Вы находитесь на странице: 1из 9

Journal of Experimental Psychology: 1988, Vol. 14, No.

4, 601-609

Copyright 1988 by the American Psychological Association, Inc. 0096-J523/88/S00.75

Lexical Ambiguity and the Timecourse of Attentional Allocation in Word Recognition


George Kellas and F. Richard Ferraro
University of Kansas

Greg B. Simpson
University of Nebraska at Omaha

In two experiments the allocation of attention during the recognition of ambiguous and unambiguous words was investigated. In Experiment I, separate groups performed either lexical decision, auditory probe detection, or their combination. In the combined condition probes occurred 90, 180, or 270 ms following the onset of the lexical-decision target. Lexical decisions and probe responses were fastest for ambiguous words, followed by unambiguous words and pseudowords, respectively, which indicated that processing ambiguous words was less attention demanding than unambiguous words or pseudowords. Attention demands decreased across the timecourse of word recognition for all stimulus types. In Experiment 2, one group performed the lexical-decision task alone, whereas another group performed the lexical-decision task during the retention interval of a short-term memory task. The results were consistent with those from Experiment 1 and showed that word recognition is an attention-demanding process and that the demands are inversely related to the number of meanings of the stimulus. These results are discussed with regard to the structure of the mental lexicon (i.e., single vs. multiple lexical entries) and the effect of such a structure on attentional mechanisms.

The process of word recognition is considered a cornerstone of language comprehension, because almost any view of language understanding requires that lexical information serve as input to syntactic and semantic processors. Consequently, word recognition research has proliferated, and models have grown in sophistication as researchers attempt to understand the basic components of lexical processing. The present research was designed to examine the attentional demands associated with word recognition. Despite frequent discussionsof the role played by attentional resources in word processing (e.g., Brown, Carr, & Chaderjian, 1987; Henik, Friedrich, & Kellogg, 1983; Simpson & Burgess, 1985), very few studies have examined directly the allocation of limited-capacity resources during word recognition. An exception is a study by Becker (1976), which used a dual-task procedure and showed that a subject's ability to respond to an auditory probe varied as a function of the frequency of a word presented as the stimulus for the primary task (lexical decision), with high-frequency words allowing faster proberecognition latencies than low-frequency words. Because printed frequency is known to have powerful effects on word
This research was supported by Graduate Research Fund Grant 3247-XO-20-0038 and Biomedical Sciences Support Grant 4382-200711-0 from the University of Kansas Graduate School (awarded to George Kellas) and in part by United States Department of Education Grant G008630072 to the Kansas Bureau of Child Research and the Learning Disabilities Center of the University of Kansas. Experiment 1 was completed as partial fulfillment of the requirements of the master's degree by F. Richard Ferraro under the direction of George Kellas. Greg B. Simpson is currently a Postdoctoral Trainee in the Child Language Program of the University of Kansas. We wish to thank Frances J. Friedrich and Thomas Carr for their comments on an earlier draft of the article. Correspondence concerning this article should be addressed to George Kellas, Department of Psychology, 426 Eraser Hall, University of Kansas, Lawrence, Kansas 66045.

recognition (e.g., Forster & Bednall, 1976; Jastrzembski, 1981; Rubenstein, Garfield, & Millikan, 1970; Whaley, 1978), its influence on a secondary task indicates that the attentional demands of word-recognition processes are sensitive to characteristics of the stimulus. In the present research we used dual-task procedures to examine attention allocation during the early operations associated with word-meaning retrieval. Just as word frequency has been shown to have strong effects on recognition, so has the number of meanings associated with a single lexical form. Words with multiple meanings represented by a single spelling (homographs) are typically responded to more rapidly than are words with only one meaning (Jastrzembski, 1981; Jastrzembski & Stanners, 1975; Rubenstein et al., 1970; Rubenstein, Lewis, & Rubenstein, 1971; but see Gernsbacher, 1984). This homography effect is usually explained by proposing that ambiguous (i.e., multiple-meaning) words have multiple representations in the internal lexicon. For example, in an elaboration of Morton's (1969, 1979) logogen model, Jastrzembski (1981) has argued (following Morton, 1979) that ambiguous words are represented by separate logogens for each meaning. The more logogens that are associated with a particular form, the greater the likelihood that one of these will reach its recognition threshold sooner than that representing a word with only one meaning. Words that possess more than one meaning (and hence more than one logogen) should therefore be recognized faster than words that possess only one meaning (and only one logogen) or no meaning (and, therefore, no logogen). An explanation provided by Rubenstein et al. (1970) is quite similar. Although couched in terms of a search through the lexicon rather than the incrementing of logogens, their explanation offers substantially the same account of the relation between a multiplicity of lexical representations and word-recognition speed. Neither of these models, however, offers any account of the attentional demands involved in processing ambiguous words.
601

