EVIDENTIALS: SOME PRELIMINARY DISTINCTIONS
James Higginbotham
University of Southern California
Keywords: evidentials, binary constructions, functional heads, first-person authority.
ABSTRACT
I raise several questions about the semantic interpretation of evidentials, supposing throughout that they are only through grammaticalization distinguished from main Verbs, and thus amenable to abstract study, even through languages that do not support evidential morphology. These questions point to distinctions that are often not made, and sometimes not even considered, in the important and growing literature on the topic. A major question is: does a person who asserts an evidential sentence say one thing, or two? Some links to well-known philosophical topics, such as first-person authority, are also explored.
______
The now large, and growing, literature on evidentials relies in considerable measure upon arduous and accurate fieldwork, undertaken from different points of view, and aiming at the classification, and the syntactic and semantic effects, of evidential constructions in the varieties of languages in which they occur. At the same time, one is free to conjecture evidentials of one sort or another in more familiar languages. In consequence, there is some controversy over how to formulate, out of the raw material, a proper cross-linguistic concept. In this article, I will note some distinctions that seem to me essential to concept formation in this area. These distinctions are indeed in some measure noted here and there in the literature on evidentials; but they are relatively neglected in much of it. In many cases, crucial data (often very hard to elicit, even if envisaged) are lacking. The semantic points can, however, be illustrated in any language, English included.
Besides the classificatory questions that I will raise here, there are some larger issues involving the semantic relations between host elements and subordinate elements, which invite inquiry into relations between propositional content and speech act, or in Fregean terms between sense and force. These issues appear in connection with evidentials, but also in a wider domain, taking in even the question of the proper syntactic description of the combination of host and subordinate elements.
In this article, following a brief introduction to the topic, I consider a family of semantic questions concerning evidentials; questions that, so far as I am aware, have not received definitive answers in the literature. My discussion will include a preliminary exploration of English data, and these data in turn will be put to use in an attempt to clarify some of the issues with evidentials.
A note that may help orient the reader to the perspective taken here: I understand the notion evidential in such a way that syntactic appearance of an evidential, e.g. as an affix, or as a main Verb, or in another guise, is not a defining characteristic of the notion. (In this respect, if I understand him correctly, I agree with Rooryck (2001a).) Evidential affixes are presumably functional heads, taking clausal complements (with different effect in view of the distinction between singulary and binary constructions, discussed below). From the point of view taken here, they might as well be main Verbs.
1. Evidential Constructions
In a number of languages, not in general grouped together by historical affinity, evidential morphemes appear optionally or obligatorily on Verbal (also in some cases Nominal) heads. They generally function so as to indicate something about the epistemic background of an assertion, but may also have other implications, for instance for scope of quantifiers (Lecarme (2003)). I will concentrate here on the Verbal case, where a typical sentence from a language with evidentials might be as in (1), with ‘’ representing the evidential:
(1) A horse ran+ through the village.
(1) might be asserted on a variety of grounds: I saw (or thought I saw) a horse; I saw some tracks that I take to be hoofprints (spaced far apart, as if made by a horse galloping); horses run through the village all the time in this season; somebody told me a horse passed by; and so on. The choice of selects amongst these. The selection may be from 2-4 categories (Willett (1988)), not by any means uniform across languages (see Faller (2002), Speas and Tenny (2004), and references cited there). Viewing (1) as asserted, the speaker is indicating something about the background leading to the belief asserted, its “source of information,” the “strength of evidence,” or something of the sort.
From a syntactic point of view, evidentials are high-up verbal affixes that signify a speaker’s relation to her grounds for the assertion of S (or, in the case of at least some interrogatives, ask the respondent’s relation to the grounds upon which he asserts an answer to the question whether S). I use “grounds for assertion” as neutral between two popular interpretations, namely “source of information” and “nature of evidence.” Willett (1988) proposed a hierarchy of up to four evidential categories: (i) personal experience; (ii) direct sensory evidence; (iii) indirect evidence; (iv) hearsay (his terms). From this description, the epistemic background could take any of several forms.
