| Summary: | <p>This thesis is about the nature of a man's immediate memory for a short sequence of verbal items. Two components of immediate memory may be distinguished: an 'active' memory component, which is progressively destroyed within a few seconds if, subsequent to presentation of the sequence, the man is required to perform a distracting task which prevents rehearsal of the items, and a residual 'inactive' component which is more permanent. My concern is with behavioural evidence relating to the question of how the attributes of a verbal sequence are represented in <em>active</em> memory subsequent to identification of the constituent items. It is possible selectively to probe the retrieval of different attributes of the sequence by the use of appropriate tasks. Evidence on the performance of normal adult subjects in various short-term memory tasks, from the literature and from my own experiments, is extensively reviewed. There emerges a fundamental contrast between two groups of tasks: on the one hand, tasks which require the subject to indicate whether a test item was present in the memorised sequence (item recognition) or to judge the relative recency with which items have occurred; on the other,tasks which require the subject to base his response on the order or precise position of items in the sequence. To account for the nature of the contrast I put forward a general hypothesis which is an amended and elaborated version of parts of a general model of memory described by Morton (1970). It is argued that subsequent to identification of an item, representations of it may temporarily be held in active memory by two distinct storage mechanisms. Thefirst is as a decaying trace at the neural unit, called (following Morton) a 'logogen', responsible for the identification of that item.</p> <p>This trace may be treated as possessing a unidimensional 'strength' whose magnitude is dependent on how recently the item has occurred. Item recognition and judgements of recency are held to be mediated by assessment of the test item's trace strength with respect to a decision criterion located on the strength continuum. Secondly, the item may also be represented as one of several maintained in a serially- organised limited-capacity storage mechanism called (following Morton) the 'response buffer' . This holds a small number of the verbal responses recently made available by the logogens, coded as a string of articulatory descriptions or commands. It has two crucial properties. Firstly, items represented in it as potential responses may be fed back in sequence to the logogens for re-identification (the process of sub-vocal rehearsal). Secondly, since the serial order in which items enter the response buffer is retained as an intrinsic property of its structure, retrieval from it provides a ready mode of access to the order or position of items in the memorised sequence, which item-traces at logogens do not.</p> <p>The thesis falls into two parts. The first three chapters contain a theoretical review of the literature. In Ch.1, the active/ inactive distinction is introduced with some reference to the historical background (1.1). Some points of technique are raised (1.2), the aims, of the present work are outlined (1.3) and the twostore hypothesis is described (1.4)</p> <p>In Ch.2, evidence is reviewed from short-term memory experiments in which accuracy is the main dependent variable, beginning with experiments in which either serial recall of the whole sequence or probed recall of a single item is required. The hypothesis that the order of items is retained in active memory as an intrinsic property of the memory structure is contrasted with theories (e.g. Wickelgren, 1972) emphasising the formation of temporary associations between the representations of items adjacent in the list and/or between item and position representations. It is concluded that inter-item associations play no major role in active memory, and that item-position associations cannot account both for the partial independence of order and item errors and for the relationship between order errors and phonemic similarity. Conrad's (1965) model is introduced as a precursor of my own. It is concluded that items are recalled primarily from the (ordered) response buffer, but that traces at logogens influence the availability of responses as guesses when items are wholly or partly lost from the response buffer (2.11). Some evidence is then described which suggests that variables may be identified which differentially influence retention of information about the order of items and about their occurrence (2.12). Experiments comparing the effectiveness of position, context or both,as cues for recall are argued to imply that order and position are coded by the same mechanism, but that access to items retained by it may be more direct given a position rather than a contextual cue (2.13). We then turn to experiments on short-term recognition and judgements of recency, which are discussed in relation to the trace strength theory of Wickelgren and Norman (1966) (2.21). It is concluded that an item's recent identification is represented as an exponentially-decaying trace located at a unique permanent representation of that item, to which access is direct (2.22). But order-recognition experiments provide little evidence that associations between items are represented in active memory in the same way (2.23). The two-store hypothesis is then applied to some general problems of the functions of active memory: in speech comprehension (2.31) as an 'address register' for inactive memory (cf. Broadbent, 1971) (2.32) and as a working memory. Finally, the logogen system is discussed in relation to Sperling's (1967) 'recognition buffer' (2.34) and evidence for articulatory as opposed to acoustic coding in immediate memory is reviewed (2.35).