The Syntagmatic Rules

Syntagmatic responses are found much less often than paradigmatic responses in word associations; and they are more difficult to characterize in rules. But there are two related rules that appear to account for the bulk of the syntagmatic responses.

The selectional feature realization rule. The list of features for a word often contains selectional features that partially characterize the meaning of thepotential contest of that word. The adjective young, for example, has selectional restrictions on the nouns it can modify, as specified in the feature [+det[+animate]be–]. Many responses to young are merely specific realizations of this feature — e. g. boy, child, girl, man and people. To produce these responses, the respondent took the partial feature list [+noun, + animate], filled it out with other features, and gave the result; the features added were often other features of young, since some responses were words with the feature [–adult] — boy, girl, and child. The rule that accounts for these responses might be started as follows: “Take the features specified by a selectional feature, adding as many features as necessary for a surface realization; in addition, restrict yourself to the significant part of the selectional feature, the portion specifying a lexical word”.

The selectional feature realization rule accounts for the differences in the number of syntagmatic responses people give to nouns, verbs, adjectives, and so on. Notice that in the theory put forward by Chomsky nouns have no selectional features although verbs, adjectives, and other categories do. So nouns should elicit relatively few syntagmatic responses in comparison to other categories. As confirmation, we see that in Deese’s large sample of stimuli and responses nouns produced only 21 percent syntagmatic responses, while verbs produced 48 per cent, adjectives 50 per cent, and adverbs 73 per cent. Several further predictions of this sort can be verified in data from Fillenbaum and Jones. First, the selectional features of adjectives specify the nouns they modify. So adjectives should elicit nouns most often, and they do, with nouns accounting for 80 per cent of syntagmatic responses. Similarly, the selectional features for verbs specify the subjects and objects that govern the verb. So nominals (nouns and pronouns) should occur most often here also, and they do. Verbs likewise select for the particles and prepositions that occur with them, as in get along, seem like, and try out: these responses also occur quite often. Prepositions select for their objects, so prepositions should elicit nominals most often as their syntagmatic responses. This is also confirmed. It is within prepositions that the dominance of the previous minimal-contrast rule over the selectional feature realization rule is best illustrated. Some prepositions have a common antonym, hence the minimal-contrast rule can be successfully applied to them. These prepositions tend to elicit far fewer nominals than other prepositions. This implies that the selectional feature realization rule is usually applied only after certain other rules failed.

The idiom-completion rule. The stimulus cottage often elicits cheese. Likewise, whistle elicits stop; white, house; stove, pipe; justice, peace; how, now; so, what and so on. The rule that generates these responses is a close cousin of the selectional feature realization rule, for it seeks out a selectional feature that has only one realization. The rule might be started: “Find an idiom of which the stimulus is a part and produce the next main word”. Without better semantic specification of idioms, this rule will have to stand as it is.

This rule might also be appealed to explain many apparently paradigmatic responses. Ham elicits eggs, bread elicits butter, and needle elicits thread, probably not so much because the responses are paradigmatic, but because they are completions of common idioms. But carried too far, this reasoning might be used to explain the associations here — there, high — low, now — then, man — woman, etc., in exactly the same way. Instead, the phrase here and there appears to be common just because here and there are simple contrasts, and it is the latter fact that explains their frequent occurrence in word associations. This interpretation is further supported by Marshall’s observation that, for example, low elicits high more often than high elicits low, in spite of the fact that the normal order of the two words is high and low. The rule usually applied in these cases is therefore the minimal-contrast rule, not the idiom-completion rule.

Syntagmatic responses are influenced in important ways by the normal left-to-right production of sentences. Cottage often elicits cheese, but cheese rarely elicits cottage, and the same is true of other idioms. The idiom-completion rule therefore works left to right, not right to left. Also consider-adjective stimuli with the selectional feature [+det[+abstract]be–]. If the selectional feature realization rule is to add as few extra features as possible to [+noun phrase, –abstract], it should often produce pronouns. But it does not. Almost all nominal responses to adjectives are full nouns. Nouns would be produced, of course, if the adjectives were taken to be in their normal pre-nominal position, where pronouns are impossible. So here again, normal left-to-right order dictates to some extent the form responses will take. Also, in the case of transitive verbs, their features restrict subjects and objects both, yet responses to transitive verbs tend to be objects rather than subjects; furthermore, the objects of transitive verbs can be produced more quickly than their subjects in restricted word-association tasks.

Nevertheless, syntagmatic associations are not merely continuous fragments of normal speech, as writers such as Saporta have assumed, but rather responses that bear only an abstract relationship to normal speech. First, note that many stimulus-response pairs would never be found in normal speech — e.g., about — house, bread — butter, on — table, etc. — for there is a missing function word in between stimulus and response. People prefer to give lexical rather than function words as responses. And the distance between syntagmatic associations and speech is also demonstrated in a comparison of (a) the nominal responses to prepositions with (b) the objects of the same prepositions in sentences people had composed. Whereas the nominal responses are pronouns 48 per cent of the time, the objects are pronouns only 4 per cent of the time. In word association, the rule that produces a realization for [+– [–abstract]], for example, does so by adding as few other features as possible, so the responses are often semantically empty pronouns, like it, them, him, her, etc. In full utterances, on the other hand, speakers tend to fill in the feature list, producing nouns as objects. So the nominal responses and the trueobjects of a preposition reflect the same selectional restrictions, but the selectional feature realization rule, when applied with time limitations, is more likely to produce simpler realizations in the form of pronouns.

Thus, although syntagmatic responses first appearto be different from paradigmatic responses, they are produced by rules that belong to the same class of rules stated for paradigmatic responses — the simplicity of production rule. To repeat that rule, “Perform the least change on the lowest feature, with the restriction that the result must correspond to an English word”. To include syntagmatic responses, we must consider the operation of isolating selectional features and filling out their feature list to be a possible “least change”, an operation more difficult, however, than changing feature signs or deleting or adding features. The various expansions of this rule are obvious. [...]

In this brief account of the word-association game, I have tried to show that any successful explanation of word associations must be formulated in terms of syntactic and semantic features. In such a theory, the explanation will consist of rules that operate on features of a stimulus to produce features of an utterable response. Examination of the data now available suggests what several of these rules must be, but further work waits on more extensive studies of the semantic features in the lexicon. Because of the limited scope of this review, I have had to omit discussion of many veryimportant studies — e. g. Deese’s work, which shows the extent of very subtle semantic information in word associations; these studies often contain rich and orderly data but have no ready explanations. Since the word-association game is so easy to play, we know plenty about the scores. We now need to find more about the rules.