Journal o f Psycholinguistic Research, Vol. 2, No. 3, 1973

Subjective Phrase Structure: An Empirical Investigation 1 Z. S. Bond 2 and John Gray 2

Received July 21, 1972

The present study represents an attempt to investigate the basis of subjective groupings of words in sentences, employing Edwin Martin's subjective phrase structure technique. Results indicate that subjects base their judgments not only on grammatical structure but also on constituent length and on stress and intonation patterns.

INTRODUCTION

Dissatisfaction with the restricted data base of many theoretical syntacticians has led us into a study directed toward eliciting subjective judgments concerning sentences. The scope of such a study is wide and much work remains to be done. However, early results are interesting enough to justify the presentation of a preliminary report, together with some tentative conclusions. Our study has its origins in a paper by Edwin Martin (Martin, 1970). Martin wanted to "discover empirically whether or not our time-honored, widely accepted views on phrase structure are in fact representative of how language users themselves organize sentences [p. 153] ." In order to achieve this aim, he suggested a technique which would produce a "subjective phrase structure tree," on the basis of the pooled judgments of a group of people. 1A shorter version of this paper was presented to the Annual Meeting of the Canadian Linguistic Association at McGill University, Montreal, May 1972. 2Department of Linguistics, University of Alberta, Edmonton, Alberta, Canada. 259 ~) 1 9 7 3 P l e n u m P u b l i s h i n g C o r p o r a t i o n , 2 2 7 West 1 7 t h Street, N e w Y o r k , N.Y. 1 0 0 1 1 .

260

Bond and Gray

The technique is quite simple. Subjects were presented with a written sentence and asked to indicate which words of the sentence form natural groups. By counting how many subjects group any two words together, it is possible to arrive at a distance measure showing how closely related those words are judged to be. This distance measure, can, in turn, be represented in a "subjective phrase structure tree." The technique proposed by E. Martin is attractive, but it can be used profitably only with a clear understanding of the basis on which subjects produce groupings. One of Martin's hypotheses is that "subjective hierarchies are invariant over people [p. 157] ." Accordingly, different subjects will give consistent groupings, although they will vary in fineness of analysis. The present study was undertaken to investigate some of the factors that influence the subjects' groupings of words within sentences. Particularly, we were concerned with the effect of the constituent length and the stability of subjects' judgments of the same constituent in different contexts.

METHOD

Sixty-four different sentences were presented, one per page in a booklet, to a group of 35 subjects, who were students in an introductory linguistics course. The sentences were assembled from a list of subject and object constituents, varying in length from one to six words. (In one case, a constituent erroneously consisted of 7 words.) The verb was always a one-word transitive verb in the past tense. The list from which we drew the constituents, and the frequency of use of each constituent, is given in Table I. The sentence varied in length from three words to thirteen, and each constituent appeared in combination with a variety of other constituents. The subjects were tested prior to any discussion of syntactic theory. We used the same order of presentation for all subjects, so that all subjects working on a given sentence would have similar prior experience with the task. No training session was given and, in consequence, early judgments were treated with reservation. The instructions to the subjects were identical with Martin's instructions. These instructions are very broad. Their latitude gives each subject the opportunity to establish his personal style of response. A more constraining set of instructions would have reduced the variation in response style and reduced the usefulness of resulting trees. The data were processed following the procedure outlined in E. Martin (1970). The subjective grouping of a sentence was coded and submitted to a Fortran computer program which carried out the steps of Martin's method and produced the tree diagram. Each subject/sentence judgment was coded as a vector of numbers. The vector had as many elements as there were words in the sentence being coded. Basically the vector consisted of t's and 2's. When a

Subjective Phrase Structure: An Empirical Investigation Table I.

