What’s eating coherence theories – and how to fix it
2.1 The Problems of Unfounded Coherence and Unfounded Connectivity
2.1.5 Coherence is not a necessary or sufficient prerequisite for utterance understanding
A simple argument against coherence as a prerequisite for discourse understanding (and not a scholarly byproduct of discourse understanding which exists only in the text) is the fact that people generally do not have any problems understanding discourse- initial or isolated utterances (Blass 1993: 94):
(8) [Note attached to empty bottles of milk at the door] Three bottles today. (from Blakemore 1988: 236)
This utterance cannot be part of a coherent discourse, as it is entirely isolated. Yet it is still understandable as a note from the owners of the house to the milk man – it is a request for three bottles of milk (ibid.).
Consider also the following utterance:
(9) Dogs must be carried. (from Blakemore 2001a: 103)
As Blakemore points out (ibid.), it is clear that this warning plastered everywhere around the London Underground escalators should not be taken to mean to mean that all passengers have to take a dog with them, but that, if you have a dog with you, you should carry him or her when using an escalator. (9) is not a coherent discourse segment – again, as it is isolated, there are no other segments to connect it with. Still, we are able to arrive at the correct interpretation despite this lack of coherence and the segment’s inherent ambiguity.
Novels which start in medias res are also often incoherent:
(10) ‘So now get up.’ (from Wolf Hall, by Hilary Mantel (2009). Fourth Estate export paperback, 2010, p. 3)
This is the first line in Hilary Mantel’s Wolf Hall. We do not know who is speaking, or to whom he (or she) is talking. We also do not have any discourse segment to relate this one to, even though (10) is marked by the DM ‘so’, which, on a coherence approach, usually indicates some kind of inference or conclusion which follows from previous segment(s). (10), then, is not coherent – but it is understandable in the sense that we, as readers, can form hypotheses of what is going on here. We can assume, for instance, that we are thrust right into the action of a fight here, or that a mother is talking to her child to get out of his or her bed, or that someone is talking to his or her dog. We can form these hypotheses despite the fact that this segment is not coherent – it is only
fighting with his son (and that he had been fighting him before the story started with (10)), and it is only a few pages later that we are provided with the information that the name of this son is Thomas.
Aristophanes’ Lysistrata begins with a DM as well (see also (26) in §1.1.3.2): (11) Ἀλλ’ εἴ τις εἰς Βακχεῖον αὐτὰς ἐκάλεσεν, ἢ ’ς Πανὸς ἢ ’πὶ Κωλιάδ’ εἰς Γενετυλλίδος, οὐδ’ ἂν διελθεῖν ἦν ἂν ὑπὸ τῶν τυμπάνων. νῦν δ’ οὐδεμία πάρεστιν ἐνταυθοῖ γυνή· πλὴν ἥ γ’ ἐμὴ κωμῆτις ἥδ’ ἐξέρχεται. χαῖρ’, ὦ Καλονίκη. (Aristophanes, Lysistrata 1-6) “[Alla] If they were trysting for a Bacchanal, A feast of Pan or Colias or Genetyllis,
The tambourines would block the rowdy streets, But now there's not a woman to be seen
Except—ah, yes—this neighbour of mine yonder. Good day Calonice.” (tr. Lindsay 1926)
Ἀλλά is usally translated as ‘but’, and its “general function” is that it substitutes a previous element for a new one (‘not X but Y’) (Drummen 2009: 152). In this case, of course, there is no previously communicated element to substitute for. As such, its function cannot be described in terms of coherence – it may still substitute for an assumption, but that assumption cannot have been linguistically communicated, and hence cannot be part of any coherent discourse.8
There are other examples of incoherent discourse being perfectly understandable. Blakemore (1988: 236) distinguishes between mention and use:
(12) A: What did Susan say? B: You’ve dropped your purse.
There is both a coherent and an incoherent interpretation for (12): on the coherent interpretation, B would be quoting Susan, and the answer would probably be denoted with quotes: “You’ve dropped your purse.” On the incoherent interpretation, B would
8 In this respect, the beginning of Xenophon’s Hellenica is also interesting. It starts with a δέ utterance, which
would seem to point to an example of discourse-initial δέ. However, most scholars agree that the beginning (Μετὰ δὲ ταῦτα…, or ‘after these things…’) forms a continuation of Thucydides’ History of the Peloponnesian War – in which case δέ would still connect the upcoming utterance to a preceding segment (Brown Ferrario 2014: 183).
point out to A that she dropped her purse – i.e., B would not provide an answer to A’s question, but would remark on something much more urgent (the fact that A has dropped her purse). In this case, it would not be possible to add quotes, of course (ibid.). Both interpretations are possible, and deriving the right one depends on context – if B points to the ground, for instance, and/or his utterance is marked by an urgent intonation, the incoherent interpretation would probably be preferred. The point here is that discourse can be incoherent but still fully and easily understandable, even when an incoherent interpretation is the only option:
(13) A: Where did you put my pen?
