Download Understanding Language and Communication: Gricean Meaning and Dialogue Annotation and more Lecture notes Reasoning in PDF only on Docsity! Felicity Conditions Performative speech is neither true nor false, as we’ve argued, but it can certainly “fail” in some sense of the word -- there’s some sense in which a performative must be uttered under “appropriate circumstances.” By way of example, to bet is not merely to utter the words “I bet …, etc.”: someone might do that all right, and yet we might still not agree that he had in fact succeeded in “making a bet,” or at least not entirely, succeeded in betting. To satisfy ourselves of this, we have only to to go to a horse race and, for example, announce our bet after the race is over. None of your interlocutors, you may be sure, will believe you to have succeeded in betting. Nonetheless, under the right circumstances, saying “I bet …, etc.” is precisely what betting is. Sometimes performative speech goes wrong and the intended act -- marrying, betting, bequeathing, christening, baptizing, etc. -- is therefore to some extent a failure. The utterance is then, we may say, not indeed false but in general unhappy. In trying to classify the ways in which things can go wrong we arrive at the Doctrine of the Infelicities. Or, as they are now know, felicity conditions. (A. I) There must exist an accepted conventional procedure having a certain conventional effect, that procedure to include the uttering of certain words by certain persons in certain circumstances, and further, (A. 2) the particular persons and circumstances in a given case must be appropriate for the invocation of the particular procedure invoked. (B. I) The procedure must be executed by all participants both correctly and (B. 2) completely. (Γ. I) Where, as often, the procedure is designed for use by persons having certain thoughts or feelings, or for the inauguration of certain consequential conduct on the part of any participant, then a person participating in and so invoking the procedure must in fact have those thoughts or feelings, and the participants must intend so to conduct themselves, and further (Γ. 2) must actually so conduct themselves subsequently. Indirect Speech Acts [...] Sometimes explicit evidence of our construal of indirect speech acts rises to the linguistic surface, and we get adjacency pairs like: A: “I’d love to help.” B: “Thanks for the offer.” Or: A: “I could eat the whole cake.” FeedbackElicitation, Stalling, Pausing, TurnTake, TurnGrab, TurnAccept, TurnKeep, TurnGive, TurnRelease, SelfCorrection, SelfError, Retraction, Completion (of partner), CorrectMispeaking (of partner), InitGreeting, ReturnGreeting, Apology, Thanking, etc. Across 8 dimensions: AutoFeedback, AlloFeedback, Time Management, Turn Management, Own Communication Management, Partner Communciation Management, Social Obligation Management, Discourse Structuring. If enough and large enough corpora are so annotated, this scheme (and others like it) obviously have the potential to give rise to rich theoretical and practical (engineering) work in the field of dialogue, broadly construed. Coming up: reasoning from context The modern work on deixis really beginning with (John) Lyons (1977 monumental work on Semantics) and (Chuck) Fillmore. The work on pragmatic implicature, conversational and conventional, really beginning with (Herbert Paul) Grice The work on conceptual metaphor primarily from (George) Lakoff, (Mark) Johnson, (Gilles) Fauconnier, (Zoltán) Kövecses, etc. The work on politeness theory primarily from (Penelope) Brown, (Stephen) Levinson, (Dan) Sperber, (Deirdre) Wilson (the last two a relevance-theoretic interpretation) Deixis (background) The concentric circles shift — take, e.g., “The book is over there [on that desk, in the classroom, in this building, at UMD]” vs “All the students are here by now [at the University of Maryland].” Clearly the speaker-proximal area is far smaller in the first sentence. In other words, the meaning of these deictic words and expressions is non-conventional, and approximately inferred. Or I might shrink my zone of intimacy/familiarity in a language that grammatical uses these notions (German Sie/du, Spanish tú/usted, etc.) — for instance my child in one moment might be addressed by du, but in the next (supposing he or she’d done something to upset me, or I needed to discipline him or her) I might address him/her as Sie, shrinking my circle of intimacy for the purpose of driving home the disciplinary nature of my action/utterance. Deixis of place Here/there — “the pragmatically given space, proximal/distal to speaker’s location at CT(, that