relate words to other words and events in an infinite regress.
Let us take an example in order to view the process at close range. Suppose I am asked to code the word PROGRESS in terms of E, P, and A. A paraphrase of the dictionary meaning of "progress" goes something like this: "moving forward or onward; a development; an advance toward perfection." The assignment of the coding can be described introspectively in terms resembling the following account: "Now let's see. Progress means moving ahead, not backwards. It'a a change for the better ("toward perfection"). Therefore it must be active and good. It'a also strong because it's a development and has the source of power within it, like a seed that develops." Thus PROGRESS would be coded E+P+A+ - which is indeed what an empirical check yields (see Footnote 13).
This example is, of course, merely an Illustration and I do not propose it as a serious claim of what really goes on inside our head when we rate a word on these kinds of scales. A formal account will have to be more specific about the choice of the features that are abstracted. For example, it's perfectly possible to imagine a situation where there is slow progress or no progress where there ought to be ("I am progressing too slowly") or where the change was for the worse ("It looked as if I was making a lot of progress, but it turned out I was really heading in the wrong direction"). It may be that when a word is presented in isolation or when the context fails to specify all the relevant elements then some principle of prior usage or saliency of features helps determine probabilistically those features that are most likely to be abstracted.
Consider the following examples:
(a) Juliet is the Sun (Shakespeare)
(b) She'll make someone a nice husband (C. E. Osgood)
(c) He is a mother to me.
I take it that none of these utterances are anomolous, and indeed, they seem to me to convey very effectively the intended conceptual event. The question I'd like to raise is how do we know which feature of the concepts "Sun", "husband", and "mother"
is to be abstracted in order to recover the intended conceptual event? Note that
in all three examples, a literal interpretation would be anomolous: Sun is not animate, husband should be male, and mother should be female. One could say that the examples "violate" semantic projection rules, and their effectiveness is precisely
because the rules are violated. I would prefer not to think about it in terms of "violation of a rule" but rather to make provisions within the projection rules to allow for such combinations (e.g. "mother" means (i) female biological parent and (ii) a person, uncoded for sex, that acts like a mother. Note that "He is a father to me" allows for an equivalent relation but means something quite different than sentence (c)).
Essentially, all three examples represent instances of metaphor by analogy. In (a) the speaker intends to say that Juliet is bright and glowing, in (b) that "She" is masculine and dominant, and in (c) that "He" takes care of "me" with loving kindness as lf "I" were a helpless child. What enables the listener to abstract in (a) "bright" and "glowing" rather than "round", "hot" or "yellow", in (b) masculine" and "dominant" rather than "provider" or "companion", and in (c) "loving care" rather than "legal guardian"? I submit that the answer lies in a principle of psychological saliency of semantic features in interaction with the usual Implicit and implicative inferences about the conceptual event represented by the utterance (see Sections 1.2. and 1.3.). Thus in (a) "brightness" is a salient feature of the sun and logical inference eliminates "roundness"; the implicative meaning of the utterance as given by the situational context (Section 1.2.5.) specifies the intent as a private tact (feeling: see Section 1.3.3.4.1.) and the implicit meaning specifies E+ affect (1.2.2.), making the choice of "bright" appropriate since it is coded as E+. In example (b), the salient features of "husband" are "masculine" and "dominant" by virtue of its opposition to "wife" ("feminine", "submissive"); the implicative meaning specifies the intent as an autoclitic of irony (see Section 1.3.3.7.2.10.), and the implicit meaning specifies P+ affect making masculine" and "dominant" (both P+) appropriate choices. (There is no need to go through the argument with (c).)
I think it can easily be shown that unless the context or situation supplies the relevant implicative cues, then utterances of the type just considered cannot be properly interpreted or, to put it another way, given utterances (a) to (c) in isolation, as was done here, their meaning can be given an interpretation only if one makes the kind of implicative inferences suggested.
The notion of saliency of semantic features needs to be related to a general theory of perception and concept formation. Saliency is undoubtedly a relative
concept: it has the property of oscillation. For example, religious membership may be a salient trait in one context (e.g. dating) but not in another (e.g, choosing a roommate). The saliency of food objects increases during hunger, decreases at other
times. Saliency of semantic features may oscillate in relation to situational and motivational factors of a similar kind but in addition, the oscillation process will be highly sensitive to intraverbal relations. Consider the following three
utterances:
(a) He is the life of the party.
(b) He is the life of the company.
(c) He is the life of the group.
"Life" has affective features of E+, P+ and A+. In (a), its relation to "party" draws out the A+ feature of "life" and makes it salient. In (b), "company" draws out the B+ and P+ features of "life" and makes them salient. In (c), "group" is coded for both P+ and A+ and is ambiguous about E and none of those can be said to be more salient than the others making for a certain ambiguity in (c): is he the most important member? is he the most powerful member? is he the most active member? Note however that (d) and (e) resolve the ambiguity:
(d) He is the life of the social group.