261

Subject, Verb, and Object Constituents a

Subject constituent Mary (8) the man (8) the happy children (8) the young playful puppies (8) the old and feeble pony (8) the girl who was pretty (8) the old frost-bitten naval officer (8) the farmer who ploughed the fields (8)

Verb noticed preferred (5) saw (7) liked (34) loved (10)

Object constituent Rover (3) yellow canaries (3) the banana peels (3) the large green socks (3) the indestructible steel nails (3) the mules which brayed loudly (3) many tantalizing steaming hot buttered rums (3) a breakfast which fills the stomach (3) Bill (5) fudge cookies (5) the friendly woman (5) the purple spotted dragon (5) the parsnip and carrot stew (5) the woman who was tired (5) the tall and droopy weeping willows (5) children who were happy and gay (4) the children who were happy and gay (1)

aThe number of occurrences of each item is given in parentheses. word was not in the same group as the preceding word, the number was changed. Numbers higher than 2 were used to signal discontinuous groups. For ease o f interpretation, word order was adjusted in tree diagrams representing sentences which gave rise to an analysis producing discontinuous constituents. Reference to Fig. 1 will indicate the form in which the results were expressed. The scale at the side o f each tree gives an indication o f the strength of tile opinion relating to a particular cluster. It does so by indicating the number of subjects n o t agreeing to a particular clustering. In the case illustrated in Fig. 1, words 1 and 5 were grouped compactly together by all but four subjects; on the other hand, the subject and verb cluster reflects the opinion o f only ten subjects (that is, the N of 3 4 - o n e had to be rejected-less the number of 24, as indicated on the scale). Two trees for each sentence were produced as a check on the data. Frequently, all the words within a cluster are not equidistant from a given word outside the cluster. When such a discrepancy is found, a minimum tree is constructed by making a consistent choice o f the minimum distance values; a maximum tree is obtained by consistently selecting the maximum distance values and using t h e m in the construction o f the tree. The two trees given in Figs. 1 and 2 are not typical. They have been chosen specifically to illustrate the result o f employing the two methods and give some idea of the kinds o f

262

Bond and Gray THE

I

PONY

5

OLD

2

AND

3 I-'--'--

~-- -

FEEBLE 4 LIKED

6

THE

7

DRAGON 10 PURPLE 8 SPOTTED 9

o

s

Io

is

20

zs

Fig. 1. Phrase-structure tree of a sentence produced by the minimum method. differences that may result. It will be seen that there is a difference in distance as well as a change in clustering. Under the minimum method (Fig. 1), the verb clusters with the subject constituent, while with the maximum method (Fig. 2), the verb clusters with the object constituent. Such qualitative changes in the distance between constituents in a sentence are usual in the trees produced by the two methods.

RESULTS AND DISCUSSION The first question to be considered is whether there are consistent groupings or whether the clustering is purely random. Our results indicate that there is a considerable degree of consistency to be found. Certain constituents appear in a regular form independently of context. In addition, our results are consistent with those reported in E. Martin (1970) for sentences of comparable grammatical pattern. An especially striking example of a consistently analyzed constituent is presented in Fig. 3. This constituent was clustered in the manner indicated in each of its five occurences. In the light of the foregoing, we are assuming that a constituent of a particular grammatical form will have a tendency toward a particular type of clustering. However, the clustering of a particular word group will be influenced by a number of factors of which internal grammatical structure is but one.

263

Subjective Phrase Structure: An Empirical Investigation THE

I

PONY

5

OLD

Z

AND

3

FEEBLE

4

LIKED

6

THE

7

,

- -

DRAGONI0 PURPLE

} ' ~

8

SPOTTED 9

--] o

s

1o

Ts

20

z~

3a

Fig. 2. Phrase-structure tree of a sentence produced by the maximum method. Our findings show that there is a considerable effect associated with the length of the constituents. Whether the verb clusters with the subject constituent or with the object constituent seems to depend to a considerable degree on the size of the subject constituent. In three-quarters of the cases involving one-word subjects, the Verb was grouped with the subject. By and large, the length of the object constituent seemed irrelevant to the grouping of the verb; only when the object was six words long, was there a tendency to group the verb with the subject. The impression is, therefore, that subjects analyze the sentences from left to right, and make the major break not only on the basis of grammatical structure but also take into account constituent length. Figure 4 shows the extent to which the verb is clustered with the subject or object constituent depending on the length of the subject constituent. The details of the clustering given to specific constituents are suggestive and interesting. For example, in constituents with the conjunction and, there is a tendency to group the conjunction with the following word in 18 out of 23 occurrences of such constituents; prenominal adjectives were grouped with a cluster consisting of article-noun rather than directly with the noun. It is difficult, however, to arrive at a reasonable estimate of the significance of such tendencies since they generally reflect the consistent judgment of only a few subjects. We can, however, suggest several factors which influence the grouping of constituents.