includes the point or location gesturally indicated).” E.g.: “Place it here,” “Place it there.” This/that — [glossed by Lyons 1977a: 647 as] “the object in a pragmatically given area close to/beyond the speaker’s location at CT” E.g.: “Bring me that book,” “That’s very delicious.” Empathetic deixis: the speaker can shift the deictic centre to that of the addressee/hearer to show emotional closeness or empathy, or perhaps to garner empathy or help one’s argument to go through by conflating one’s own thoughts with those of the hearer. Such a shift has to be recognized contextually; and appropriate inferences may be drawn therefrom. Various languages discretize space along this deictic dimension differently; English has only two pragmatically given discrete locations, the NW Amerindian language Tlingit has four (right here, nearby, over there, way over there), and Malagasy has a six-way contrast along the same dimension. But it still serves to indicate to the hearer — to assure them, in a sense — that the intended referent is available, even in some sense salient in the contextual space. Consider the utterance: “The cathedral was built by the Medicis.” What cathedral? It depends on context, obviously; but note that now how we search for the intended referent also depends on context. If we’re in Italy, we’ll likely assume that the proper cathedral is deictically available along the spatial axis — perhaps it’s ‘the cathedral in this city we’re currently in.’ If we’re in a classroom in America, during a history class, and we fell asleep, we might assume that the intended referent is available along the discourse axis — perhaps it’s ‘the cathedral that was grounded a few utterances ago.’ Or if we just watched a movie that involved a cathedral, perhaps the cathedral is temporally proximal — ‘the cathedral that we just saw a few minutes ago in the film.’ In any case, we can agree that the indicates that the correct referent is salient, in the sense (now that we can formally articulate it) of deictically relatively proximal. Salience examples: salience decays over time (if two women were grounded, probably more recent one in absence of more explicit description); more likely to refer to say, someone in the room or someone mentioned recently than say, to suddenly refer to Hitler or Mahatma Ghandi, although he’s certainly somewhere in the common ground (as a part of our collective cultural knowledge). Frame Semantics Cf. also the notion of profiling found in Ron Langacker’s theory of cognitive grammar (CG), for which consult the eponymous Cognitive Grammar. Now at first blush you’re likely to think, “Well this is some kind of philosopher’s utopia — no one talks like this.” And certainly people don’t always tell the truth, and sometimes we ramble on about irrelevancies, and sometimes we get confused and use more words than were strictly necessary to convey a thought. But this isn’t what Grice is talking about. That’s all at a very superficial level. The fact of it is that even when we’re at each other’s throats verbally, we’re cooperating in the most wonderful way. Suppose I say something to you in the course of a conversation that’s blatantly false — I violate the maxim of quality — “Queen Victoria was made of solid steel.” All of a sudden a metaphor springs into existence — that is, you assume I’m not saying what you and I know to be false and instantly find an interpretation that renders what I say adherent to the maxim of quality. Or suppose I appear to say something completely uncooperative and irrelevant — you ask “What time is it?” And I say “Well the milkman just came.” Strictly speaking, this is a non sequitur if we’re to take the second sentence at literal face value. But you never even consider that I might not be cooperating. You find a set of inferences that connects the two sentences. Or suppose we were British, and you say in the course of a political conversation: “What if Putin were to blockade the gulf and keep all the oil from us?” — “Oh come now, Britain rules the seas!” Of course that hasn’t been true since perhaps before WWII, but you don’t suppose I’m uninformed or lying — you try to find a nearby proposition that I might really be trying to get across, and you hit upon the exact negation of what I said — “Britain has no naval clout anymore; we’re screwed if that happens.” Other quality flouting: “Tehran’s in Turkey, isn’t it, teacher?” “And London’s in Armenia I suppose.” Some quantity flouting: proposition (in a sense that we haven’t time to define or formalize) that he/she could safely infer we meant. In general, this led to theoreticians beginning to increasingly notice