(e) He is the life of the fund raising group.
"Social group" draws out the A+ feature of "life" making it salient "fund raising group" draws out the B+ feature ("important") making it salient.
3. Selection factors in the act of composing.
In this section I would like to discuss some of the factors that must be considered if we are to understand the process by which a speaker is able to engage in communicative behavior. Communication can take place in the absence of linguistic competence, as we know from studies of lower animals and from investigations on paralinguistic phenomena and kinesics; but I am concerned here with communication via the medium of language, and therefore I will assume linguistic competence as having been already acquired. l4
3.1. Intention: This provides the motivation for engaging in communicative behavior.
The type of intention involved in any particular communicative act affects the surface form of the utterance. For example, mands (see Section 1.3.3.1.) will tend to be short and direct as compared to public tacts (see Section 1.3.3.5.); self-tacts (1.3.3.6.2.); autistic utterences generally (1.3.3.6.) will often skip surface phonological actualizatlon and be less affected by some intraverbals
(notably, grammaticalness, see Section 1.3.3.3.1.) and more by others (e.g. word associations, see 1.3.3.3.2.) while style factors relating to effectiveness (2.3.) will be less important since there is no concern here with message fidelity.
A serious study of how intention affects the selection of utterances must concern itself with the mental nature of "selection." The latter implies conscious effort or volition, and while volitional factors cannot be left out of the account, it is not necessary to assume that all selection processes are volitionally controlled or even that the speaker must be aware of their operation. Certainly it is the case that some implicative aspects of utterances reveal intentions rather than communicate them directly, often despite the efforts on the part of the speaker to hide them. There is a distinction here which approximates closely that discussed by Goffman (1959) between impressions that are "given" and those that are "given off." When a speaker intends to control the action of a listener and attempts to do so by uttering a mand, he may select the advice using the surface form appropriate for auch a mand (1.3.3.1.5.). He thereby makes a public claim that he has the listener's interest at heart. This is the "given" intention. However, the context (1.2.4.) or the situation (1.2.5.) may provide clues that contradict this claim and the intention that is "given off" may be as revealing as it is involuntary. The girl who on the night of the first encounter eagerly says "Yes" to her date's request of "would you like to see the painting in my bedroom?" is playing a game that involves manipulating the intention that is "given off," and her assent has nothing to do whatsoever with the aesthetics of works of art. (See also Footnote 9).
3.2. Audience: The nature of the audience has both direct and indirect selectional effects upon utterances. The indirect effect is through the prior selection of intention. For example, when the listener is of higher status or power than the speaker a command or advice is an inappropriate intention to "give" and may be either inhibited (that is, it is not encoded) or relegated to the status of an intention "given off" either inadvertently or skillfully through a disguise (e.g. "Hey, Boss! Did you know the boys bought a case load of Scope today?'')l5
Direct effects of the audience upon selection of utterances include dialect, tonality, style, rate and amount of speech output, stress pattern, and the like. For example, when adults talk to infants and pets they restrict their verbal repertoire to a few characteristic utterances that are then repeated over and over again with
variations in phonetic stress pattern (e.g. "Aren't you cute! Yes. So cute. Yes. Aren't you cute. Yes. Isn't that right! Isn't that right!" etc.). The analysis of this kind of output could be quite complex. There is no intention on the part of the speaker to communicate to the infant (or pet) some message that is carried by the linguistic output, yet the linguistic meaning of the utterance is not irrelevant. One could for example imagine someone who steps out of tbe house with his pet saying: "Oh, it's raining, we can't go. You poor little thing, we can't go. Oh, you poor little thing, we can't go!" etc. I think utterances addressed to infants and pets are basically autistic (see Section 1.3.3.6.) but their linguistic form is learned in the manner of conventions (1.3.3.3.3.).
Skinner refers to "positive" and "negative" audiences depending on whether the listener is a source of reward or punishment. Nodding of the head, "sparkling" of the eyes, a physical posture of attentiveness, verbal retorts such as "Mmm! Hmmm!", "Yeah," "I'm with you" etc., are characteristic behaviors of positive audiences that encourage verbal output. A stern face, distracted looks, silence, are charactaristics of negative audiences that discourage verbal output. Under some conditions verbal output is maintained in the absence of audience encouragement or feedback (e.g., when the situation defines the nature of the speaker-listener relation as in lectures, public speeches, radio programs, or when the utterance is autistic such as the speech of hebephrenics).
3.3. Situation: Conventions (see Section 1.3.3.3.3.) are defined by the situation which may also have inhibitory effects on speech output (cf. the "No talking" rule which prevails in libraries, churches, ceremonies, etc.). The situation may also represent potential audiences (e.g. state secrets are not discussed in U.S. embassies except in special rooms reserved for this purpose) or double audiences (e.g. a repartee on stage that has one meaning for the addressee and another meaning for the theater audience).