602

G. KELLAS, F. FERRARO, AND G. SIMPSON

Indeed, the relation between attention and the initial processing of word meaning has generally been neglected, although the role of attention in the resolution of ambiguous word meanings in context has been examined (see Simpson, 1984, for a review). Becker's (1976) results, however, showing greater capacity demands as well as longer response latencies for low-frequency words, may be taken to suggest that singlemeaning words require more attention for their recognition than do ambiguous words. In Becker's verification model, a pool of word candidates generated from an initial analysis of the input is searched serially until a match with the input is found. The set is ordered according to frequency, with more common words undergoing comparison (verification) before lower-frequency words. Because each comparison is assumed to demand resources, the earlier verification of high-frequency words results in lower attentional requirements as well as shorter latencies to respond to the word. In the case of ambiguous words, their faster recognition in comparison to single-meaning words (even with frequency controlled; Jastrzembski, 1981; Rubenstein et al., 1970) suggests that they too have priority in a verification set. If we again assume that each meaning has its own representation in memory, then presumably each of these is present in the candidate set, increasing the probability of one of them being verified very rapidly. This view suggests that the ambiguity effect functions in a manner similar to word frequency. Because Becker has shown that both shorter response latencies and lower resource demands characterize the processing of high-frequency words, the model may also predict that recognition of ambiguous words places a lesser burden on attention. In the first experiment to be reported, subjects made lexical decisions about words with either a single meaning or multiple meanings or about pseudowords (with no meaning). In addition, they were required to indicate the presence of a tone. It was expected that responses to the probe would be slower when performed in conjunction with the lexical-decision task than when performed alone (Becker, 1976). If attentional resources are also sensitive to factors associated with the early stages of meaning access, then we further expected probe reaction times (RTs) to vary as a function of the number of meanings of a stimulus. In the second experiment lexical decisions were performed in the presence or absence of a memory load (Logan, 1978). In this case, if lexical decisions are resource demanding, they should be slower in a memoryload condition. This method was used to provide converging evidence on the attentional demands of word recognition under circumstances in which the response requirements of the two tasks were not as similar as in Experiment 1. Experiment 1 In the first experiment, subjects made lexical decisions about ambiguous words, unambiguous words, or pseudowords. A tone served as the probe stimulus for the secondary task and followed the onset of the letter string by a stimulus onset asynchrony (SOA) of 90, 180, or 270 ms. These times were chosen to examine attentional demands across the time course of the early stages of word recognition. If these processes are attention demanding, then the performance of the

