Response to Spronck and Nikitina “Reported speech forms a dedicated syntactic domain”

I am grateful for the chance to respond to this interesting and valuable study. The ubiquity of reported speech constructions in human languages is a remarkable fact about them, bearing out Bakhtin’s (1984: 143) dictum that that we “live in a world of others’ words”. But despite its ubiquity and functional distinctiveness, as Spronck and Nikitina (S&N) show us, the category of reported speech (RS) is harder to pin down than we might think. First of all there are problems with the term itself, given that what RS “reports” may include thought as well as speech, and even when it is (re)presented as speech, may not ever have actually been spoken. Notwithstanding those problems with the term “reported speech”, in practice it seems that the range of phenomena to which it has applied do match up closely with those referred to by alternative terms such as “reported discourse”, “represented speech”, and “constructed dialogue”. Given that, and the fact that “reported speech” is the most commonly used term for it nowadays, S&N’s decision to stick with it seems sensible.

A more serious problem (which is of course perennial in linguistic typology) is that it is difficult to draw a boundary around the range of phenomena to be included under the term. Everyone can agree on such central cases as (1) ^[1] John said: “Look, there is marmalade here!”, but what about cases like (32) “He believes that there is no tooth fairy”, (33) “He sees that she is entering the room” and (34) “I am telling you that he is in for a surprise”. For those cases I take S&N’s position to be that there are no purely formal criteria for whether or not the sentences as such are to be counted as instances of RS. Rather, particular situated uses of them can be determined to be RS or otherwise, according to the three semantic criteria stipulated in their definition: “demonstratedness”, evidentiality and an evaluative epistemic relation between the represented utterance and its speaker. A paradoxical result of that way of determining the matter is that (34), which in form resembles canonical reported speech more than (32) or (33), is found not to be. (More on this below.)

Another aspect of S&N’s argument that on the face of it seems paradoxical is that although their criteria for delimiting the domain of reported speech are all ultimately semantic, their primary aim is to establish that reported speech “constitutes a dedicated syntactic domain” [italics added]. But that aspect of their argument perhaps becomes less paradoxical when we see that, in practice, their way of making their case is to show that, across a wide range of languages, the range of constructions which are typically (though not exclusively) used for reported speech (as defined by S&N’s semantic criteria) are ones which show a number of features that set them apart from other constructions (length of the elements involved, atypical prosody etc.). Nonetheless, in my view it remains to be clarified in what sense the object of their analysis is a class of “constructions”.

However that may be, S&N have made an important contribution to linguistic typology by:

1. developing an abstract, cross-linguistically applicable model of RS, involving two elements: R (the “reported”) and M (the “matrix”), where R can be of any length and M need not correspond to a clause, but may be expressed by a single morpheme, or remain unexpressed;

2. bringing together from the literature a number of interesting cross-linguistic regularities in the properties of M and R, and in relations between elements in M and ones in R.

I am convinced by S&N’s argument that the constructions used for framing reported speech and thought tend to share a number of features across languages, and that the relation between the M element and the R one is generally quite different from other sorts of syntactic relations, but I think it is important to remain attentive to ways in which it may resemble them. In the rest of this contribution, drawing in part on Rumsey (2010) I will briefly present what I regard as some particularly interesting examples of such resemblance, and draw out some of their implications for S&N’s argument.

While there has been much debate about whether RS “complements”, and complement clauses more generally, behave like grammatical objects of the verb, it has been recognized at least since De Roeck (1994) that languages widely in the extent to which that is the case, and that in some languages the resemblance is striking. For example, in the Australian language Martuthunira, a Pama-Nyungan language spoken in the Pilbara region of Western Australia, reported speech complements of the verb wangka “say” are marked with the accusative case, just like more ordinary sorts of grammatical objects. An example is (1).

Martuthunira being a nominative-accusative language, note that in both lines of the above example the subject of the verb wangka, yartapalyu, occurs without any overt case marking, which is how the nominative case is marked (Dench 1994: 66). In that respect it behaves just like the A argument of a transitive verb with a nominal object, as in (2).

This compares interestingly with what happens in Bunuba, a non-Pama-Nyungan language spoken approximately 1000 km to the east/northeast of Martuthunira, in the Kimberley region of Western Australia. Accusative case marking for reported speech is not an option in Bunuba, since there is no accusative case. Rather, case marking follows an “optional ergative marking” pattern, in which S and O NPs are zero marked and A NPs are usually ergative marked – in approximately 90% of clauses with bivalent verbs (Rumsey 2000: 107, 2010: 1657). The semantic and pragmatic correlates of ergative marking in Bunuba are multiple and complex, but one of the necessary (albeit not entirely sufficient) conditions is that in addition to its subject the clause must have a second participant which is a notional patient or goal (Rumsey 2000:108–9). An example is (3).

