaning, and not a subcategory of any modality. However, it is hard to find diagnostic tests, especially reliable morphosyntactic tests, to decide whether a particle is a modal or an evidential. This paper offers a case study that aims to tackle this problem by probing different syntactic behaviors of modals and evidentials in Nuosu Yi. With these differences in view, Nuosu Yi gives extra weight to the position that evidentiality is a category in its own right.

Two dedicated evidential particles are found in Nuosu Yi.1 2 The particle di34 is used as a hearsay and quotative evidential. The other particle, dzo34bu33, is used as an inferred and assumed evidential. In Nuosu Yi, the first-hand source of information, both visual and sensory, is not marked. In terms of Aikhenvald's (2004) semantic parameters listed in (1), Nuosu Yi does not have overt markers for either (1a) or (1b), whereas (1c) and (1d) are marked by the inferred and assumed evidential dzo bu , and (1e) and (1f) are marked by the hearsay and quotative evidential di . A similar system is found in another Tibeto-Burman language, Magar (Grunow-Hårsta 2007).2

The evidential system of Nuosu Yi, especially the particle di34, has already been discussed in a number of research works (Chen & Wu 1998; Gerner 2013; Hu 2002; Liu & Gu 2008; Walters 2010). Most research on the evidential system in Nuosu Yi has been focused on the particle di34, and there has been little discussion about the inferred evidential. This paper aims to provide a comprehensive picture of the Nuosu Yi evidential system, serving as another detailed case study of the evidential systems found in Tibeto-Burman languages. We show that the emergence and development of the reported evidential in Nuosu Yi can be traced back to the verb of speech. In addition, we argue that the epistemic modal in Nuosu Yi is distinct from the inferred evidential based on some morphosyntactic tests.

The paper is organized as follows. In §2, we provide a description of the inferred evidential and the reported evidential in Nuosu Yi, and examine the co-occurrence of the two evidentials. In §3, we illustrate the grammaticalization pathway of the reported evidential. In §4, the inferred evidential is compared with the epistemic modal, and two syntactic tests are put forward to tease evidentiality apart from modality. Section 5 is a brief summary.

2. The evidential system of Nuosu Yi

In Tibeto-Burman languages, evidentiality can be expressed either by inflectional affixes as in Qiang (LaPolla 2003), by copulas and auxiliaries as in Lhasa Tibetan (DeLancey 2001), or by particles as in Akha and Lisu (Thurgood 1986; Yu 2003). Nuosu Yi uses particles to mark evidentiality. The evidential system of Nuosu Yi is found to consist of two particles: di34 marks reported and quotative evidentiality, and dzo bu marks inferred and assumed evidentiality. First-hand (visual or sensory) information is not marked. In this section, we shall describe the two evidential particles and their interactions in terms of double evidential marking.

2.1 The inferred and assumed evidential marker

Nuosu Yi does not make a distinction between the inferred evidential and the assumed evidential. The evidential particle dzo bu can be used to indicate that the speaker makes a statement based on evidence he has gathered from observing the result of an action or from general knowledge relevant to a particular event.

(2) i si ma ha dzi dzo bu .

just rain fall INF

'Apparently, it just rained.'

Sentence (2) can be used in a context where the speaker did not witness the rain, but found that everything was wet outside his house. Based on this evidence, he infers that it has just rained. It is important to note that what the speaker actually saw was that everything outside his house was wet. He did not see the process of raining. The semantic core of dzo bu is the source of information, indicating that the information is inferred from some clues. Sometimes it is very difficult to make a distinction between the inferred evidential and the epistemic modal. For example, the epistemic modal tço34dzi21 in (3) indicates the speaker's evaluation of the probability of rain. 3

(3) e sa ma ha azi la tco azi .

soon rain fall come might

'It might rain soon.' (Chen & Wu 1998:155)

The epistemic necessity modality and the inferred evidentiality represent an overlap of modality and evidentiality in the following way: an epistemic necessity modal encodes the sense of necessary truth judgment, and implicates that the information source is an inferential process. Despite their semantic connections, the two categories, evidentiality and modality, have a different semantic focus. Evidentiality mainly deals with the source of information, while (epistemic) modality is mainly concerned with estimation of the likelihood that (some aspect of) a certain state of affairs is/has been/will be true (or false) in the context of the possible world under consideration. Their semantic difference can be observed by using the diagnostic test of 'implicature cancellation', which was originally proposed by Faller (2002:9-10).4 The rationale of this test is that although an epistemic modal implicates that the relevant source of information is inferential, such an implicature can be cancelled in rigid contexts. For example:

(4) a. ~ja tsi thu a dzo , ti i si ma ha dzi dzo bu

lsg evidence NEG-have but just rain fall INFNEG

Ì don't have evidence, but apparently it just rained.'

b. ~ja tsi thu a dzo , ti e sa ma ha tco dzi

lsg evidence NEG-have but soon rain fall come might

Ì don't have evidence, but it might rain soon.'

Since dzo bu encodes the source of information, such encoded information cannot be cancelled, as is shown in (4a), but tco34dzi21 only implicates a source of information, and such an implicature can be cancelled, as is shown in (4b). This test clearly shows that dzo34bu33 and tco34dzi21 carry different encoded information (semantic focus).