secondary task (detection of the probe tone) should decline when performed in conjunction with the primary task in comparison to its performance in isolation. In addition, if attentional processes are implicated in the early stages of meaning retrieval, then the nature of the lexical-decision target should have further effects on secondary task performance. Finally, the SOA manipulation should inform us about changes over time in the attentional requirements of word recognition. Method
Subjects. Fifty-nine undergraduate students participated as volunteers or for course credit. Data from 11 subjects were eliminated, three because of excessive errors on the lexical decision task (greater than 10% errors), three for mean reaction times that were greater than two standard deviations above the group mean, and five because of equipment failure. All were native speakers of English and reported normal or corrected-to-normal vision and normal hearing. Sixteen subjects were randomly assigned to each of the three conditions (primary task alone, secondary task alone, and combined tasks). Pilot data indicated that there were asymmetric transfer effects among conditions when the tasks were manipulated within subjects. Therefore, separate groups of subjects were used in each task condition (see Poulton, 1982). An additional 72 subjects rated the stimuli for number of meanings, and 47 more rated them for experiential familiarity. (The rating procedures are described below.) These subjects were drawn from the same population as the subjects participating in the main experiment. Stimuli. An initial pool of 576 letter strings was developed. This pool included 198 ambiguous words taken from Nelson, McEvoy, Walling, and Wheeler (1980) and Simpson (1979), 192 unambiguous words selected from Webster's Seventh New Collegiate Dictionary (1976) and Roget 's II New Thesaurus (1976), and 186 pseudowords taken from Juola, Ward, and McNamara (1982) or formed by changing one vowel of a word from the dictionary or thesaurus. All pseudowords were orthographically legal (Massaro, Taylor, Venezky, Jastrzembski, & Lucas, 1980), and none was derived from any word in the pools of ambiguous or unambiguous words. All words and pseudowords were pronounceable, one syllable, and between three and seven letters in length. These 576 letter strings were then rated for the perceived number of meanings by two groups of subjects (37 rated 376 of the stimuli, and 35 rated the remaining 200). Letter strings were presented randomly in booklet form, with 20-25 letter strings per page. Adjacent to each string were the numbers 0, 1, and 2. Subjects were instructed to read each letter string, decide if it had no meaning (0), one meaning (1), or more than one meaning (2), and circle the appropriate number. From these ratings, a set of 60 ambiguous words, 60 unambiguous words, and 120 pseudowords was selected. The criterion for selection was that a stimulus show at least 70% agreement across subjects as possessing no meaning, one meaning, or more than one meaning. A one-way analysis of variance indicated that the rated number of meanings differed for the three stimulus types, f\2,237) = 6,792.00, MS, = 0.01, p < .01. Newman-Keuls comparisons showed that those words initially identified as ambiguous were indeed rated as having more meanings (1.86) than unambiguous words (1.15), which, in turn, had more meanings than did pseudowords (0.09). The three stimulus types did not differ, however, on either the positional bigram frequency (Massaro et al., 1980) or the number of letters (both Fs < 1). In addition, the two word types did not differ in printed frequency (Kucera & Francis, 1967), r(118) = .05,/.10. Finally, all word stimuli were rated for experiential familiarity. Gernsbacher (1984) has demonstrated the importance of this factor

ATTENTION AND WORD RECOGNITION

603

by showing that several common effects seen with word stimuli (including the tendency for ambiguous words to be recognized more quickly) change when familiarity is controlled. Familiarity ratings were obtained by the same method as that used by Gernsbacher (1984). Forty-seven subjects were instructed to rate each of the ambiguous and unambiguous words on a 7-point scale, with 1 corresponding to very unfamiliar and 7 corresponding to very familiar. Rated familiarity did not differ between the two word types (6.61 and 6.56 for ambiguous and unambiguous words, respectively), S(\ 18) = 1.62,p> . 10. In addition to the experimental stimuli, 30 letter strings (10 of each type) not selected for the experiment were used in practice trials, and 12 (4 of each type) served as buffer trials. All letter strings were typed in uppercase Orator typeface (IBM), centered on 0.80 x 1.28-cm plain white index cards. A blank space separated each letter, so that at a viewing distance of 82 cm (the distance from the stimulus card to the facepiece of the tachistoscope), stimuli subtended maximum visual angles of 0.28 vertically and 1.96 horizontally. A plus sign was typed at the center of another card and served as the warning stimulus for all trials. Design and procedure. The design for the experiment was a 3 x 3 x 3 x 2 mixed factorial. The condition (primary task alone, secondary task alone, and combined task) was manipulated between subjects. Stimulus type (ambiguous, unambiguous, and pseudoword), SOA (90, 180, and 270 ms), and block (Block 1 and Block 2) were within-subject factors. Subjects were seated in front of a tachistoscope and wore headphones, regardless of the condition to which they had been assigned. The subjects in each condition completed 282 trials. The first 30 trials were practice, followed by two blocks of 126 trials each (6 buffer trials preceding 120 experimental trials). They were given a 1-min rest after the practice trials and after completion of half of the trials in each experimental block. A 5-min rest was given between blocks. Accuracy feedback was given during the rests. Subjects committing errors on more than 10% of the trials preceding a rest were asked to concentrate more on accuracy. On each trial of the primary task in isolation condition, a plus sign warning signal was presented in one field of the tachistoscope for 500 ms. One second following the offset of the warning signal, a letter string was presented for 500 ms. Onset of the letter string initiated a millisecond timer, which was stopped when the subject responded by pressing one key to indicate WORD or the other to indicate NONWORD. Subjects were instructed to respond to the letter string as rapidly as possible but to avoid making errors. They used the index and middle fingers of one hand for their lexical decision responses. Half of the subjects used the right hand to respond, and half used the left hand. The mapping of WORD and NONWORD response to the middle and index fingers was also counterbalanced within response hand. The response automatically started a second timer, which timed a 15-s intertrial interval. In the secondary task in the isolation condition, subjects again saw a plus sign at the beginning of each trial for 500 ms. A 500-ms auditory probe (70 dB and 4 KHz) was presented through headphones 1,090,1,180, or 1,270 ms after the offset of the plus sign. These values correspond to the time that elapses between the lexical-decision stimulus and the tone in the combined task condition (90, 180, and 270 ms). Onset of the probe started a millisecond timer, which was stopped when the subject responded by pressing the probe key. Subjects were instructed to respond to the probe as rapidly as possible. Half of the subjects used the right hand for this response, and half used the left. This response started the intertrial interval, and the next trial followed after 15 s. In the combined condition, each trial again began with the presentation of a plus sign for 500 ms. One second after the offset of this signal, a letter string was presented, and a timer was started, which