As discussed in Rumsey (2010), the correlates of ergative vs absolutive case marking in Bunuba include a range of factors of the kind that were shown by Hopper and Thompson (1980) to be cross-linguistically associated with clause-level semantic “transitivity”. One of them is the degree of what Hopper and Thompson, after Timberlake (1977), call “individuation” of the object, which has two components: definiteness and referentiality. In Bunuba the absence of ergative marking on subjects of bivalent verbs is associated with low individuation. Compare for example (3) with (4).

In (3) thinga “foot” is definite and referential, being positively identified as the speaker’s foot by the 1SG object marking in the verb biyga-nginda. By contrast, in (4) wiyi does not refer to any particular woman or women, just “women”.

As in many languages with ergative case marking, one of the places where it occurs in Bunuba is on the subject of verbs of saying. An example is (5).

Just as in clauses with NP objects such as (3) and (4), ergative marking is “optional” on the subjects of verbs that frame reported speech. An example where it does not occur is (6).

The presence vs absence of ergative marking on the subject in (5) vs (6) can be accounted for in a broadly similar way to the difference between (3) and (4). In both cases, the relevant concept from Hopper and Thompson (1980) is that of individuation. But whereas in the case of nominal objects the relevant dimensions of individuation are definiteness and referentiality, in the case of reported speech the relevant variable is the extent to which the framed material is a more-or-less extended locution, the uttering of which is being focused on as a discrete speech event within the narrative frame. ^[2]

Interestingly, as shown in Rumsey (2010), the same patterns are evident in a completely unrelated, Papuan language, Ku Waru, which is spoken in the Western Highlands of Papua New Guinea. Ku Waru too has “optional” ergative marking, the distribution of which shows a statistically significant correlation with object individuation in ordinary two-argument clauses, and with the subjects of verbs of speaking when they are used to frame reported speech which is lower in individuation than in canonical examples of direct quotation such as in (7).

An example of Ku Waru reported speech that does not have ergative marking on the subject of the framing clause is (8).

In keeping with the absence of ergative case marking on na, the reported locution makuna nyikimil here is less individuated than in (7), since, as in example (6) from Bunuba, it pertains to a class of purportedly repeated utterances rather than to a single one. Furthermore, in this case the class of utterances that it pertains to actually includes part of the present one, i. e. the words makuna nyikimil, which are not only being “reported” but reasserted in the here and now of the framing speech event.

At the bottom end of the scale of individuation among “reported” locutions are ones which are anchored entirely in the here and now, i. e. what S&N refer to as the “performative” use of RS.

A Ku Waru example is (9).

The main points that I want to draw from this discussion of Martuthunira, Bunuba and Ku Waru are that:

1. in all three languages the grammatical means that are used for framing reported speech are tightly integrated with those that are used for constructing transitive clauses with NP objects;

2. in the Bunuba and Ku Waru cases, where the presence or absence of “optional” ergative marking is associated with different degrees of transitivity, some of the same kinds of semantic and pragmatic factors figure in those differences both in ordinary transitive clauses and in framings of reported speech;

3. with respect to the factor of “individuation”, the grammar of RS in both Bunuba and Ku Waru treats the relative degree of individuation of speech events as analogous to that of the referents of NPs. In other words, the meta-linguistic relationship between one utterance and another (the framing one and the framed one) is treated as analogous to the relationship between “Agent” and “Patient” within other kinds of clauses.

In view of these points, while I think that Spronck and Nikitina have made an important contribution by highlighting what is most distinctive to reported speech constructions and schematizing in it on semantic grounds, I would not see their perspective as supplanting ones which focus on what they may have in common with other kinds of syntactic relations.

As a final example let us consider the case of performative constructions such as my example (9) (“I say thank you very much”) and S&N’s example (34) (“I am telling you that he is in for a surprise”). Based on the “demonstratedness” and “evidentiality” criteria in their definition of RS, S&N conclude that (34) does not qualify as in instance of it, because it is not a stylized demonstration, but that the speaker “actually means to do the utterance ‘he is in for a surprise’”; and because it does not entail “a perception event somehow contrasting or interacting with the current speech event”. By contrast, drawing on the analogy provided by Hopper and Thomson’s scalar approach to object individuation, I have analyzed (9) as a true instance of RS, placing it at the bottom of a scale of RS individuation that includes (7) at the top and (8) in the middle. Whatever drawbacks that may have from S&N’s perspective, it is advantageous from at least one other perspective: that of child language development. For while performative uses of RS verbs may seem like marginal instances of RS from the perspective of adult speech, they are widely reported (Ely & McCabe 1993; Diessel & Tomasello 2001; Rumsey 2017) to be the first ones learned by children, whose subsequent mastery of more canonical forms of RS such as S&N’s example (1) and my (5) and (7) develops in parallel with children’s emerging awareness of other people’s minds and intentions. Again, my point is not that a more unifying approach to RS is preferable overall to the interesting one developed by S&N, but rather, that each may be more useful for certain purposes than the other. Vive la différence!