B: Oscar’s just brought in a mouse! (adapted from Blakemore 2004: 238)
In this case, B’s answer cannot be taken to be an answer to A’s question – instead, B is (depending on the context) either very excited or distraught about the fact that Oscar has brought in a mouse. Either way, B’s feelings about Oscar’s actions are such that B feels this information is more important than answering A’s question; as a result, B’s utterance is incoherent – his or her answer has nothing to do with the question, or with the context opened by A in posing his or her question. Yet A will have no problems in deriving the correct interpretation of B’s utterance (assuming that he or she knows who Oscar is).
Examples like (8)-(13) leave one of two options: either discourse-initial, isolated and otherwise incoherent utterances are somehow different from multisegment and coherent utterances in how hearers/readers understand them, or coherence is not a necessary prerequisite for understanding (Blass 1993: 94; Blakemore 2001a: 103). As there has not been any attempt to develop a double-pronged theory of utterance understanding (i.e., the first option), the second option seems to be the only logical outcome (Blakemore 1988: 237; Blass 1990: 20-21).
Utterances, then, can be understandable without being coherent. On the other hand, utterances can be coherent without being understandable:
(14) John was late. The station clock had struck nine. It was time for Susan to start work. She took the first essay from the pile. It was by Mary Jones. Mary had not been well for weeks. The doctor told her to take a holiday. The problem was that she couldn’t afford one. Living in London is now very expensive. All central government subsidies to the Greater London Council have been abolished. Paradoxically, this might be seen to follow from the premises of Libertarian Anarchism. The minor premise might be difficult for the reader to discern. Our theorem-providing program does this using a ‘crossed-syllogism’ technique.” (from Blakemore 1988: 233)
Every segment in (14) is related to the previous one via a coherence relation, but the result is nonsense (ibid.). In the same vein, the different coherence relations can be construed in varied ways. Compare (15a) and (15b):
(15) a. Pyongyang has also been sending anti-South leaflets over the DMZ in balloons. Some that landed north of Seoul Tuesday, according to local media, featured an unflattering caricature of Park and warning of "divine vengeance" for what it said were her failures, including the Sewol ferry crisis, the high suicide rate and the Mers outbreak. (from Washington Post, ‘South Korea’s Park vows all-out effort to punish North for provocations,’ by Anna Fifield. 16th of February, 2016.9)
b. Go down Washington Street. Just pick up your left foot, place it down in front of your right foot, transfer your weight from right to left foot, lift your right foot… (from Blakemore 2001a: 102)
Both examples contain Elaborations in Mann et al.’s (1989: 53) sense – they provide additional detail about the situation presented in the first discourse segment. In (15a), we are given further information about the leaflets presented in the first clause; in (15b), we are given further information on how to proceed down Washington Street. However, while (15a) is acceptable, (15b) is not. If both are Elaborations, but they are not both acceptable, there is obviously something else going on which precludes one of them from being a felicitous utterance.
The same is true for Restatements:
(16) a. [In a review of one of the episodes of the Mad Men television series:] Pete Campbell exits the office like he does everything in his life: like an extremely irritating guy who gets everything he wants but doesn’t seem to understand he’s not owed even more. In other words: my hero. (from The A.V. Club, ‘Mad Men: “Mystery Date”’, by Todd VanDerWerff. 9th of April, 2012.10)
b. It’s a mouse; that is, a small grey furry rodent. (adapted from Blakemore 2001a: 103)
In both cases, we get explicit markers of a Restatement (‘in other words’ and ‘that is’; see Blakemore 1993: 103). In (16a), the description of Pete Campbell is reformulated as overlapping with the writer’s conception of his hero; in (16b), the concept of ‘a mouse’ is
9 Accessed via https://www.washingtonpost.com/world/south-koreas-park-vows-all-out-effort-to-punish-
north-for-provocations/2016/02/15/f4c6aece-4d3f-4378-b9c0-eccef80c1966_story.html?hpid=hp_hp- cards_hp-card-world%3Ahomepage%2Fcard, on the 16th of February, 2016.
reformulated. Again, both examples are perfectly fine examples of Restatements; yet only (16a) is felicitous, as the Restatement in (16b) contains information which is presumably already manifest to the speaker’s interlocutor. Coherence theories do not offer a solution to this problem – on a coherence view, it is not apparent how we recognize (16b) and (16b) as being infelicitous despite the obvious coherence relations between the different discourse segments.11
Even the presence of a DM, which, on a coherence approach, would explicitly indicate which coherence relation would hold between discourse segments, does not guarantee understanding. Blass takes the following example (1993: 98; see also example (10) above):
(17) There is mist on the hills, so the spirits are cooking.
As in example (10), ‘so’ indicates that the discourse segment it marks should be understood as some conclusion, or the result of an inference. But it is not clear in what way ‘the spirits are cooking’ is a conclusion:
“[A]re we to assume, in interpreting this utterance, that mist is like smoke in the spirit world, so that whenever it is misty on the hills, this is a sign that the spirits are cooking? Or could it be that the spirits of the hills always cook when it is misty, in order to stay invisible? Different assumptions yield different interpretations, even though in both cases so introduces a conclusion.” (ibid.)
In other words, the assumption ‘there is mist on the hills’ can be taken in one of two ways as input to the inference which leads to the conclusion ‘the spirits are cooking’: one interpretation is that mist is like smoke in the spirit world, the other is that mist is a cover for the spirits to stay hidden while they are cooking. Without knowledge of the right context, we will not be able to reach the speaker-intended interpretation of the utterance, despite the fact that we can recognize the correct coherence relation between the two segments due to the presence of so (Blass 1993: 97-98). On a coherence approach, however, “the role of the context is restricted to establishing coherence
11 It should be noted that contexts can be thought up under which (16b) would be felicitous – for example, if
the speaker had to repeat himself a couple of times, became irritated, and wanted to ridicule her interlocutor for his lack of understanding. However, in most instances, (16b) would be downright strange. The fact that a context can be found under which (16b) makes sense, should not be considered to be evidence for the position that understanding requires coherence; rather, it should be seen against Travis’ (2008: 56) insistence that “[r]egardless […] of what words mean, and of what might actually be said in them, a speaker may mean anything by them, or mean to say absolutely anything at all in using them, hoping to drive home, or insinuate, etc., any imaginable point”. It’s all about context.
relations” (Blakemore 1988: 236; Blass 1990: 20) – this would not help us here, as the coherence relation has already been established by use of the DM ‘so’.
In sum, then, it does not seem plausible to propose a link between the recognition of coherence relations, i.e., coherence (let’s call this proposition P), on the one hand, and utterance understanding (proposition Q) on the other, of the sort P Q. Recognition of coherence relations is neither necessary nor sufficient to attain discourse comprehension (Blakemore 2001a: 103). Yet utterance production is not a free-for-all – although coherence does not seem to be a precondition for utterance understanding, this does not mean that just anything can be uttered (see Blakemore 2001a: 104). Compare (18) with (13) above:
(18) A: Where did you put my pen?
B: Bob Dylan has recorded more than twenty albums.
The difference between (18) and (13) is that B felt that his or her answer in (13) trumped A’s question – B is so enthusiastic (or disgusted) about Oscar bringing in a mouse that he or she thinks this to be more important and urgent than answering A’s question. In (18), by contrast, the information provided by B doesn’t seem to be ‘newsworthy’ in the sense the information provided in (13) is. Yet even here, a context can be thought up in which B’s answer in (18) would be acceptable. Let’s say that A and B had just had an impassioned discussion about how many albums Bob Dylan has released. If B thought that he had recorded more than twenty, for instance, and A did not, then B might utter his or her answer in (18) triumphantly, having looked up Dylan’s discography on Wikipedia, for instance. In that case, B’s answer will be understandable, as A will probably still remember the discussion he or she had had with B. Again, the acceptability of an utterance does not depend on whether it is coherent, but on whether it ‘fits into the context’ in the sense outlined here (see also footnote11 supra). Note also that hearers will usually strain to derive an interpretation which ‘fits into the context’ in this sense, even if an utterance might seem obscure or inappropriate. In other words, discourse understanding seems to be constrained by some principle, even if it is not the principle of coherence (see Blakemore 1988: 245). Coherence would then be a byproduct – a “superficial symptom of something deeper, which is itself the key to comprehension” (Blass 1990: 24; Blakemore 2001a: 114; 2004: 237; Reboul & Moeschler 1996: 66-67; see also Hobbs 1985: 23).12 Many utterances will be ‘coherent’ in the sense
12 See also Hobbs & Redeker (1997: 2), who raise the question “of why coherence relations are needed at all in
discourse processing, since the inferences they induce can and must generally be derived by other means.” Givón (1995: 59-61) also regards coherence as epiphenomenal.
that the context made available in (the) previous utterance(s) will be important in establishing the interpretation of the next one. As we have seen, though, this is not necessary – discourse can still be perfectly understandable without this contextual continuity, and hence without coherence. If only a subset of all understandable utterances is coherent, then coherence cannot be the key to a psychologically plausible account of utterance understanding.