was stopped by the subject's lexical-decision response. The letter string remained visible for 500 ms. Following the onset of the letter string by 90, 180, or 270 ms, the auditory probe was presented for 500 ms. Onset of the probe started a second timer, which was stopped by the subject's pressing of the probe key. The second key to be pressed initiated the 15-s intertrial interval. Subjects were instructed to respond to both the lexical decision and the probe as rapidly and as accurately as possible. They were instructed to divide their attention equally between the two tasks.1 The lexical-decision responses were made with the index and middle fingers of one hand, whereas the probe response was made with the index finger of the other hand. Assignment of task to the two hands and assignment of lexicaldecision response to the index and middle fingers were counterbalanced across subjects. The order of presentation of the letter strings was identical in the primary task in isolation and the combined conditions. Likewise, the SOA order was the same when the probe was presented alone and when presented in combination with the lexical-decision task. The order of letter strings and probe intervals was random, with the constraint that no more than three consecutive occurrences of letter-string type or SOA be allowed. Half of the subjects received the stimuli in one order, and half received them in the opposite order.

Results Several analyses were performed on the lexical-decision and probe data. For analyses including lexical decisions from the combined condition, only those trials on which the probe response was rendered before the lexical decision were included in this analysis. Responses to the probe that occurred after the lexical-decision response could not discriminate between the attentional requirements of the primary and secondary tasks, because the probe response could be affected by the decaying memory trace of the auditory probe. Any apparent effect of stimulus type on probe RT, therefore, could simply be a reflection of different lexical-decision times for the different types of letter-string targets. Subjects responded to the probe first on approximately 60% of the trials, and this percentage was constant across both stimulus type and SOA.2 This is the pattern that would be expected with the unbiased divided attention instructions given to the subjects. Of the 16 subjects originally participating in the combined task, data from 3 were excluded completely because they responded to the letter string first on all trials. After the elimination of all such data, the resulting analysis was based on 1,872 observations.
1

Pilot testing showed that divided-attention instructions were necessary to ensure invariance between speed of lexical decisions performed in isolation and in the combined condition. Our use of primary to describe the lexical-decision task reflects our focus on word recognition processes and not any task bias deliberately induced in the subject. 2 Although an analysis including all trials is not as clearly interpretable as one performed only on those trials on which the tone response was given first, an analysis performed on all of the data from Experiment 1 was fully consistent with that reported above. As in that analysis, the main effects of SOA and stimulus type were significant, with no further main or interactive effects. In addition, the variability in the order of responding provided evidence against the possibility that subjects were executing a single molar response, in effect treating the combined condition as a single more complex task.