264

Bond and Gray

THE

1

WOMI~,N

3

FRIENDLY

2

Fig. 3. A stable constituent (minimum method). All occurrences of this constituent received an identical analysis, with the determiner and noun forming a closer group than the noun and adjective. The maximum method produced a comparable configuration.

1. A long constituent is less "stable" than a short one in that it is more prone to variations within its internal structure in different contexts. This is especially true of long prenominal modifiers. An example of a "stable" constituent is provided in Fig. 4 and of an "unstable" constituent in Fig. 5. 2. The placement of the major constituent boundary either before or after the verb seems to correlate with the domain of a phonological phrase demarcated by intonation and stress. Thus, the major constituent break will

Verb clusters with subject constituent

verb clusters with object oonstituent

i00

90

80

70

//A

60

//A //A ///1

50

//

//',4 ~ -

8

//11

//A m~

//.

A

///

40

//A //A

30

//

20

///1 //A //A

/ . .// ///

///I //A // A ///I

~/1

i0

A

oY/A

,'''2 / / / // / /// ///

2

3

4

5

Number of words in subject constituent

Fig. 4. Clustering of the verb with the subject or object constituent.

Subjective Phrase Structure: An Empirical Investigation

265

THE

1

THE

l

YOUNG

2

YOUNG

2 --

PLAYFUL 3

PLAYFUL 3

PUPPIES 4

PUPPIES 4

THE

I

THE

|

YOUNG

2

PUPPIES 4

1 I

--~-~

|

- - J

1

-

!

-

PLAYFUL 3

YOUNG

2:

PUPPIES 4

PLAYFUL

3~ ]

J~ | L

/

Fig. 5. Four different analyses of an unstable constituent (minimum method). The maximum method gives comparable configurations. tend to be placed where a pause might be placed in ordinary speech, between phrases spoken with one intonation contour. In connection with the last suggestion, we should like to mention a particularly interesting case. In one sentence, a six-word constituent was inadvertantly typed with an additional determiner and the error was not detected. Thus, we were able to compare the seven occurrences of the regular constituent with the single occurrence of the erroneous one. The regular object constituent was: children who are happy and gay. The erroneous constituent was: the children who are happy and gay. In the regular version, we consistently found the cluster "children who." In the other version of the constituent, "the children" appeared as a basic cluster, and the relative pronoun was the first word of the next basic cluster. I f the effects of constituent length are considered, our findings are consistent with E. Martin's. In one of his sentence sets, with a one-word subject constituent, the verb clustered with the subject. In another set, with subject constituents containing a relative clause, the verb clustered with the

266

Bond and Gray

object constituent. E. Martin speculated "that the [S(VO)] organization of the second set of sentences results from the presence of the relative clause [p. 162] ." Similar results have also been obtained by a rather different technique, simply asking subjects where junctures (pauses) should be placed in written sentences (J. Martin et aI., 1971). They found that the division of the utterance into phrases was dependent upon the length of the constituents. Long subject phrases produced juncture between the subject and the verb, whereas short subject phrases produced juncture between the verb and the object. In summary, we can say that our work has led us to the following tentative conclusions: Martin's technique is an interesting and potentially useful one in the exploration of production/perception phenomena. However, it seems to be quite irrelevant to any considerations concerning the psychological reality of the linguist's phrase marker.

REFERENCES

Martin, E. (1970). Towards an analysis of subjective phrase structure. Psychol. Bull. 74, 153-166. Martin, J., Kolodziej, B., and Genay, J. (1971). Segmentation of sentences into phonological phrases as a function of constituent length. Jr. Verb. Learn. Verb. Behav. 10, 226-233.

Subjective phrase structure: An empirical investigation.

The present study represents an attempt to investigate the basis of subjective groupings of words in sentences, employing Edwin Martin's subjective ph...
340KB Sizes 0 Downloads 0 Views