564 Lecture 1 Aug. 24 1999

(1) Formal semantics = the study of how language encodes meaning.

(2) The communication process:

Speaker thinks of message --> Speaker encodes message --> Speaker transmits message --> hearer receives message --> hearer decodes message --> hearer gets message.

Semantics is the study of the encoding process: finding out what meanings correspond to what parts of the code, and how they combine to give the final total message.

Both speaker and hearer have to know the code. Note that not only are there different linguistic codes across cultures, there are different codes of body language used to convey meaning. (Solicit some). (Navajo chin pointing, Inuktitut eyebrow raising (for "yes"), different rude gestures across languages.) Some facial codes are universal — smiling, frowning, crying, surprised expression, disgusted expression.

(3) Consider the following situations. Who’s producing meaningful communication? What’s the message? Who’s using the code of English? Is the message received?

Encoding and decoding messages into English (or whatever language you happen to be speaking) is the job of the language faculty. We’ll be studying speakers’ competence in English as our primary subject matter, trying to abstract away from things like speech errors and processing limitations.

We’re not trying to understand how we decide on a particular message, or how we know that the substance we call water is liquid and the animal we call a dog has four legs, or how we know that things with feathers form a cohesive group which deserves a name (birds) while things with sharp pointy parts don’t form a cohesive group which deserves a name (cactuses, porcupines, hedgehogs, carnivores, barbed wire). These are all good and interesting questions, and are studied as part of cognitive science /psychology /psycholinguistics. (The question of what things are nameable will be part of what we look at in 522 next term… plug, plug.)

What we’re going to do is more like helping an English speaker understand Spanish by giving a translation of a Spanish string into English. After that, the English speaker will know what a Spanish speaker means whenever they hear that string. So, for instance, if I said, "The Spanish word mañana means ‘tomorrow’", whenever you heard someone say mañana in Spanish you would know they meant tomorrow.

We’re going to try to translate English (or sometimes other natural languages) into a formal language. What makes a formal language formal is having a "full and explicit interpretation procedure" for all expressions that might occur in the formal language. That is, any you can find the meaning of any expression that is a well-formed part of a formal language by following a given set procedure, and you’re guaranteed that by the end of the procedure you’ve got the meaning of the expression. If we can translate an English expression into a formal language, we’re guaranteed to have a procedure that will give us the meaning of that expression. If we do it right, the procedure will predict when a given sentence of English is incoherent, because the interpretation procedure will break down at some point and not generate a well-formed response.

In that respect, the interpretation procedure of the formal language will be like the procedure used by the mind to interpret sentences. The interpretation procedure used by the mind tells you when a (grammatical) sentence of English is incoherent, that is, has no meaning. Short of enormous advances in neuropsychology (and the future field of theoretical neurosemantics), we can’t find out directly about the actual interpretation procedure the mind uses. The best we can do is to try to model it, which is what our translation into the formal language will accomplish. (deSwart p. 7 last para).

(Now you can see why we’re trying to abstract away from processing errors and so on — just because we can’t interpret center embeddings doesn’t mean they’re not meaningful sentences of English; rather, they just create processing overload.)

Now, in order to model the interpretation procedure the mind uses, we have to know what it does. So our data will be English sentences, and our experiments will be generally asking, "what are legitimate interpretations of this sentence?" Once we know what the legitimate interpretations are, we’ll try to translate the English sentences into the formal language in such a way as to generate all and only the legitimate interpretations. This means semanticists are often doing things like asking, "Can sentence Y mean P?" and introspecting madly.

One of the most obvious kinds of element that we need in our theory is a list of the basic expressions that bear meaning: a sort of dictionary or lexicon. These basic expressions are built up into complex expressions by a set of rules which you’ll know as syntax, producing a complex structure. The meaning of a whole sentence is computed by combining the meaning of the basic expressions in the manner directed by the structure given by the syntax.