Their semantic differences can also be transparently understood by looking at the internal structure of the two words.5 The word tco34dzi21 is composed of two morphemes, tco34 and dzi21; tco34 can be used as a postposition meaning 'toward the direction of', and dzi21 is the copular verb meaning 'become'. The whole word, roughly speaking, means that the speaker holds an attitude that something has a certain degree of probability of developing in some direction. As an epistemic modal, the word tco34dzi21 encodes the degree of the speaker's commitment to a proposition. The word tco34dzi21 does not directly say anything about the source of information.

The word dzo bu is quite different. It is composed of the two morphemes dzo and bu ; dzo can be traced back to the noun dzu 'footprint', and bu is an existential verb related to the existence of footprints. The whole word, roughly speaking, means 'having left a footprint'. When the word is used in sentence-final position, it encodes the meaning that the speaker makes a statement based on some evidence. In other words, dzo bu encodes the source of information. When dzo34bu33 is used, the proposition expressed by the clause is inferred based on some evidence; it implicates that the speaker is not fully committed to the validity of the proposition.

Apart from having different encoded meanings and different compositional meanings based on their internal morphological structures, the two words also differ drastically in syntactic distribution. In Nuosu Yi, modals are immediately adjacent to the predicate verb, but the scope of evidentials extends to the whole proposition. We shall detail their syntactic differences in §4.

Sometimes without direct evidence, a speaker may make a statement based on his world knowledge, marked by the use of sentence-final dzo34bu33. For example:

(5) tsni íce di vi ko ci la o dzo bu .

3sg-GEN family guest LOC arrive come PFT INF

'Apparently, the guest of his family has arrived.'

Sentence (5) may be uttered when the speaker knows that his neighbor is waiting for a very important guest. According to local custom, important guests should be welcomed with a firecracker display, and on hearing the sound of firecrackers, the speaker makes a judgment that the guest has arrived. It is important to note that this example should not be regarded as a case of direct auditory evidence, because the speaker does not make the judgment purely based on the sound of firecrackers. Rather the judgment is also based on the local traditional custom that a family should set off firecrackers to welcome the most important guests. With such a custom as a general assumption, the speaker makes the judgment at the moment of hearing the sound of firecrackers.

Take sentence (6a) as another example. The meaning of this sentence is that the speaker makes an inference that Muga is a smoker, either from observing that Muga's fingers are yellowish, or from logical reasoning based on the fact that the speaker has heard from others that Muga often buys cigarettes. The scope of the sentence-final evidential particle can extend over an action, as well as over a state. For example, in (6b), the scope of dzo34bu33 extends over a state, as is shown in (6c).

(6) a. mu ka ~i ndo d~o bu

Muga tobacco drink INF

Àpparently, Muga smokes.'

b. mu ka zi ndo ma ?JW dzo bu

Muga tobacco drink CL COP INF

Àpparently, Muga is a smoker.'

c. mu ka zi ndo ma ?JW

Muga tobacco drink CL COP

`Muga is a smoker.'

The evidential particle can be peeled off from the sentence without affecting its grammaticality. For example, (6b) and (6c) are both grammatical sentences. The only difference between them is that (6b) has an extra evidential meaning.

2.2 The reported and quotative evidential marker

The particle di34 in Nuosu Yi can be used either as a hearsay (reported) evidential (with an unspecific quotative source) or a quotative evidential (with a specific quotative source). When the particle is used as a hearsay evidential, the exact authorship of the information is not specified. When it is used as a quotative evidential, the exact authorship is specified. In other words, the particle is used as a quotative evidential if the reported information has an overt reference to the quoted source, and as a hearsay (reported) evidential marker if the quoted source is not mentioned.

In (7), di34 is glossed as hearsay (HEARSAY), because the 'author' of the information is not specified.

(7) a ndi hi ma ha azi di .

yesterday rain fall HEARSAY

'It rained yesterday (it is said).'

If the speaker of sentence (7) chooses to specify who told him the information, a direct (or indirect) speech construction or a sentential-topic sentence, which is semantically similar to the English verb-complement clause, has to be used, as shown in (8a) and (8b) respectively. If the source of information is specified, then di34 functions as a quotative evidential marker, glossed as quotative (QUO), as in (8a).

(8) a. mu ka hi ko a ndi hi ma ha d~i di

Muga say SENT.TOP yesterday rain fall QUO

`Muga said that it rained yesterday.'

b. mu ka hi ko a ndi hi ma ha d~i

Muga say SENT.TOP yesterday rain fall

'Muga said that it rained yesterday.'

According to some of our informants, the quotative marker in (8a) in the sentence-final position can be dropped.6 The resulting sentence, (8b), becomes a sentential-topic construction, a special sentence structure in Nuosu Yi. Verbs of cognition can also occur in such sentential-topic constructions.7

(9) a. ya tha ko vo tshi ma yi ha tci dzo .

1sg guess SENT.TOP pig this CL two hundred CL have

'I guess that this pig weighs two hundred kilograms.'

b. ya hm ko a na mu hm -a -sa .

1sg look SENT.TOP more good-NEG-looking

'It seems to me that it is not very pretty.' (Chen & Wu 1998:185-186)

As a reported/quotative evidential marker, di34 can mark the indirect speech (of a specific source) or the direct speech (of a specific source). Example (8a) illustrates the use of di34 in indirect speech, and Example (10) illustrates the use of di34 in direct speech. In a direct speech sentence, the sentence-final quotative marker cannot be dropped. This can be easily observed when first-person or second-person pronouns are used in a quotation. Since the quotative markers in (10) immediately follow direct speech clauses, they cannot be dropped. Otherwise, the sentence would become ungrammatical.

