Главная Случайная страница


Категории:

ДомЗдоровьеЗоологияИнформатикаИскусствоИскусствоКомпьютерыКулинарияМаркетингМатематикаМедицинаМенеджментОбразованиеПедагогикаПитомцыПрограммированиеПроизводствоПромышленностьПсихологияРазноеРелигияСоциологияСпортСтатистикаТранспортФизикаФилософияФинансыХимияХоббиЭкологияЭкономикаЭлектроника






Frames and schematic activation

As we have seen, the distinction between two different levels of competence is important from the point of view of interpretative processes; this distinction iden­tifies the set of interpretations conventionally associated with linguistic forms, and which constitute the object of study of lexical semantics. I would like now to analyze in more detail how the semantic-linguistic component contributes to comprehension processes. The schematic models discussed in the previous chapter are particularly suitable for this purpose.

As we saw, individual terms are connected to a scene, or standard context, of reference. For example, the Italian verb pranzare (which translates as "to (have) lunch") activates all the elements of the very general scene, "to eat," including those that cannot be directly subcategorized by that particular predicate. In the case of pranzare, the object of the action of eating cannot be lexicalized, because pranzare is an intransitive verb. It is not therefore possible to say:

3. * Oggi ho pranzato un polio. (* Today I lunched a chicken.)

However, each time the term to lunch is introduced into discourse, it activates the whole scene of "eating," of which lunch is a sub-specification relating to a particu­lar moment in the day, contrasting in this respect with to have dinner. Because the scene is inherently transitive, one of its constituent roles being an edible entity which is the object of the action, to lunch, inasmuch as it is part of the same scene, inherits what one could define as a conceptual ttansitivity. When inserted into a sentence, it activates even those roles that cannot be directly lexicalized. It is on the basis of this conceptual transitivity that in the sequence:

4. Ho pranzato volentieri. IIpolio era ottimo. (I was quite happy to have lunch. The chicken was excellent.)

we attribute the object role in our lunch to the chicken. At the level of under standing, this is very important, because it allows us to explain links of coherency and anaphoric connections between sentences. We have no difficulty in (4) in connecting the chicken to the object role activated by the frame "eating," and this explains why chicken can be introduced by the definite article. Normally the tied-

nite article is used when referring to identifiable entities, typically entities that have already been mentioned in the discourse. However, even an entity that has not been mentioned can be referred to with the definite article if, as in our example, it has been implicitly introduced into the discourse through the activa­tion of a frame.

The relation between the meaning of an individual lexical entry and its un­derlying scene also explains why a sentence like (5) seems anomalous:

5. ? Aveva pranzato senza mangiare. (? He had had lunch without eating.)

To lunch necessarily implies "eating" in that it is part of the same scene. Cases like this are very frequent and can be explained adequately by a schematic semantic model able to give an account of the relation between the representation of lexical units and their textual insertion.

A great deal of attention has been devoted to this type of phenomenon in the field of artificial intelligence, which has traced the regularity of our comprehen­sion processes back to underlying schemata of knowledge: the structures for the representation of knowledge, for example frames, are composed of so-called slots that operate as default values, automatically activated in the absence of explicit indications to the contrary. In the by now famous example of the "restaurant" script,10 a series of slots describes the typical restaurant and its constituent ele­ments (waiters, tables, menu, etc.), together with the typical sequence of actions that occurs in restaurants (ordering, eating, paying, leaving, etc.). From a program that contains this script it is possible to infer that if someone enters a restaurant, orders, eats, and leaves, that person will probably also have paid, unless otherwise specified. Default values are activated, one might say, ceteris paribus. The concept of default values has been widely used to explain the contribution of lexical con­tent to textual comprehension.11 It is undoubtedly very useful but it needs to be outlined more clearly.

Default values and typicality

If we look more closely at the examples considered thus far, we can see that al­though in each case the term activates a frame, the properties activated are not of the same kind. It is here that the distinction between typical and essential prop­erties proves so productive. There are restaurants without menus or without wait­ers serving at the tables, but a restaurant where you could never eat would not be a restaurant. In other words, the slots of a frame, to adopt the terminology of artificial intelligence, relate to values that have different roles in the representation of a given term; some are typical values while others are essential to the definition of the meaning. We saw in the last chapter that both essential and typical prop­erties can be included in frames, as in prototypes. Now, although each term acti­vates all the properties of its frame or typical occurrence, not all of them can be

considered default values: only typical components are default values of the repre­sentation.

We must remember that in the representation formulated by artificial intelli­gence, default values are those values assumed to be valid in the absence of explicit indications to the contrary, indications which can, however, always be introduced without thereby altering the meaning. What this means is that the fundamental characteristic of default values is their erasability. The typical restaurant has menus, waiters, and tablecloths, but each of these elements is a property that can be erased, leaving simply an atypical restaurant. Green lemons, albino tigers, and white whales are all non-typical exemplars of their respective categories, but they are still respectively lemons, tigers, and whales. Typical color is a default value. In the absence of information to the contrary, we assume that lemons are yellow, tigers have stripes, and whales are dark-colored, but these properties can always be erased in that they are typical but non-essential. On the other hand, the facts that lemons are fruits, tigers are feline, and whales are mammals are not default values but properties that cannot be erased without renegotiating the meaning of the terms. Something similar occurs in complex scenes activated by ptedicates: pran-zare necessarily implies that something edible is consumed. This feature cannot be erased and is not assumed by default.

The difference between the two types of property clearly emerges if we use the test of the adversative but, as we saw in chapter 6. But can block typical properties, the subject of probable inferences, but not essential ones, without producing a semantic anomaly requiring subsequent explanation or textual elabo­ration:

6. It is a tiger but it is albino.

(I.e., it does not have stripes which, by default value, we would typically expect it to have).

7. It is a lemon but it is green.

(I.e., it is not yellow as, by default value, we would typically expect).

8. * It is a tiger but it is not an animal.

9. * I had lunch but I did not eat.

Although typical values are not homogenous and it is possible to distinguish be­tween different forms of typicality according to the various lexical categories,12 they are all characterized by their erasability—typical properties are the set of default values of a frame or prototypical scene. Differentiating between typical properties and essential properties is thetefore important from the point of view of comprehension, in that we can then specify the components of a semantic frame that are operating as default values, that is, as a probable but not certain inferential base.

 

Последнее изменение этой страницы: 2016-08-11

lectmania.ru. Все права принадлежат авторам данных материалов. В случае нарушения авторского права напишите нам сюда...