2. Evidence versus Source
There appears to be a divergence between evidentials that have to do with strength of evidence, as opposed to those that make reference to source of information. The two notions will inevitably be hard to disentangle in some cases, but can be separated directly in cases of what is commonly referred to in philosophical discussions as first-person authority.1
The domain of first-person authority comprises assertions, or occurrences of belief, advanced without anything that could count as evidence (though there may be an experiential basis for them), where (in the normal case) the truth of the (sincere) assertion or belief goes without saying. Examples: “My finger hurts;” “I’m thirsty;” “I doubt John will come;” “‘The child seems sleeping’ isn’t a sentence for me;” and the like. Naturally, viewed as utterances, they involve the first person, and the present tense. Thus “My finger hurt yesterday” does not carry first-person authority (perhaps I am misremembering); the assertion “His finger hurts” is always, if reasonable, based on evidence (his behavior, for instance). But I don’t look at my own behavior to determine that my finger hurts, or that I am thirsty, or that I doubt John will come. And when I pronounce that so-and-so is or is not a sentence of my native language, I don’t produce evidence, even though I may be wrong.
Of the writers I have seen, only Garrett (2001) (discussing Tibetan) observes any cases involving first-person authority, mentioning for instance “I am hungry” (where in the normal case it is nonsense to ask, “What’s your evidence?”). This assertion takes what he calls the Tibetan “immediate knowledge” evidential. That label nicely straddles the distinction between source of information (that feeling of hunger in the speaker’s stomach) and any question of evidence. However, I am not aware that the full domain of first-person authority has been exploited for data. Besides the examples above, that domain would include statements ‘I want …’, or ‘I firmly believe ...’, and the like, and would depend upon the first person for the relevant evidential. In this and in other respects to be noted below, the morphosemantics of the evidential appears to be underdetermined by what is now known.
The above purely reflective semantic diagnostics of first-person authority can be associated with the strangeness of the use of examples normally displaying it when they are put in conjunction with English epistemic must. Now, much of the literature on epistemic must takes it to be a modal attaching to the whole of a declarative, as in (2):
(2) Max must be in his office.
indicating at the same time that evidence is in some sense “indirect” (see for instance Izvorski (1997) and references cited there). I find this assumption doubtful, because it does not explain why (2) is not a stronger assertion than the simple (3):
(3) John is in his office.
(a property that epistemic must shares with English qualifiers such as doubtless, or bound to be).2 In any case, epistemic must does indeed indicate that the situation is evidentially odd with respect to contexts that normally fall under first-person authority. Thus the somewhat strained interpretations of assertions of (4) and (5), each of which suggests that the speaker has evidence for something that normally is known without it:
(4) I must be hungry.
(5) I must believe that I am a Communist.
(with the modal understood epistemically). It would be interesting to know whether the “indirect” evidentials give rise to similar interpretations in languages with evidential morphemes, or whether they are simply rejected out of hand.
To sum up: evidentials that invariably signify evidence, rather than source, should be incompatible with normal cases of first-person authority (though of course syncretism remains a possibility). Conversely, if there are evidentials that can be shown only to characterize sources, but never evidence, these should be incompatible with the counterparts of assertions such as (4) and (5), where normal first-person authority is specifically denied.