(10) ~ja ko tco : [nw ci nji ?] di ija ko mu

lsg 3sg-Acc to 2sg what name QUO ask when

tshi di ko : [vu ka mi ] di

3sg-Nom say SENT.TOP Vuga name QUO

'When I asked him, 'What is your name?', he said, 'I am called Vuga.' (Zhang & Cai 1995:161)

Interestingly, when a third-person pronoun is used in a quotation, things become very tricky in Nuosu Yi. Let us consider the situation where a girl named Aguo, as the external speaker, said the following two sentences.

(11) a. mu ka hi ko tshi i ip a kha di

Muga say SENT.TOP 3sg-Nom today uncomfortable QUO

Muga. said that he. was uncomfortable today.'

b. mu ka hi ko i i ip a kha di

Muga say SENT.TOP LOG today uncomfortable QUO

`Muga said that he was uncomfortable today.'

The third-person pronoun in Example (11a) can only refer to a person different from Muga. When the external speaker, Aguo, wants to report what Muga said about himself, she has to use the logophor i , as in (11b). When the logophor is used, it conveys a strong sense that the external speaker, Aguo, is repeating what Muga said about himself.

Aikhenvald (2004:133) mentions that a number of African languages use logophoric pronouns as tokens of indirect speech. In Nuosu Yi, however, we are surprised to find that the logophor i33 can also occur in semi-direct speech in narratives. The following two examples are found in two short stories in Chen & Wu (1998). The ordinary and logophoric pronouns are underlined in the 1998;217) two sentences.

(12) a. ni55 kh' 33 i33' 33 ka33 si33 k' 34' a33 o34 di34.

your dog LOG pound dead completely PFT QUO

I have already pounded your dog dead' (the elder brother said). (Chen & Wu 1998:217)

b. si33 z?? 33 ma34 su33 hi21 ko33 tsh?? 21 mu33 o34ko33,

b. si33 z?? 33 ma34 su33 hi21 ko33 tsh?? 21 mu33 o34ko33,

god CL ART say SENT.TOP this.way if

n?? 33 ???? 21??i21 d?? i21 ko33n?? 33 ???? 33tsha33 thi21 si34

2sg tomorrow wake when hot.water scoop take

i33 b?? 34 la33, i33 si21 ko34 phu33 mo33 di34.

LOG give come LOG take LOC spray SFP QUO

'The god said, "If this is the case, when you wake up tomorrow, you scoop some hot water and give it to me, I will take it and spray it around."' (Chen & Wu 1998:267)

It is worth mentioning that we also found in the same grammar book many cases in which the first-person singular pronoun ??a33 is used in direct speech. Notice that when the first-person pronoun ??a33 is used in quotation, there may be ambiguity between direct speech and indirect speech, and the ambiguity, as one of the reviewers suggests, can be eliminated by some phonological clues. If Example (13a) were intended as a direct quotation, there would be a small pause after ko33 and a rise in pitch for the direct quotation. If it had been intended as an indirect quotation, there would be no rise in pitch. Depending on whether it is a direct or an indirect quotation, the first-person pronoun would be the internal speaker (direct quotation) or external speaker (indirect quotation). When (13a) is interpreted as direct speech, it has the same meaning as (13b), but when (13a) is interpreted as indirect speech, the first-person pronoun refers to the external speaker, Aguo in this case.

(13) a. mu ka ni ko ya i qi a kha ai .

Muga say SENT.TOP 1sg today uncomfortable QUO

'Muga said that I was uncomfortable today.'

'Muga said, "I am uncomfortable today."'

b. mu ka ni ko i i qi a kha di .

Muga say SENT.TOP LOG today uncomfortable QUO

'Mugai said that hei was uncomfortable today.'

2.3 Co-occurrence of the two evidentials

As manifested in many languages, two evidentials can co-occur in a sentence. Aikhenvald (2004:88) addresses this phenomenon, saying, 'marking evidentiality more than once is different from the multiple expression of any other category: it is never semantically redundant. Having several evidentiality markers in one clause allows speakers to express subtle nuances relating to types of evidence and information source, either interrelated or independent of one another.' LaPolla (2003:69-70) gives the following illustrative example of double marking of evidentiality in Qiang:

(14) oh, the: 2ib3 %ete-k-u!

oh 3sg drum beat-INF-VIS

'Oh, he WAS playing a drum!'

In (14), -k is the inferred evidential, and -u is the visual evidential. The situation in this sentence is as follows: the speaker first guessed that someone was playing a drum next door, and then went next door and saw the person holding a drum or drumsticks. The combination of the two evidentials has the sense of 'as I had guessed and it's now pretty well confirmed'.

The double marking of evidentiality in Nuosu Yi can be realized by two co-occurrence patterns: reported + inferred, or inferred + reported. The two patterns have different meanings because of the different scopes of evidentials. For example:

(15) a. tshi bo o di

3 sg-Nom go PFT REP

`He left (it is said).'

b. tshi bo o d~o bu

3 sg-Nom go PFT INF

Àpparently, he left.'

c. tshi bo o di d~o bu

3sg-Nom go SFP REP INF

'Apparently, it was said that he had already had the intention of leaving the place.'