3.4. Style: As discussed in an earlier section (2.) style deals with selection rules on the basis of a number of relevant criteria. In this section I would like to give a few additional specifications about internal factors that affect choices in composition.
3.4.1. Intraverbal Influences: Skinner cites the following example from T. S. Eliot " . . . what will the aplder do, Suspend its operations, . . ." in which the choice of "suspend" is related to the meaning of its possible variants such as "cease" and "desist in" but at the same time it has a relation to "spider" which the other variants do not. In the same poem, Eliot writes "The tiger springs in the new year" where "springs" relates both to the meaning of the variant "pounces" (or "jumps") and, intraverbally, to "tiger" (note the possible variation of "The tiger leaps in the new year" which maintains an analogous relation).
Instances of "code switching" represent violations of intraverbal relations (e.g. "Hi, Professor Jakobovits," with which an undergraduate once greeted me, contains elements from two different codes). Bilingual interference may be considered as a special case of code switching.
3.4.2. Formal Constraints: Speakers will commonly begin an utterance without prior planning of the sentence as a whole and then find under the pressure of grammatica1 constraint that they must complete the sentence in a given way which is ineffective, inappropriate, or lacks fidelity (followed sometimes by an attempt at correction using an autoclitic of negation: see 1.3.3.7.2.2.).
Selection of a word in a syntactic frame is restricted by formal constraints of the given frame. The task known as the "Cloze" procedure requires the speaker to supply words that have been systematically deleted from the text. "Pattern practice" commonly used today in college foreign language courses restrict the speaker's contribution to the utterance to substitution of lexical items within a given syntactic frame.
In poetry there are many instances of formal constraints which dictate the selection of words in a line (sound quality, syllabic pattern, length, etc.).
3.4.3. Blendings: Certain utterances (especially of the expression type: see 1.3.3.4.) appear to be made up of a combination of two separate utterances: e.g. "That's what I think so" ("That's what I think" + "I think so"), "You're probably true" ("You're probably right" + "It's probably true"),
"For that matter of fact" ("For that matter" + "Az a matter of fact").
Skinner, who citez these examples (1957, p. 297) calls them "phrase-blends." Deliberate use of blending and distortion is a device used for punning and
witting (sic). Word-blending allows for the coinage of new words:
"intrinsicate" (intrinsic + intricate) and "wind'ring" (winding + wandering) are from Shakespeare and I have once heard a teenager assert that "fantabulous" (fantastic + fabulous) is a "real" word (perhaps it ought to be!).
4. The structure of the message.
In this sectlon I shall concern myself with two types of factors related to the structure of what it is that's being said: the syntactic characteristics of the utterances that make up the message and the conceptual events that precede the act of composition (encoding) including the derived or inferred aspect of these as available to the listener (meaning). The discussion will take us into an examination of the nature of thought and language, their development and interrelation.
4.1. Syntactic Characteristics: Personal observation of children as well as controlled experimental studies (see for example McCarthy, 1954) have established the fact that as children get older the also of their vocabulary (both active and passive) and the length of the sentences they produce get progressively longer. What is more revealing, however, is the study of the specific changes in the structure of children's utterances since there are many ways by which sentences can get longer and these alternatives are by no means equivalent. Thus the lengthening of an utterance by concatenation of simple main clausea using the conjunction "and" would seem to be a lesser degree of linguistic maturity than the construction of a main clause with subordinate clauses and this in turn represents a lower degree of complexity than the embedding of nominal and adverbial constructions within main clauses. A review of a recent study by Hunt (1965) will enable us to examine the above contention in greater detail.
Hunt compared the syntactic structure of written materials produced by children in Grades 4, 8, and 12 to each other and to the writings of "superior adults" in guality magazines with a national circulation. One of the most interesting findings he reports concerns the length of the "T-unlt." This refers to the "minimal terminable unit" and is defined as the shortest sentence within a longer utterance. An example will clarify the concept. Consider the following hypothetical account as it might have been written by a ten-year old:
"The man was tall and funny-looking he wore a red coat and
he tried to catch the big beach ball it had red and white
stripes the man fell over backwards."