That is, the meaning of "The dog licked the girl" isn’t just a concatenation of the meaning "the", "dog", "licked" and "girl". If it was, it’d mean the same thing as "The girl licked the dog". The fact that it doesn’t is a reflection of the fact that it has a structure, which dictates the ways in which the basic elements are combined, which means that the elements in the two sentences are combined in different ways and hence result in a different meaning. The study of the structure of language is outside this course; we’ll be assuming that the (fairly well-agreed-upon) structures for basic sentences which syntactic theory has produced are correct. You don’t necessarily need to understand the rationale behind a given structure, although it’ll help if you do, but you do need to understand the basics of the structure.

So we’re not trying to motivate the actual structures (that’s a syntactician’s job), but the way that the structures control the combination of the meanings of the basic expressions. Syntax will come in again and again, especially towards the end of the course. (Sometimes semantics can help decide between possible alternative structures for a sentence; syntacticians appeal to it all the time).

Another thing we’re not studying, although it’ll also come in again and again, is pragmatics, which is the relationship between meaning, structure and context. The message which is communicated usually depends greatly on the context surrounding the communication. To take one kind of example, consider the sentence, "She is here".

Pragmatics: the study of context-dependent meaning.

Your language faculty will tell you that this means that some independently identified female person is near the speaker. Where exactly the speaker is (and hence the meaning of the word "here") is not specified by the language faculty, but the language faculty tells you how to consult your non-linguistic knowledge to find out where "here" is. The set of instructions that tell you how to find out the reference of "here" is in fact the meaning of "here", and the actual location assigned to it is its reference. (More on this next class). The fact that the reference of "here" has to do with where the speaker actually is at the moment of utterance means that "here" is a deictic expression. Other deictic expressions, whose reference is determined by the time or place of the utterance, include ‘yesterday’, ‘now’, ‘tomorrow’, ‘I’, ‘you’, and tense marking (you know that "John ate a sandwich" happened before the time the sentence was uttered, whenever that was).

Similarly, who exactly "she" is is not necessarily determined by the language faculty, but again, your language faculty will tell you how to consult non-linguistic knowledge (e.g. your list of salient female individuals that you’ve built up in the conversation that this sentence is a part of ) to find out. These are cases where semantics and pragmatics clearly overlap to a certain degree.

Resolving the reference of pronouns is clearly sometimes a linguistic matter and sometimes not; we’ll be looking in depth at some of the theories which deal with the linguistic side of the question.. This whole field is of central interest to computational linguists, who develop algorithms for computers to do this kind of pronoun (or anaphora) resolution.

Another kind of pragmatic information has much less to do with the semantics, and more to do with inferences about what a speaker might mean by conveying a certain message in a certain context. Consider the following situation: Dr. Stevens is a candidate for a philosophy position, and Professor Williams has written a testimonial about him for the hiring committee to read. It says:

In the context of Dr. Stevens applying for a philosophy job, this sentence conveys the idea that Dr. Stevens is not a very good philosopher — if he was, the hiring committee hypothesizes, Dr. Williams would certainly have said so. Dr. Williams knows that she’s supposed to be commenting on Dr. Stevens’ philosophical abilities, and the fact that she doesn’t implies that she doesn’t want to state her opinion — but the fact that she wrote at all must mean that she wants to convey them in a non-direct way. The conclusion is that she thinks Stevens is no good at philosophy.

Or take the example from the book, "It’s warm in here". Uttered upon entering a room from a cold winter’s day, it could simply be a statement of fact, or even a grateful approval of the temperature of the room. If uttered upon entering a room on an unpleasantly hot summer’s day, it could be a criticism. The kinds covenants between speakers which mean that speakers expect certain things about the utterances of other speakers (e.g. that speakers are cooperative and maximally informative) have been dealt with seminally by Grice; they’re often called Gricean principles.

The study of the non-literal conveyance of messages is part of the subject matter of pragmatics, including irony and metaphor. Knowing what the literal interpretation of the sentence is, however, is part of developing a non-literal interpretation, and knowing literal interpretations is the business of semantics.

So we’ve got essentially the following picture, when asking about the relationship of semantics to other parts of linguistic theory: