Psych 124/290B

Ecology of Language

Lecture for Wed., Jan. 28, 2004

Language as a system

Eve Sweetser

Dept. of Linguistics

University of California at Berkeley

sweetser@cogsci.berkeley.edu

Classic view:

Language is system of relationships between FORM and a MEANING.

Saussure SIGNIFIANT SIGNIFIE

FORM MEANING

Tree ÒTree conceptÓ

Arbre

Gwezenn

É..

(Query: is the Òtree conceptÓ the same in all these cases???)

The arbitrariness of the sign.

Could you guess that arbre, tree, or gwezenn meant ÒtreeÓ without knowing the relevant linguistic CONVENTIONS?

Iconicity - refers to cases where there is actual similarity between the form and the meaning - classic area of vocabulary to look for this is ÒonomatopoeticÓ or sound-symbolic words:

bow-wow, woof, meow, ding-dong,

cockadoodle-doo

BUT English cockadoodle-doo still is not identical to French cocorico, even though both are sound-symbolic for a roosterÕs crowing. So convention is relevant even here.

Language as a system with an internal ecology of its own, independent of (though not necessarily unrelated to) the rest of human thought and action. (A Structuralist view.)

Suppose I have a language (American English) where I have a word sweater, which covers cardigans and pullovers, and another very similar language (British English) where there is no such general word, just the two sub-labels.

In American English, I can just be vague in saying I need to put on a sweater (and indeed that is the norm - cardigan is a rather technical term not used in everyday American speech), but in British English I apparently have to be specific.

Similarly, compare English use of grandmother, grandfather, uncle, aunt, cousin with other languageÕs kinship terms - in English you CAN say maternal grandmother or paternal uncle or male first cousin on my fatherÕs side - but you DONÕT normally.

In some languages, there is no general term for grandmother or cousin. In Scandinavian languages, where the words for ÒgrandmotherÓ normally mean Òfather motherÓ and Òmother mother,Ó these words are literally equivalent in REFERENCE to English maternal grandmother and paternal grandmother. That is, they identify the same entities in the world.

HOWEVER, because we also have the general word grandmother, maternal grandmother plays a different role in the English system of contrasts between words than Òmother motherÓ would in Swedish.

What about contrasting a language with words like English red and yellow and orange, with a language which has only two words, one of them meaning Òred and reddish orangeÓ and the other meaning Òyellow and yellowish orange.Ó

Even if the best examples of these two color terms were bright red and bright yellow, you still might feel that they donÕt mean the same thing as English red and yellow, because they arenÕt opposed to an orange category.

Perhaps even more interesting - think of grammatical categories as part of such a system.

Imagine a language with only two tenses - Past and Non-Past. Does ÒPastÓ mean the same thing it does in English? Does ÒNon-pastÓ have a good translation into English?

The very assumption that there can be categories which are best described as Ònot XÓ like Ònon-pastÓ is in itself an admission that the real meaning of that grammatical category is not a meaning of its own, but somehow a meaning that comes from its place in a system of oppositions or contrasts.

Some very important aspects of language to linguists are:

(1) Arbitrary form-meaning relationships. (see above)

(2) Compositionality.

That is, you take smaller forms with meanings, you put the forms together, and the meaning of the new larger form is systematically related to (predictable from?) the meanings of the pieces.

At a really simple level, The cat is on the mat means something different from The dog is on the mat because of the predictably different contributions made by the meanings of dog and cat to the meaning of the larger form.

(3) Recursion.

Linguistic systems involve embedding similar kinds of units in each other - sentences in sentences, for example.

As far as we can tell, there is no principled and systematic limit to how many embeddings you can have of ÒJohn knows that Mary thinks that Joe said that Sue believesÉ..Ó

Does this mean that grammar is infinitely recursive?

Certainly not in USAGE - human memory and processing limitations severely limit the number of such embeddings in real language use.

(4) Another important distinction:

Grammar vs. usage

Langue vs. parole

Question: if human limitations on memory, processing, etc. (Òusage constraintsÓ) limit ALL human language use, where would a human get access to a grammar unlimited by usage?

One answer: it could be innate.

Many linguists also think Language is MODULAR.

This means (1) that it is in principle separable (in study? or in actual neural processing?) from other systems of human cognition, perception, action. For example, general memory constraints are not part of Grammar.