3. The Object of Evidentiality
Do the complements of evidentials refer to (or quantify over) propositions, or events? It has been noted that some evidentials have historically developed out of, or as extensions of, perception Verbs (Botne (1997)). Garrett (2001) counts one evidential form as that of what he calls direct perception. Now, the perception verbs (in English, particularly ‘see’ and ‘hear’, but also ‘feel’ in the strict sense of apprehension by touch, and including some uses of ‘remember’ and ‘imagine’) may be understood epistemically or non-epistemically. The epistemic case is illustrated by (6), the non-epistemic by (7):
(6) John saw/heard [that Mary was crying]
(7) John saw/heard [Mary cry/crying]
In (6), ‘see’ signifies roughly come to know by using one’s eyes (or metaphorical extensions thereof), and ‘hear’ signifies come to learn that it was said or rumoured. The construction is referentially opaque, and so does not admit substitutivity of identity for singular terms, or existential generalization. In (7), however (following Higginbotham (1983), the extension to remember and imagine being given in Higginbotham (2003)), the perception verbs appear in propria persona, alleging the seeing or hearing of things; i.e., events or states. The tenseless and complementizerless objects serve as existential quantifiers over events e, classified in our examples as events of Mary’s crying. Hence, substitutivity of identity holds without exception, and existential generalization is allowed with respect to all argument positions. But more than this, the truth conditions of the constructions shown in (7) have nothing to do with the agent’s epistemic situation. Thus it might be true that John heard Mary crying even though he thought it was the wind in the trees. Again, if the agent asserts (8), and the complement is perceptual, then the evidential must modify the main clause (as the object does not express a proposition, but rather describes an event):
(8) I saw+direct evidential Mary cry
The evidential would attach to the whole of (8), the truth of Mary cried being an immediate logical consequence of its truth.
Now, Garrett (2001) remarks that the Tibetan direct evidential does not readily occur with complement predicates outside what he, following Kratzer (1995), calls “stage-level” (or, perhaps equivalently, to the active or transient in the sense of Higginbotham (1983)), meaning by this those that are, or are normally, understood to apply to objects from time to time, rather than being permanently included or excluded. This property is characteristic of the complements to perception V as in (8), and thus provides some evidence that the complements are descriptions of events, rather than propositional. It should be possible to test the matter further, but so far as I am aware this task has not been undertaken.
3. Singulary or Binary?
I introduce, or anyway give a name to, a very general distinction. Consider a construction ‘..._...’, having a certain content (which may be more or less complex, as specified further below), and with nothing at all in the position marked by ‘_’. Let X be an element introduced in this position, so that we have constructions
..._..., and ...X...
our question is whether the introduced element X contributes to the content of ‘...X...’, or rather constitutes a further remark, on the side as it were, made in addition to that made by ‘..._...’. In the first case, I shall say that the construction ‘...X...’ is singulary with respect to the augmentation of ‘..._...’ by X, and in the second case that it is binary, with X constituting further material, to be adjoined to whatever content and force attaches to ‘..._...’. (Ambiguity is possible, so that it may be in principle that a construction is either singulary or binary; in which case the question would arise whether the phrase structure is univocal.)
The full characterization of singulary and binary constructions is more complex than that indicated by the sketch of the last paragraph: for we must consider not only the possibility of ambiguity as above, but also whether the host construction ‘..._...’ is itself binary; and if so, how X contributes to the whole. However, even the broad classification just given allows for illustrations of the distinction.
The most obvious example of a binary construction is that in which the element X is an appositive relative clause. Consider (9):
(9) Bill, who is my friend, is coming to dinner.
In asserting (9), I have asserted two things: (i) that Bill is coming to dinner, and (ii) that Bill is my friend. The assertions are not equal in weight: intuitively, (ii) is by way of a “side remark” to the main assertion (i). But (ii) is likewise asserted. Suppose the appositive appears with interrogatives or imperatives, said with appropriate force, as in (10) or (11):
(10) Is Bill, who is your friend, coming to dinner?
(11) Let Bill, who is your friend, come to dinner if he wants to!
The force of (10) is that of asking, we may suppose, and the force of (11) that of exhorting. But the appositive forms no part either of the question or the exhortation; rather, it constitutes an assertion, as it did in (9). These latter observations underscore the correctness of taking the appositive outside the main assertion in (9).
