Word class and serial position in the immediate recall of sentences

Word class and serial position in the immediate recall of sentences ALEXANDER J. WEARING* t Yale University, New Haven, Conn. 06510 A divided-attention task was used to study the immediate recall of sentenees and to determine whether or not nouns were reealled better than modifiers, verbs were reealIed better than nouns, and anormal skewed U-sbaped serial position eurve eould be detected. The answer to all three questions was a qualified yes. Conventional wisdom has it that when a sentence is learned, the core meaning [which term usually involves the agent (logieal subjeet), objeet, and verb] is stored, while modifiers such as adjectives and adverbs are quiekly forgotten (Mandler & Mandler, 1964; Martin, Roberts, & Collins, 1968; Martin & Walter, 1969). Belief in the storage of the core meaning usually presumes that normal serial position effects (Crowder, 1969) are overridden and that those terms eomprising the eore meaning are all stored equally weil. However, neither the role of serial position nor the possibility that the eore meaning may not be equally strongly carried by the subject, objeet, and verb has been thoroughly investigated. Fodor, Garrett, & Bever (1968) present evidenee for a central role for the verb in sentence processing, and Gladney & Krulee (1967) found that ehanges in the verb caused more disruption in sentence recognition and recall than did changes in the subject or the object of the sentence. Healy & Miller (1970) showed that when sentences are sarted on the basis of similarity of meaning, the verb is a mueh better predictor of sarting decisions than either the subjeet or the object. On the other hand, Wearing (1969a, b, 1970) found that in both immediate (3Q-min) and 48-h recall, the subjeet and object were both reeognized more easily and were more effective recall cues than the main verb of the sentenee. Martin, Roberts, & 'Collins (1968) found that for four different kinds of sentences, at 0 and 10 sec after presentation, the verb was recalled worst on five of the eight occasions but never was recalled best. For Martin & Walter (1969), with *This study was supported by NIMH Grant MH·17706-01. The author is grateful to M. B. Pazker and Rosemary Wearing for recording the sentences and digits, to Bruce Patker for collecting the data, and to Bruce Patker and Rosernary Wearing for assistance in its analysis. tPresent address is: Department of Psycho!ogy, University of Melbourne, Patkville, Victoria 3052, Australia. 338 three kinds of sentences at 0- and 1 Q-sec reeall intervals, the verb was recalled worst on two of the six occasions and never recalIed best. A careful examination of Mandler & Mandler's (1964) figures also suggests that recall of the verb was inferior to that of the no uns, at least on the early trials. But perhaps the most notable unstated result of these last three experiments is that far from heing unimportant, the normal serial position effects are quite strongly manifested. In the Martin et al experiments, the single best predictor of recall is serial position, and in the Mandler & Mandler (1964) experiments, serial position effects are also strongly pronounced. The purpose of the present experiment was to answer the questions of whether or not form class and serial position effects can be reliably 0 bserved in sentenee processing. The most frequently used proeedure for studying this in short-term memory has been the procedure due to Peterson & Peterson (1959), but in the present case it seemed desirable to use a technique which permitted immediate recall, yielded a reasonable error rate, and, in addition, simulated normal speech proeessing eonditions to the extent that attention was only partly directed to the speaker. A procedure that meets these requirements is a divided-attention task of the kind described by Wearing & Crowder (1971) in which the listener is required to attend to same other competing input as weIl as the sentence itself. SUBJECTS The Ss were 26 undergraduate volunteers from Yale University. MATERIALS Thirty experimental sentenees were embedded in a list of 38 sentenees. All sentences were 11 words in length. All the experimental sentences had the same grammatical structure, being of the form: Preposition + Deterrniner + Adjeetive + Noun + Deterrniner + Adjeetive + Noun + Verb + Deterrniner + Adjeetive + Noun. The semantie conte nt of the sentences varied widely, and no sentences had any content words in common. The main purpose of the buffer sentences (also 11 words in Iength) were to absorb warmup effeets at the beginning of the task and to provide same variation in the course of the list. The experimental sequence was the same for all Ss. The sentences and the digits were reeorded simultaneously with a stereo tape recorder, the sentences being spoken by a male and the digits by a female. Normal speech rhythms were used, and the location of the 10 digits relative to the 11 words is shown in Fig.1. PROCEDURE The Ss, who were tested individually, were seated in front of a two-speaker stereo system and were told that they would hear a senten ce from one of the speakers at the same time as they heard a set of digits from the other and that after presentation they would have to recall both sets of materials in order. Ss were assigned alternately to one of two experimental groups. They were told that they must first reeall the sentence and then the digits, or vice versa. Iri both eases, they were instructed to focus their attention equaIly on both inputs. Balancing order of recall seemed a prudent control in the light of Epstein's (1969) finding that a procedure sirnilar to the present one yielded results that were sensitive to the order of recalI. Eaeh group coniained 13 Ss, and each S received alI of the senten ces in the same order. RESULTS The proportion of digits eorreetly reealIed in each serial position is 'oe - . .- '0- . i I o- s 5 I ~. N_ . :l ""'· .... • -th·h : C "''' :J- J ;- Fig. 1. Proportion of words and digits correct as a function of order of recall and serial position. Psychon. Sei., 1971, Vol. 25 (6) Table 1 Correlations (With Regard to Correctness of Recall) Between Successive Words in the Sentences for the Digits Recalled First and Sentences Recalled First Groups Separately Form Class Preposition Det. Adj. Noun Det. Adj. Sen tence First Digits First .20 .58 .50 .51 .29 .52 .21 .38 .24 .39 shown in Fig. 1. Recall of digits was better when the digits were recalled first, F(1,24) = 70.34, P < .01, and the overall effect of serial position was also highly significant [F(9,216) = 80.21, p < .01]. The corresponding recall curves for the sentences are also shown in Fig. 1. Sentences were recalled better if they were recalled first, F(1,24) = 27.80, p < .01. There were no significant interactions between sentences and digits at recaH. The presence of aserial position effect and the relative goodness of recall of the' nouns, on the one hand, and their modifiers, on the (...truncated)