</p> <p>In Ch.3, I introduce the experimental paradigms pioneered by Sternberg (1966), which involve presentation on each trial of a different sub-span list for memorisation, followed by a probe to which reaction-time (RT) is measured. Several models are outlined of the nature, dynamics and format of the representations mediating performance in Sternberg's item-recognition paradigm (IRn), in which the subject must indicate whether the probe was or was not a member of the list. Predictions are derived from Sternberg's two scanning models and two versions of a trace strength hypothesis and compared to data available in the literature. The evidence favours the hypothesis that subjects perform the IRn task by judging the strength of the trace at the amodal representation (logogen) to which the probe provides direct access (3.1). Evidence from a version of the IRn paradigm in which the memory set remains constant from trial to trial is then reviewed; it is argued that the results obtained are necessarily equivocal with respect to <em>active</em> memory; nevertheless, the concept of trace strength may be of use in this context also (3.2). Next, we see that Sternberg's own experiments on contextual recall (CR), in which the subject must respond with the item following the probed item in the list, yield a pattern of results very different from that obtained in IRn. His own explanation is in terms of different strategies of search within a single store. In view of the failure of the exhaustive scanning model of IRn, the differences may be better accounted for by the two-store model. Since these RT paradigms provide a relatively pure and sensitive way of probing the retrieval of different attributes from active memory, a strategy is suggested of looking for factors which differentially influence performance in IRn, CR and related tasks. Evidence already available suggests that these factors may include serial position, phonemic similarity and the availability of a position cue.</p> <p>The second part of the thesis describes my own experiments. These employ Sternberg's varied-set IEn and CR paradigms and a task in which the subject must indicate the location of the probe in the list presented on that trial (Ln). In Ch.4, I raise some methodological issues inherent in the use of reaction time as a measure and give details of the apparatus and general procedure common to the experiments.</p> <p>In Ch.5 are described three experiments on the effects of phonemic similarity between items of the memory set on performance in the three tasks. Five-consonant lists were presented visually at a fast rate; the probe followed immediately and subjects were instructed not to rehearse; vocabulary and other factors were held constant across tasks. In Exp.I (CR) and Exp.III (Ln) phonemic similarity increased both RT and error frequency. The relationship between serial position and the effects of similarity in both tasks suggested a slow forward serial search through a temporally-ordered phonemically-coded store, save perhaps for terminal items of the sequence. The effects of similarity were in both tasks diminished when the probed item was phonemically distinct from the list items and, in CR, when the response item was similarly isolated. In Exp. II, no comparable effects upon IRn performance were found. Moreover, whereas marked increases with serial position probed were obtained in both CR and Ln, in the IRn task RT showed an accelerating decrease with serial position probed. The contrast between the two patterns of results is argued to be incompatible both with Sternberg's account in terms of two retrieval strategies and with Wickelgren's associative theory.</p> <p>The first experiment described in Ch.6 (Exp.IV) demonstrated that IRn was not facilitated by a cue presented along with the probe which informed the subject which half (in terms of ordinal position) of the four digit memory set was relevant to his decision. This result is compatible with the proposal that access to the active memory representations involved in IRn is direct, and does not require a search through an ordered store. Exp.V compared the effects upon IRn and CR of giving with the probe a cue telling the subject in which of four spatio-temporal positions the probed item had been presented (if, in IRn, it was a positive probe). Experimental parameters and subjects were the same for both tasks. Information provided by the position cue had a marked facilitatory effect on CR, none on IRn (as in Exp.IV). The difference is held to reflect the fact that location in the list of the probed item and retrieval of its successor, in CR, requires a serial search of the response buffer, a search which may however be abbreviated if the subject knows at what position to begin the search.</p> <p>In Ch.7 I give an account of a single experiment (Exp.Vl) intended to test predictions critically distinguishing the models of IRn outlined in Section 3.1. Memory sets of 1-4 consonants were visually presented at a fast rate; subjects were told not to rehearse, Two variables were manipulated: 1) the recency of occurrence, prior to the present trial, of items presented as negative probes; 2) the recency of items in the memory set. The latter was varied by presenting the probe either immediately or following a short delay during which rehearsal was prevented by a vowel-naming task. It was found that the more recently a negative probe had occurred, the slower and less accurately it was rejected; the effect increased with set size. The introduction of a filled delay between list and probe reduced the slopes of the serial position and set-size functions. The former showed monotonic recency effects. These and other aspects of tho data are incompatible with any scanning model and favour the trace strength inspection hypothesis.</p> <p>In Ch.8, two early exploratory experiments are described in which the task was again IRn. Mixed memory sets of digits and letters were presented; the probe was randomly either a letter or a digit. The question was whether the subject could use the semantic class of the probe to restrict a search to the appropriate subset of the list; such a strategy would not be possible according to a pure trace strength hypothesis. While Exp. VII provided no evidence for such a 'directed search' strategy, the data from Exp.VIII, in which the conditions were changed so as to maximise the possibility of such a strategy, showed an increase in RT with the size of the subset of the same class as the probe. However, it transpires that given the conditions of the experiment, this result does not rule out an explanation in terms of the trace strength inspection model. Changes are suggested which would make the experiment more decisive.</p> <p>The last chapter is devoted to concluding comments,speculations and plans. The necessity for exploring the adequacy of the two-store model in relation to other areas of research is acknowledged; these include speech production perception and comprehension, attention, non-verbal memory and the study of patients with memory disorders due to brain damage. Further experiments are suggested to compare the effects in the various short-term memory tasks of semantic similarity, rate of presentation, and interpolated material. In conclusion it is suggested that a greater emphasis on the representation of motor as well as perceptual events does much tc clarify the nature of active memory.</p>
|
|---|