It also means (2) that the sub-systems of Language are separable from each other. It should be possible to describe, for example, Syntax (arrangement of formal elements into larger clausal units - word order, etc.) without reference to Meanings (Semantics), or interactional context (Pragmatics).

Coming up later in this class - a discussion of why the gestural medium needs to be taken into account, for us to really understand language, and how that fits in with broader needs to see language as part of a multimodal communicative system in a context of interaction.

Right now - letÕs focus on some more ÒcoreÓ linguistic questions.

LetÕs come back to COMPOSITIONALITY.

One really classic test case is the English modifying adjective construction, ADJ NOUN.

By classic truth-conditional and truth-functional accounts, linguistic reference works by picking out the sets of entities truthfully described as X. Blue picks out the set of blue things, Ball picks out the set of balls.

More complex expressions pick out things like sets of situations where something is true - e.g. the classic Snow is white picks out the set of situations wherein the set of things denoted by snow overlaps with the

set of white things.

The first set of problems with this approach lie in the use of sets as a model for human categories.

For example, color terms such as blue and white are fuzzy-bounded categories - it is a lot easier for speakers to agree on their central best examples than on their precise boundaries. Human linguistic categories are at any rate not always much like classical sets, where there is a clear boundary between X and ~X, and all members are equally members.

But more interesting problems arise, even in interpreting really ÒsimpleÓ examples like red ball.

ACTIVE ZONES.

What part of the ball is red? The outside surface?

The inside material? An identifying dot on the surface?

Nothing in my simple idea of intersecting categories of red things and balls could have prepared me for this problem.

Further, if you asked me to name all the red things in a picture, and the picture contained a white ball with a red dot on it, I would probably include the dot but NOT the ball on my list.

And yet, when you think about how humans actually USE color terms, all this makes perfect sense.

Given a set of otherwise similar objects, we might well use differently colored sub-areas as a way to differentiate them (as with tops of pens).

And in general, human construal involves focusing on relevant aspects and affordances of an object. Of course one has to know what those affordances are - many a human, seeing a bean-bag chair (or an ergonomic rubber-ball seat) for the first time, has had no idea what that object ÒwasÓ, even though its basic physical characteristics were clear enough to a casual inspection.

CATEGORIZATION, then, is not a simple matter of getting yes-no answers to questions about objective properties of objects.

ItÕs fuzzy, itÕs based on experience, and it focuses on human affordances in interaction with the entity.

Active zones are related to the entities in question via METONYMY (part-whole relationships) - the red surface, the red interior material, the red dot are all parts of the ball, and can thus apparently be referred to when we say red ball.

Sometimes what is involved is in fact FRAME METONYMY. For example, red ball could mean ÒThe ball belonging to the team wearing the red

uniforms.Ó

The ball, the team and the red uniforms all belong to a conventional FRAME of team sports, wherein we have conventional ROLES for teams, players, uniforms, balls, etc.

(A FRAME is a conventional scenario of action or situation, which is accessed cognitively as a whole, but has slots for ROLES of participants in the scenario. When I say that it is accessed as a whole, I mean that once I mention ball game, I can then refer to the players, the ball, the uniforms Ð as if I had already mentioned them. All these roles were cognitively activated by mentioning the frame theyÕre part of.

And FRAME METONYMY here means that both the ball and the teamÕs uniforms are parts of a single frame Ð which is how I can somehow say red ball and mean that the red applies not to the ball itself, but to the teamÕs uniforms in the same frame. )

Once again, thereÕs no objective characteristic of the ball which is red in this case. I have to know its broader cognitive/cultural framing to relate it to something red.

METAPHOR comes in as well.

Red ball could mean Òball made in a communist countryÓ

or Òball belonging to the communist countryÕs team.Ó

Again, not an objective characteristic of the ball.

But things get more complicated.

Consider SAFE HOUSE.

Safe evokes a frame involving

(1) a valued object

(2) a potentially dangerous entity which could harm

the valued object

(3) protection; something preventing such harm

House could potentially fill any of those roles!

Safe beach is perhaps less ambiguous - though once again

one can imagine multiple mappings -

This is a safe beach to take your toddler to.

This beach is on the Environmental Protection AgencyÕs list as a safe beach - itÕs not likely to be damaged by major water pollution soon.