In (15c), both the reported and the inferred evidentials are used. The difference between (15b) and (15c) is that the former indicates that the person referred to by the subject-say, Muga-seems to have left, but the latter only indicates that Muga has the intention to leave, but has not yet left. When uttering (15c), the speaker feels a little disappointed and sad as it seems that Muga has already had the intention to depart. Maybe the speaker heard that Muga himself said so, or someone else said so. The speaker first has the impression that Muga wants to leave. Then, after observing Muga's preparations for departure, the speaker utters (15c) to show disappointment and sadness.9

When the reported evidential and inferred evidential are stacked as inferred + reported, the inferred evidential is in the scope of the reported. For example:

(16) mu ka bo o dzo bu di .

Muga go PFT INF REP

'Muga apparently left (it is said).'

The situation of this sentence is that the external speaker-say, Aguo-heard from someone-say, Vuga-that Muga seemed to have left. Aguo's source of information is Vuga. It is Vuga who made the inference that Muga had left, based on his observation that Muga was not in the hotel. The reported evidential marks Aguo's source of information, whereas the inferred evidential indicates that the departure of Muga is apparent to Vuga.

If the source of information is specified, the sentence will be in the form of a reported speech construction, as in (17a).

(17) a. vu ka hi ko [mu ka bo o d~o bu di ].

Vuga say SENT.TOP Muga go PFT INF QUO

`Vuga said that Muga had apparently left.'

`Vuga said, "Muga apparently left."

b. vu ka hi ko [mu ka bo o d~o bu ].

Vuga say SENT.TOP Muga go PFT INF

`Vuga said that Muga had apparently left.'

If we delete the sentence-final quotative marker, we shall have a sentential-topic construction (17b), and the comment clause is marked with the inferred evidential.

3. Grammaticalization of the reported evidential

As an alternative to the use of a reported evidential, a 'direct speech' construction can also serve the function of specifying the source of information. In (18), the exact source of the reported information, Muga, is specified by being the subject of the verb of speech.

(18) mu ka hi ko [i a -bo o] di .

Muga say SENT.TOP LOG NEG-go SFP QUO

'Mugai said that hei would not go.'

The subject of the embedded clause in (18) is the logophoric pronoun, which must be co-referential with the subject in the sentential topic. If the first-person pronoun jja is used in the comment clause, it may refer to either the external speaker or the internal speaker. For example, the pronoun ya in (19) may refer either to the person who utters the whole sentence, or to the internal speaker, Muga.

(19) mu ka hi ko [ya bo de -a -du ] di .

Muga say SENT.TOP 1sg-Nom go NEG-need QUO

'Muga said that I need not go.' (External speaker reading)

'Muga said, "I need not go."' (Internal speaker reading)

The contrast shown in (18) and (19) suggests that i is like a reflexive pronoun bound by the antecedent. As long as there is a semantic binder, no matter whether the binder is syntactically overt or not, the use of the logophoric pronoun is valid. For example, (20a) is also a grammatical sentence, as long as the referent of the logophoric pronoun is identifiable from the context. Examples (20a) and (20b) have almost the same meaning. The only difference between them is that (20a) indicates that the information is simply a direct quotation. Suppose (20a) and (20b) are said by Muga. Then (20a) indicates that Muga has just heard another man, Vuga, say that Vuga himself would not come. Muga is just repeating what Vuga has said. In contrast, (20b) indicates that Muga has heard (somebody say) that Vuga will not come.

(20) a. i a -la o di .

LOG NEG-come SFP REP

'He will not come (as he himself said).'

b. tshi a -la o di .

3sg-Nom NEG-come SFP REP

'He will not come (it is said).'

An alternative way to specify the source of information would be the use of the third-person pronoun rather than the logophor, as shown in (21).

(21) mu ka hi ko [tshi a -bo o ] di .

Muga say SENT.TOP 3sg-Nom NEG-go SFP QUO

'Mugai said that hek would not go.'

There are two different ways in Nuosu Yi to express reported speech. The first way is shown in (21), where the subject and the verb 'say' are packed with a sentential-topic marker, followed by a comment clause. The second form is shown in (22):

(22) a. mu ka ~ja tco [i a -la o ] di

Muga lsg to LOG NEG-come SFP QUO

`Muga~ said to me that he~ would not come.'

b. mu ka ~ja tco [i a ta nja mo ma ?JW ] di

Muga 1 sg to LOG father teacher CL COP QUO

`Muga said to me that his father was a teacher.' (Chen & Wu 1998:182)

In each of the above sentences, there are three arguments altogether. Apparently, the sentence lacks a three-place predicate to link the three arguments. This forces us either to say that the sentencefinal di34 serves as the main verb in (22), or to say that the main verb is missing in (22). Although sentences like (22a) and (22b) can be easily found in a number of Nuosu Yi grammar books, our native speaker informants felt uncomfortable with such sentential structures. They informed us that they felt that something was missing in the sentence. A verb of speech has to be added. It would be better to change example (22a) into the following form:

(23) mu ka ~ja tco hi ko i a -la o di

Muga lsg to say SENT.TOP LOG NEG-come SFP QUO

`Muga~ said to me that he~ would not come.'

This example clearly shows that the sentence-final di34, at least for some native speakers, cannot be

taken as a lexical verb of speech. It also shows that there are at least three different forms of reported speech in Nuosu Yi, summarized in (24). The main verb mi21 and the sentence-final di34 seem to

have formed a discontinuous frame. The collocation force between them is so strong that the sentencefinal di34 can even compensate for the absence of the main verb mi21.

(24) a. subject + mi21 + SENT.TOP [...] di34 (without a recipient)

b. subject + [to sb.] mi21 + SENT.TOP [...] di34 (with a recipient)

c. subject + [to sb.] [...] di34

Another interesting point to note is that di , with a different tone, can be used as a lexical verb of speech.10 For example:

(25) a. mu di ko tshi a -la o di , ?JW ~iw ?

2sg say SENT.TOP 3sg NEG-come PFV QUO COP COP

`You said that he would not come, right?'

b. mu di ko mu shi di ?

2sg say SENT. TOP what QUO

`What did you say?'

In (25) the verb of speech is di . Of special interest is the fact that in (25b) the lexical verb of speech and the sentence-final quotative marker, which differ only in tones, co-occur. Taking these two pieces of evidence into consideration, we argue that with some phonological (tonal) change, the original verb of speech has grammaticalized into a quotative marker, and from the quotative marker into a reported evidential marker.

Nuosu Yi offers us an illuminating example of how reported speech expressions give rise to the reported evidential, since the links in the chain of development can be easily identified in this language. The origin of the reported evidential in Nuosu Yi can be traced back to the lexical verb of speech, a typical phenomenon found in many Tibeto-Burman languages.

Gerner (2013) makes a comparative study of the development of the verb of speech (SAY). According to his analysis, the 'say' verb *di in proto-Yi developed a specific quotative use in Proto-Northern Yi, and further developed an unspecific quotative use in Pre-Nuosu. Based on the first-person constraint he observed in Nuosu Yi, which states that the future tense marker mi34 can only be used with the first-person pronoun, Gerner argues that from the unspecific quotative use, *di branched into *di (unspecific quotative) and *ni (future time reference) in Old Nuosu. In Modern Nuosu Yi, the 'say' verb di21, the reportative di34, the quotative di34, and the future tense marker mi34 can all be traced back to the 'say' verb *di in Proto-Yi. Such a diachronic development is also supported by data from other Northern Yi dialects, such as Neasu and Nasa. Gerner's focus in that paper is on his proposal that the verb of speech can develop into a future tense marker, according to some cross-dialectal evidence and the evidence of the first-person constraint observed in Nuosu Yi, which is reminiscent of evidential markers. In the rest of this section, we shall briefly comment on the grammaticalization of the reported evidential from the perspective of syntactic change.

As we have shown in (25), di21 can be used as a main verb in reported speech. Since di21 and hi are almost interchangeable, as illustrated in (26), the redundancy will trigger and force one of them to change its meaning and function.11

(26) a. su hi ko nm qi ko tçho ai .

others say SENT.TOP 2sg also 3sg-Acc participate QUO

'Others say that you also participated in it.' (Hu 2002:246)

b. su ai ko nm qi ko tçho ai .

others say SENT.TOP 2sg also 3sg-Acc participate QUO

'Others say that you also participated in it.'

If we assume that the origin of the reported speech marker is the lexical verb of speech, then there would be two 'say' verbs in (26b). The hypothetical sentence would be (27a).

11 We wish to thank one of the reviewers for pointing out to us that ai , the older form, is in the process of being replaced by hi , since ai is not a fully-fledged speech verb any more in some contexts. For example:

(i) tsh~ do hi o

3sg word say PFT

`He has spoken.'

tsln do di o

3sg word say PFT

Intended meaning: `He has spoken.'

(27) a. su di ko [nw ip ko tclio ] di

others say SENT.TOP 2sg also 3sg-Acc participate say

Intended meaning: Òthers say that you also participated in it.'

b. su th ku [nw ip ko tclio ] di

others say SENT.TOP 2sg also 3sg-Acc participate say

Intended meaning: Òthers say that you also participated in it.'

c. su di ko [nw ip ko tclio di ].

others say SENT.TOP 2sg also 3sg-Acc participate QUO

'Others say that you also participated in it.'

Example (27a) is a typical SVOV serial verb construction. Either of these two verbs can be grammaticalized. If the first verb is grammaticalized and removed together with the sentential-topic marker, a sentence with an SOV word order will appear, as in (27b). This word order is not found in Nuosu Yi. Therefore, we are not going to explore this possibility.

If the second verb in the sentence-final position is grammaticalized, due to the semantic redundancy with the first 'say' verb, the sentence-final verb will become a functional element. In this process, the tone of the sentence-final verb changed from a primary tone into a sandhi tone, while its syntactic function was reanalyzed as a quotative marker. At this stage, the erstwhile verb of speech began to assume the function of marking a quotation.

The reported evidential may be derived from the bi-clausal structure illustrated in (27c). When the dependent quotative clause undergoes de-subordination, the erstwhile quotative marker in the sentence-final position may become a reported evidential. Aikhenvald (2004:135) refers to the de-subordinated clause shown in (28b) as free indirect speech. When the free indirect speech becomes accepted as an independent sentence, the quotative marker then becomes a reported evidential, as shown in (28c).

(28) a. mu ka hi ko [i a -/a o di ].

Muga say SENT.TOP LOG NEG-come SFP QUO

`Muga~ said that he~ would not go.'

b. i~m*~==h~ ku [i a-/a o di].

Muga say SENT.TOP LOG NEG-come SFP QUO

`(Muga. said that) he~ would not go.'

c. muka a-/a o di.

Muga NEG-come SFP REP

`Muga will not come (it is said).'

During this process, the 'say' verb and the sentential-topic marker, together with the subject, are removed. Thus, a bi-clausal structure becomes a simple clause. This kind of syntactic change is in conformity with Roberts & Roussou's (2008) idea of grammaticalization via structural simplification. The motivation for deletion of the main clause might be a pragmatic need for vague specification of the source of information. With such a de-subordination process, the erstwhile embedded reported clause becomes the matrix clause marked by the sentence-final reported evidential. The grammaticalization pathway of the Nuosu Yi reported evidential can be schematized as shown in Figure 1.

From the figure, we can see that there is a clear relation between the 'say' verb di , the quotative marker (in both direct speech and indirect speech), and the reported evidential. We propose that both the quotative marker and the reported evidential are grammaticalized from the 'say' verb di in Nuosu Yi. The grammaticalization process is triggered by the semantic redundancy between two 'say' verbs: hi and di . By a de-subordination process which fulfills the purpose of obscuring the exact source of information, the erstwhile embedded reported clause becomes the matrix clause, and correspondingly the erstwhile sentence-final quotative marker becomes the reported evidential.

4. Morphosyntactic tests to differentiate the inferred evidential from the epistemic modal

Nuosu Yi has a set of modal verbs, both epistemic and deontic, to express meanings related to what is possible given what the speaker knows and meanings related to what is required or permitted. Between these two types of modals, epistemic modals are more related to evidentiality. Because of their semantic similarity, it is difficult to differentiate an epistemic modal from an inferred evidential, despite their clearly distinct semantic focus: epistemic modals mainly deal with speakers' evaluation of the chance of an event occurring, while inferred evidentials mainly deal with source of information. For example, the meaning of (29a) is that there is a possibility that it will rain in a while; while the meaning of (29b) is that, based on some evidence, the speaker makes an inference that it rained just now.

(29) a. E sa ma ha d~i la tco dzi

soon rain fall come might

Ìt might rain soon.'

b. i si ma ha d~i d~o bu

just rain fall INF

Àpparently, it just rained.'

Aikhenvald (2004), among many others (e.g. De Haan 2005; Lazard 1999; Speas 2008; Willett 1988), argues for the position that evidentiality is a category in its own right with source of information as its primary meaning, and not a subcategory of any modality, or of tense-aspect. However, so far, no clear-cut morphosyntactic tests have been proposed to determine whether a particle is a modal or an evidential. We found that the epistemic modal and the inferred evidential in Nuosu Yi are different in their syntactic behaviors.

We can use the following two morphosyntactic tests to differentiate evidentiality from modality in Nuosu Yi. With these differences in sight, Nuosu Yi gives extra weight to the position that evidentiality is a category in its own right. Firstly, epistemic modals can be negated, but evidentials cannot be negated. This can be observed in the following example. In (30a), the epistemic modal can be negated by the negative infix. However, the inferred evidential in (30b) cannot be negated by the negative infix.

(30) a. E sa ma ha d~i la tco -a -cizi

soon rain fall come NEG-might

Ìt might not rain soon.'

b. i si ma ha d~i d~o -a -bu

just rain fall NEG-INF

Intended meaning: `Not apparently, it just rained.'

Secondly, epistemic modals can be reduplicated to form a reduplicative polar interrogative, but evidentials cannot, as is shown in (31b).

(31) a. E sa ma ha d~i la tco d~i -dii ?

soon rain fall come might-RED

`Might it rain soon?'

b. i si ma ha d~i d~o bu -bu ?

just rain fall INFR-RED

'Did it apparently rain just now?' (Intended meaning)

In Nuosu Yi, when a question is formed on a clause marked for evidentiality, the event (action/state) of the clause can be questioned, but the source of information cannot. For example, the situation described in (32) might be that the speaker met Muga's father, and asked him whether Muga was ready to go. In this case, what is being questioned is not the source of information, but Muga's readiness to go.

(32) mu ka bo bo o di ?

Muga go-RED PFT REP

`People wonder whether Muga is ready to go.'

If the speaker has heard that Muga has left, he may ask for confirmation from other people as 12 demonstrated in (33).

(33) mu ka bo o di da ?

Muga go PFT REP SFP

'(It is said) Muga has left, right?'

In this case, the whole sentence is more like a statement rather than an interrogative. Pragmatically, the speaker can tell sensational news in such a 'distancing' manner. Although the speaker is quite sure of the validity of the news, he does not vouch for it. In this sense, the interrogative sentence (33) is more like a declarative sentence. It is obvious that the speaker is quite sure of Muga's departure, and he assumes too that the hearer also has some information about the same event. The speaker simply wants to ask for further confirmation from the hearer. In this case, the speaker is not questioning the source of information as well. He is discussing the event by pretending that he is not quite sure about Muga's departure. If the hearer turns out to be completely innocent of the event, the hearer will question the speaker's source of information as in (34), which does not contain the reported evidential marker.

(34) kha55di33 hi33 su33 qm33?

who say NMZ be

'Who said it?'

Examples (32) and (33) show that when a question is formed on a clause marked with evidentiality, only the event described in the clause can be questioned.

5. Conclusion

This paper examines the evidential system in Nuosu Yi. The evidential system in Nuosu Yi has two terms: inferred and reported. We detail the semantics of the two evidentials, and the interaction between them in terms of double evidential marking.

We also argue that in Nuosu Yi there is a close relation between the 'say' verb di and the quotative and reported evidential. We propose that both the quotative and the reported evidentials are grammaticalized from the 'say' verb di in Nuosu Yi through structural simplification. The grammaticalization process is triggered by the semantic redundancy between two 'say' verbs: hi and di . By a de-subordination process which fulfills the purpose of obscuring the exact source of information, the erstwhile embedded reported clause becomes the matrix clause, and correspondingly the erstwhile sentence-final (in)direct speech marker becomes the reported evidential.

In Nuosu Yi the inferred evidential cannot be negated, nor can it be reduplicated to form interrogatives. By contrast, epistemic modals can be negated and reduplicated to form interrogatives. Based on these syntactic differences, together with some semantic and morphological evidence, we differentiate the epistemic modal from the inferred evidential in Nuosu Yi, thus substantiating the claim that evidentiality constitutes an independent category distinct from modality.

Footnote

* This article continues our line of inquiry about evidentiality in Nuosu Yi starting from Liu and Gu (2008). In that article, we explicitly proposed that di34 is a reported evidential marker. This paper furthermore offers a comprehensive account of the evidential system of Nuosu Yi. The research has been supported by the MOE Youth Project for Humanities and Social Sciences "A Contrastive Study of Nominal Phrase Structures across Sino-Tibetan Languages" (13YJC740054) and the University of Macau Start-up Research Grant on "Syntactic Structures of Complex Predicates: A Cross-linguistic Perspective" (SRG2015-00001-FAH). We are grateful for all of their support. We would like to thank the anonymous reviewers of our paper and Prof. Hu Suhua for their comments and criticisms, which we were able to take into account in the revision and have substantially improved our analysis in the final version. We are also grateful for discussions of relevant data with Gu Yang, Wu Da, Hu Suhua, Yap Foong-ha, Huang Chenglong, Lama Ziwo, Li Xuping, and Zhang Qingwen. We would also like to thank our informants Hxi Lyrgursse, Qubi Atguop, Chen Quan, and some of their family members in Sichuan Liangshan Yi Autonomous prefecture. This paper would not have been possible without their assistance. Needless to say, all the interpretations and errors are solely our own responsibility.

1 Yi is mainly spoken in the south-western provinces of Sichuan, Yunnan, Guizhou, and Guangxi in China. Across the four provinces, the Yi language has six geographical dialects: northern, eastern, southern, western, southeastern, and central. The data used in this paper are from the Shengzha subdialect of the northern dialect, which serves as a lingua franca in Sichuan Liangshan Yi Autonomous Prefecture. The northern dialect is also called Nuosu Yi, Nosu Yi, Northern Yi, Sichuan Yi, or Liangshan Yi.

2 The only difference is that Magar makes a distinction between the hearsay (reported) and the quotative, and the two evidentials are assumed by two different morphemes. According to Grunow-Hårsta (2007), the hearsay and the quotative can co-occur in a Magar sentence, which is not possible in Nuosu Yi.

3 Abbreviations used in the Nuosu Yi examples are as follows: person is indicated 1,2,3; Acc accusative; ART definite article; CL classifier; COP copula; GEN genitive; INF inferred evidential; LOC locative; LOG logophoric pronoun; NMZ nominalizer; Nom nominative; PFT perfective aspect marker; QUO quotative speech marker; REP reported evidential; RED reduplicant; SENT.TOP sentential topic; SFP sentence final particle; pl plural; sg singular.

4 This test is attributed to an anonymous reviewer.

5 This crucial contrast was brought to our attention by Hu Suhua (personal communication on 12 April 2014). We wish to thank her for her insightful and valuable comments about internal structure and all the information she gave us about the origin of the two words.

6 Actually, during fieldwork, we recorded different views from different informants as to whether sentence-final di34 can be omitted or not. Some informants told us that sentence-final di34 definitely cannot be omitted, while others were more tolerant, saying it can be omitted.

7 Liu & Gu (2008) analyze the word ko33 in (8a) as a complementizer, and (8a) as a verb-complement sentence. Such an analysis cannot explain why the whole sentence exhibits an SVO word order, which is contrary to the general SOV word order in Nuosu Yi. We wish to thank one of the reviewers for offering the following solution to this problem: While most languages construe complement clauses with a complementizer on the complement clause, Nuosu Yi uses a structure in which the subject, the matrix predicate, and ko are packed into the extra-clausal topic position, so the English verb-complement sentence {I said that it rained yesterday.} is expressed in Nuosu as {'In my saying, it rained yesterday.'}. Semantically, (8a) might be a verb-complement sentence, but syntactically, it is a sentential-topic sentence. The complement clause looks like a comment clause that is about the sentential topic 'in my saying'. The same applies to (9) with the sentential topics 'in my guessing/in my understanding'.