the (1) pervasiveness and (2) strangeness of NL metaphor. Combined with notions from cognitive psychology, in particular a series of studies and papers in favor of the perceptual symbol system view of cognition, and the concomittent assumption of the fundamental importance of embodiment to human cognitive processes and reasoning, these observations began to coalesce into an important field of linguistics known as conceptual metaphor theory. We mention here briefly simply because it rounds out our picture of the hardness of a general regime of natural language processing techniques, and (for our purposes) completes our short-list of things that a dialogue system will have to be able to cope with. Can you think of any other metaphor schemata? An argument is a building. We’ve got the framework for a solid argument. If you don’t support your argument with solid facts, the whole thing will collapse. He’s trying to buttress his argument with a lot of irrelevant facts, but it’s still so shaky that it will easily fall apart under criticism. With the groundwork you’ve got [foundation you’ve laid], you can construct a pretty strong argument. Arguments can be undermined ... An argument is a journey. So far, we haven’t covered much ground [terrain]. This is a roundabout [circuitous] argument. We need to go into this further in order to see clearly what’s involved. As we go further into the topic, we find … We have come to a point where we must explore the issues much more deeply. Understanding is seeing [combines with the above to produce …] [combine with journey metaphor] Having come this far, we can now see how Hegel went wrong. We will now show [guides do this] that Green misinterpreted Kant’s account of will. We ought to point out [a guide does this] that no such proof has been found. Dig further into his argument and you’ll discover a great deal. We can see this only if we delve into the issues. Shallow arguments are practically worthless, since they don’t show us very much. [combine with building metaphor] We can now see the outline of the argument. If we look carefully at the structure of the argument … [combine with container metaphor] That is a remarkably transparent argument. I didn’t see that point in your argument. Since your argument isn’t very clear, I can’t see what you’re getting at. Your argument has no content at all -- I can see right through it. Embodied Construction Grammar (ECG) This is of course work undertaken by (Nancy) Chang and (Benjamin) Bergen -- originally under the direction of Jerry Feldman at UC Berkeley. It builds on construction based approaches to syntax/semantics, which are due to Fillmore, Kay, Goldberg, Croft and a number of others -- this is a huge and (still) productive area of research that’s crossed over into NLP via the algorithmic notion of a unification grammar. An insertion sequence might look like: ... C: “Can I get a scone?” E: “Vanilla or strawberry?” C: “Uhhhh — strawberry” E: “Sure thing, sir.” If E were at this point to begin asking about methods of payment, or talking about the weather, we’d consider this the start of a new conversation unit, a new sequence. But there are ways to extend it without opening a new sequence — using a post-sequence expansion, like a farewell/farewell: … E: “Sure thing, sir.” C: “Have a nice day!” E: “You do the same.” At this point, it’s probably impossible to start a new sequence (on E’s part) without some kind of pre-sequence like another summons/answer (“Hey — umm”, “Oh yes?”). Or another very common pre-sequence opening that you’re probably very familiar with is: “Are you doing anything tonight?” The dreaded question. The whole point of the exchange is what’s coming next turn — some kind of dreaded invitation to go do something. In fact, we often answer “Yes” even when it’s not true — and the only reason we do this is because we recognize this as a common, polite pre-expansion to the invitation/accept or invitation/reject sequence. We know what’s coming. Or consider the following sequence collapse, which only occurs because the speakers are subconsciously familiar with the structure of common sequences: “Do you smoke?” “Nah I left them [my cigarettes] at home.” He’s not answering “nah” to whether or not he smokes — he’s answering “nah” to the part of the sequence that was collapsed, the fpp (first pair-part) of the request sequence (“can I bum a cigarette?”). What’s been collapsed is the answer to the pre-expansion context question, which serves no other purpose than to determine whether the request is likely to be felicitous, or is applicable, as well as the fpp (first pair-part) of the request sequence itself.