The presidential candidate couldnÕt find a safe beach to hide out on for the weekend - the media found her wherever she went.

Fake gun.

This whole class of examples has been problematic for a long time. Obviously a fake gun is not a fake thing - it is also not a gun. SO it canÕt be the intersection of the sets of guns and fake things.

Fake seems to set up a frame, involving two MENTAL SPACES - one is the understanding of the knowing person who realizes that this is not a gun; the other is the understanding of the dupe, who thinks it is or might think it is.

(A toy gun is not a fake gun unless, for example, itÕs used in a play or a holdup.)

Domain modifiers.

intellectual sleeping pills

political suicide

economic muscle

Obviously here as well, we canÕt say these are the intersection of intellectual things and sleeping pills, economic things and muscle.

Instead, we are challenged to find the metaphoric economic COUNTERPART of muscle, or the metaphoric intellectual counterpart of sleeping pills (the original use applied to sermons!).

So among the things we need to do, JUST to understand English ADJ-NOUN pairings, are:

(1) Find an appropriate ACTIVE ZONE of the NOUNÕs referent, to which ADJ can apply. This may involve physical part-whole metonymy or broader frame metonymy. In either case, we are instructed to look for active zones RELEVANT in the current interactional scenario.

(2) Simply to understand word meanings like safe or fake, bring up multiple relevant world scenarios such as one wherein the valued object suffers damage or loss, and another where it is preserved from such damage or loss - and multiple world-construals such as the dupeÕs belief space and the con manÕs belief space.

(3) Know that the relevant ACTIVE ZONES may be counterparts in (e.g.) METAPHORIC relationships.

Previously established cultural models will be accessed as needed to supply counterpart relations in such cases.

Economic muscle might not make sense to someone

living in a very different economic environment

from ours.

Likely candidate

A candidate who is likely to win the election.

A person who is likely to declare candidacy for the election.

A candidate in the election who is likely to be available for appearance on our news show.

Likely textbook

A textbook which is likely to be written/published.

A textbook which is likely to be adopted for courses/for this course.

A textbook which is likely to contain the answer to my current

question.

A textbook which is likely to hold up the projector at the right

angle for use in this lecture.

Likely appears to refer to an assessment of the odds on some scenario matching a reality (often a future reality).

That means I have to figure out what scenario it is that IÕm trying to match with some reality.

Some nouns REFER to things which have obvious scenarios connected to them, as active zones.

Likely story

Likely scenario

But, as the differing readings of likely textbook and likely candidate show, we are not RESTRICTED to interpretations which are CONVENTIONALLY part of the frames set up by the nouns. CONTEXTUAL frames may be relevant too, as the referent of likely.

The problem, then, is that we canÕt seem to put linguistic meaning in a modular box.

We have a choice. We COULD say that English ADJ-NOUN constructions are semantically compositional, but that MOST of them, even red ball, are simply bizarre and their interpretation is part of pragmatics rather than semantics.

Or we could say that ALL of them are regularly compositional; that is, IÕm regularly putting the meaning of the ADJ and the meaning of the NOUN together in one regular way to get the meaning of the whole ADJ NOUN phrase. - BUT I then have to say that the meanings I am

Òputting togetherÓ are not all present in some context-free modular representation of Language.

A context-free dictionary and grammar (how to put ADJ and NOUN together) could not help me to interpret pairings which are easy for English speakers -

economic muscle

red ball = red-uniformed teamÕs ball

likely candidate (various readings)

JUST to do Adj-Noun comprehension, I need to be able to refer to contextual affordances which create new possible active zones for reference. And those affordances may be imagined mental spaces - beliefs that something is a gun or isnÕt a gun, beliefs that a textbook will or wonÕt

contain desired information, etc, etc.

I DO NOT have to put all these things into words every time.

So, returning to the big picture:

To get a compositional syntax, I canÕt have a modular view of language which separates linguistic meaning from other kinds of meaning.

Linguistic meanings are shaped by real human categories, how humans conceptualize the world. And linguistic meanings are interpreted in constant interplay with the interactional context - physical, social, cognitive, cultural, gesturalÉ

ItÕs ecology all the way down. Linguists can only create models that meet some of their criteria (notably compositionality) by sacrificing other less realistic criteria (e.g., modularity).