8 We wish to thank one of the reviewers for pointing this out to us. In semi-direct speech, some but not all deictic centers are converted. The speaker integrates two perspectives. Semi-direct speech typically conveys an expressive or emotional value. Here are some English examples. Examples (i)-(iii) are taken from Radford (1988:299). We treat (iv) as a semi-direct speech example, because the first-person pronoun is used to refer to the subject in the matrix clause, which is similar to the case in direct speech.

(i) 'Will I get a degree?' John wondered. (direct speech)

(ii) John wondered whether he would get a degree. (indirect speech)

(iii) Johni wondered would hei get a degree. (semi-indirect speech)

(iv) Johni wondered would Ii get a degree. (semi-direct speech)

9 In Nuosu Yi, the morpheme o34 can be used either as a perfective aspect marker or a sentence-final particle indicating a change of state. It is similar to the ambiguous Mandarin Chinese -le. The sentence {Ta zöu-le. (he leave LE)} means 'he left'. In this sentence, the verb-final -le is used as a perfective aspect marker. In contrast, the sentence {Wo zou-le. (I leave LE)} means 'I want to leave'. In this sentence, the morpheme -le is used as a sentence-final particle, indicating that the speaker has changed his idea from staying in the place to leaving the place.

10 There are three citation tones in Nuosu Yi: 55, 33, and 21. There is another sandhi tone, 34, involved in various phonological processes. Nuosu Yi has only these four tones.

12 We wish to thank Lama Ziwo (personal communication on 2 November 2012) for suggesting to us that da is an interrogative particle asking for confirmation of the speaker's presupposition in Nuosu Yi.

References

References

Aikhenvald, Alexandra Y. 2004. Evidentiality. Oxford & New York: Oxford University Press.

Chafe, Wallace L. 1986. Evidentiality in English conversation and academic writing. Evidentiality: The Linguistic Coding of Epistemology, ed. by Wallace L. Chafe & Johanna Nichols, 261-272. Norwood: Ablex.

Chen, Kang, & Da Wu. 1998. Yiyu Yufa [Yi Grammar]. Beijing: Minzu University of China Press.

De Haan, Ferdinand. 1999. Evidentiality and epistemic modality: setting boundaries. Southwest Journal of Linguistics 18.1:83-101.

De Haan, Ferdinand. 2005. Coding of evidentiality. The World Atlas of Language Structures, ed. by Martin Haspelmath, Matthew S. Dryer, David Gil & Bernard Comrie, 318-321. Oxford & New York: Oxford University Press.

DeLancey, Scott. 2001. The mirative and evidentiality. Journal of Pragmatics 33.3:369-382.

Faller, Martina T. 2002. Semantics and Pragmatics of Evidentials in Cuzco Quechua. Dissertation. Stanford: Stanford University.

Gerner, Matthias. 2013. Yi future: tense or evidential? Language and Linguistics 14.1:167-192.

Grunow-Hårsta, Karen. 2007. Evidentiality and mirativity in Magar. Linguistics of the Tibeto-Burman Area 30.2:151-194.

Hu, Suhua. 2002. Yiyu Jiegou Zhuci Yanjiu [Structural Particles of Nuosu Yi]. Beijing: Nationalities Press.

LaPolla, Randy J. 2003. Evidentiality in Qiang. Studies in Evidentiality, ed. by Alexandra Y. Aikhenvald & R. M. W. Dixon, 63-78. Amsterdam & Philadelphia: John Benjamins.

Lazard, Gilbert. 1999. Mirativity, evidentiality, mediativity or other? Linguistic Typology 3.1:91-110.

Liu, Hongyong, & Yang Gu. 2008. Liangshan Yiyu de yinyu biaoji he shizheng biaoji [The speech marker and evidential marker in Nuosu Yi]. Minzu Yuwen [Minority Languages of China] 2008.2:16-23.

Palmer, Frank R. 1986. Mood and Modality. Cambridge & New York: Cambridge University Press.

Radford, Andrew. 1988. Transformational Grammar: A First Course. Cambridge: Cambridge University Press.

Roberts, Ian G., & Anna Roussou. 2008. Syntactic Change: A Minimalist Approach to Grammaticalization. Cambridge & New York: Cambridge University Press.

Speas, Peggy. 2008. On the syntax and semantics of evidentials. Language and Linguistics Compass 2.5:940-965.

Thurgood, Graham. 1986. The Nature and Origins of the Akha Evidentials System. Evidentiality: The Linguistic Coding of Epistemology, ed. by Wallace Chafe & Johanna Nichols, 214-222. Norwood, New Jersey: Ablex.

Walters, Susan. 2010. Plenty of post-verbal particles in the Nuosu Yi clause. Payap University Working Papers in Linguistics 6.2:26-52. Chiang Mai: Linguistics Department, Payap University.

Willett, Thomas. 1988. A cross-linguistic survey of the grammaticalization of evidentiality. Studies in Language 12.1:51-97.

Yu, Defen. 2003. Evidentiality in Shibacha Lisu. Paper presented at 36th International Conference on Sino-Tibetan Languages and Linguistics, LaTrobe University.

PZhang, Yurong, & Kun Cai. 1995. Xiandai Liangshan Yiyu Yufa [Modern Nuosu Yi Grammar]. Beijing: Minzu University of China Press.