The abstract picture suggested by our example is that in a singulary construction, viewed as uttered with a certain force F, the element X of ‘...X...’ falls under the propositional content uttered with that force; whereas in a binary construction X appears outside ‘..._...’ under F, and is uttered with a force F’, which may be the same as or different from F. We thus have, for the singulary case, something we may put as (12), and for the binary (13):
(12) F[...X...]
(13) F[..._...]; F’[X]
Using ‘A’ for assertive force, ‘Q’ for asking, and ‘I’ for the various speech acts that fall under the imperative mood, the abstract forms of (9)-(11) are as in (14)-(16), respectively:
(14) A[p]; A[q]
(15) Q[p]; A[q]
(16) I[r]; A[q]
Besides the appositive relative, various parenthetical remarks clearly form binary constructions with their hosts. Consider (17):
(17) John is going to Hollywood, by the way.
In asserting (17) one asserts that John is going to Hollywood, and that this fact is “by the way of” whatever is under discussion. The parenthetical may be embedded on the surface, say within the antecedent of a conditional; but it is asserted with a force wholly outside the conditional, say as in (18):
(18) If John goes to Hollywood, by the way, he’ll become a star.
Here it is the whole conditional that is “by the way.” Similar remarks apply to cases that Rooryck (2001) classes as parenthetical, for example (19) and (without the complementizer) (20):
(19) The doors open at 8.00, very likely.
(20) (That) John is a nice fellow, I admit.
With the complementizer, (20) is just a case of inversion, or topicalization. Without it, the construction is necessarily binary. Likewise, (19) must be sharply distinguished from (21):
(21) Very likely, the doors open at 8.00.
inasmuch as (21) is singulary, and freely embeddable, for instance as in (22):
(22) John knows that, very likely, the doors open at 8.00.
In (23), however, the adverb attaches preferably, and perhaps only, to the entire assertion:
(23) John knows that the doors open at 8.00, very likely.
The binary nature of appositive relative constructions, and parentheticals such as ‘by the way’, is not undermined by cases where the content of the appositive or parenthetical interacts with that of the main clause. The following example is due to Frege (1892):
(24) Napoleon, recognizing the danger to his right flank, himself led the guards
againstthe enemy position.
Frege worries a bit about whether the main clause and the appositive are to be taken as just separately asserted, or whether, as intuitive understanding would suggest, the truth of the appositive is to be taken as supplying a reason for the truth of the main clause. (In Frege’s system, the latter would imply a shift from direct to indirect reference.) Ultimately, he concludes that the speaker merely “hints” at the connection, so that (24) must be distinguished from, say, (25):
(25) Napoleon, because he recognized the danger to his right flank, himself led the
guards against the enemy position.
Abbreviating the main clause by ‘p’ and the complement to ‘because’ by ‘q’, the structure of (25) is as in (26):
(26) A[p]; A[p because q].
In (24), however, we have from the linguistic structure just (27):
(27) A[p]; A[q].
This circumstance allows incorporation of what Frege calls a “hint,” which we would generally understand today as speaker’s implication, as in the last assertion in (28):
(28) A[p]; A[q]; A[p because q].
(which would, incidentally, fully justify Frege’s further remark that the clauses should be taken “twice over,” once for their ordinary reference, as in the two assertions, and once for their indirect reference, as in the speaker’s implication).
An example of (the very beginning of) more complex recursion is (29):
(29) John, who is my friend by the way, is coming to dinner.
In this case, I believe, the parenthetical can serve to dismiss the implication that John’s coming to dinner has anything to do with his being a friend of mine. The sequence of assertions is then as in (30):
(30) A[p]; A[q]; A[q is by the way of p].
3. Application to Evidentials
I turn now to the question whether evidentials are singulary or binary with respect to their host clauses, building upon and endeavoring to put in a general perspective some remarks of Faller, Garrett, and Rooryck in work cited above. Consider again our abstract example (1), repeated here:
(1) A horse ran+ through the village.
and let the evidential represent (say), “Somebody told me.” If (1) is a singular construction, then the evidential falls under the assertive force of the whole, so that we have what we may give as (31):