By adult standards fourth graders are lax with punctuation marks and the above "sentence" can be broken up into several T-units each of which is composed of a main clause plus whatever subordinate clauses are attached to it or embedded within
it. Thus:
1. The man was tall and funny-looking.
2. He wore a red coat.
3. He tried to catch the big beach ball.
4. It had red and white stripes.
5. The man fell over backwards.
The average length of the T-unit in this sample is 6.2 words. Actually, Hunt found the average for children in Grade 4 to be 8.6 words. This increases steadily so that in Grade 8 it is 11.5 words, in Grade 12, 14.4 words, and for the superior adults it is 20.3 words. Another way of looking at this is to compute the percentage of utterances that are made up of "short T-units" defined as 8 words or less: these percentages turn out to be 43%, 21%; 10%, and 6% respectively for the four samples it is significant that the greater length of T-units for superior adults
is not achieved by increasing the number of subordinate clauses; instead, adults write longer clauses. In fact, the most distinctive difference between the writing of superior adults and that of twelfth graders is the clause length, on the average
36% more words per clause. Thia accounts for three quarters of the difference between superior adults and twelfth graders in sentence length and this gap is larger than the difference between twelfth graders on the verge of graduation sud
fourth graders who are just beginning to develop writing skills.
In order to appreciate the significance of the findings just discussed, it is necesaary to examine the syntactic devices involved in-the conaolidation of separate shorter clauses into longer ones. In another recent report, O'Donnell, Griffin,
and Norris (1967) present a transformational analysis of the syntax of kindergarten and elementary school children for samples of utterances in both oral and written codes of expression. These authors describe the lengthening of T-units and conaolidat$on of clauses in terms of "sentence-combining transformations." This may be conceived of as the embedding of a sentence within another by means of deletion, substitution, and expansion transformation ruleas. O'Donnell and his
associates observed significant overall increases in the incidence of three major types of constructions formed by consolidation of clauses in speech from kindergarten through Grade 7 and in writing from Grade 3 through Grade 7. These were nominals, adverbials, and coordinations within T-units. Let us look at the previous sample utterance given above and consider some possible sentence-combining transformations. The five T-units can be consolidated into three by using a relative adjective clause in two places:
1. The man who was tall and funny-looking wore a red coat.
2. He tried to catch the big beach ball that had red and white stripes.
3. The man fell over backwards.
Now we have three T-units (instead of five) with an average length of 10.3 words
(instead of 6.2). This is still below the average for eighth graders, so let us do some more consolidation using an adverbial construction:
1. The tall funny-looking man in the red coat tried to catch the big beach ball with the red and white stripes.
2. The man fell over backwards. Now we have two T-units (instead of three) with an average length of 15.5 words (instead of 10.3) which is just slightly above the level of twelfth graders. One additional adverbial sentence-combining construction transforms the utterance into a single long T-unit:
The tall funny-looking man in the red coat fell over backwards as he tried to catch the big beach ball with the red and white stripes.
This is now a T-unit of 31 words in length, much longer than the average for superior adults (20.3 words) and is almoat too long for effectiveness (it strikes me as a bit unwieldy).
It is quite obvious from these examples that sentence-combining transformations increase the information content of syntactic units. It is a 'unitizing' process that packs increasing bits of information into the T-unit. This is a general design feature of the type of information processing device that the human mind represents (see Miller, 1967). If we look upon the T-unit as not only a syntactic unit but alao a communicative unit, then sentence-combining transformation is in effect a recoding device for "chunking" (see Miller, 1956) information into larger and larger chunks. There is no theoretical limit to such a procedure since the grammar of language allows for infinite embeddings, but I would guess that for communicative
purposes the efficient limit for the length of a T-unit is somewhere between 20 to 30 words.
It may be instructive to determine the number of communicative elements in a T-unit by decomposing it into the maximum number of assertions it makes. The 31-word long T-unit considered above contains at least seven separate assertions:
1. The man is tall.
2. The man ia funny-looking.
3. The man wears a coat.
4. The coat is red.
5. The man fell over backwards.
6. The man tried to catch the beach ball.
7. The beach ball has red and white stripes.
The problem for a performance model of language is, as I see it, to account for the manner in which these simple aasertions are combined into syntactic unite of increasing complexity containing 1arger and longer chunks of information. It is evident that it will be necessary to concern ourselves with the conceptual nature of simple assertions and how these are compounded into more complex assertions. I shall discuss this problem directly (see next section) but first let me point out that empirical studies of a descriptive nature such as that of Hunt and of O'Donnell et.al. are likely to be very useful in that they provide clues for the development of increasingly complex cognItive operations as they manifest themselves in syntactic constructions of increasing flexibility. Take for example the finding of O'Donnell et.al. that not all types of nominaI conatructions produced by sentence-combining transformations increase in occurrence in the writings of upper grade elementary school children (Grades 5 and 7 in their sample) compared to lower grades. The nominal constructions that increase significantly include a noun modified by a genetive form (man's coat), a participle (falling leaf), an infinitive with subject (The sun made the flowers bloom), and a structure containing a gerund phrase (The dove kept him from being drowned). All these constructions involve deletion transformations (as opposed to additions or substitutions). Could it be, as O'Donnell et.al. suggest, that deletion rules represent a difficult cognitive operation? (Thus, "The dove saw that the ant was drowning" appears to be easier (and more primary, developmentally) than "The dove saw the ant drowning".)(continue here)