604

G. KELLAS, F. FERRARO, AND G. SIMPSON

First, we report a comparison of lexical-decision times when this task is carried out in isolation and when it is part of the combined task condition. Next, a similar comparison is made for probe reaction time. Finally, the probe times from the combined task condition are analyzed to assess the effects of word-recognition processes on probe detection. In all analyses involving stimulus type (ambiguous, unambiguous, and pseudoword), all three levels were included, because the stimuli were scaled for meaning by the number-of-meanings rating procedure. Therefore, pseudoword stimuli serve as the appropriate no-meaning control. However, analyses excluding pseudoword stimuli were performed as well and in every case showed the same results as those to be reported below. Trials on which lexical-decision times were less than 200 or greater than 2,000 ms were treated as outliers and eliminated from analysis, as were probe reaction times less than 100 or greater than 500 ms. On the basis of these criteria, 0.89% of the responses were lost from the primary task alone, 1.46% from the primary task in the combined condition, and 1.50% from the secondary task in the combined condition. Lexical decision. Mean lexical-decision times and error percentages for ambiguous words, unambiguous words, and pseudowords are shown in Table 1, both when the lexical decision task was carried out alone and in conjunction with probe detection. In order to interpret probe RT as an index of the attentional demands of word recognition and to ensure that both tasks tap a common pool of resources, it is necessary to demonstrate that the secondary task does not interfere with the primary task of lexical decision (Kerr, 1973). Therefore, the lexical decision times were first analyzed in a 2 x 2 x 3 mixed analysis of variance, with the between-subjects factor of task condition (isolated and combined), and within-subjects factors of block (1 and 2) and stimulus type (ambiguous, unambiguous, and pseudoword). A parallel analysis was conducted by using stimuli as a random factor. Only the main effect of stimulus type was significant, /"n(2, 131) = 59.28, p < .001. Neither the effect of task condition nor the interaction of condition and stimulus type approached significance in either the subject or stimulus analyses, which indicated that the probe task did not interfere with the primary task of lexical decision. There were also no main or interactive effects of block. In fact, this factor did not show any significant effects in any analysis of this research and will not be discussed further. A final analysis identical to those above was conducted on error percentages. Only the effect of stimulus type was signif-

Table 1 Mean Lexical Decision Times, Standard Deviations, and Percentage of Errors (PE) as a Function of Condition and Stimuli From Experiment 1
Stimuli Ambiguous Primary task Isolation Combined Note. M 725 744 SD 97 94 PE 0.6 1.0 Unambiguous M 763 782 SD 98 106 PE 1.2 1.9 Pseudoword M 858 867 SD 110 119 PE 1.8 3.1

icant, F(2, 30) = 24.14, MSC = 2.69, p < .01. Errors increased in the order ambiguous, unambiguous, and pseudowords. Probe reaction time. Next, a comparison was made between probe reaction time when performed alone and when performed in combination with the lexical-decision task. Comparisons were made separately for ambiguous words, unambiguous words, and pseudowords, as stimulus type is not a factor when the secondary task is performed alone. Therefore, for each stimulus type, a 2 (isolated and combined task) x 2 (block) x 3 (90, 180, and 270 ms SOA) analysis of variance was performed. In each case, the main effects of task condition and SOA were significant, as was the interaction of these factors, Fs(2, 30) = 148.88, 90.16, and 142.21 for the comparisons involving ambiguous words, unambiguous words, and pseudowords, respectively (respective MS,s 678.03, 1,166.58, and 686.63; all ps < .01). This pattern, which is displayed in Figure 1, suggests that limited capacity resources are used for both tasks. Consequently, probe RTs are longer when the probe occurs during word recognition than when it occurs in isolation. Mean RTs for the secondary task in isolation were submitted to a 3 (SOA) x 2 (block) analysis of variance. Only the main effect of SOA was significant, F(2, 30) = 9.99, MSC = 30.84, p < .01. Reaction times to probes presented at SOAs of 180 and 270 ms were faster than those at the 90-ms SOA. The difference between probe RTs at 180 and 270 ms were not significant, although responses were slightly faster at 180 ms. This V-shaped SOA function is often seen in simple reaction time research that uses irregularly presented preparatory intervals and is interpreted frequently in terms of subjects' expectancy for the average of the intervals presented (Kellas, Baumeister, & Wilcox, 1969). Combined task condition. The analysis of greatest interest is that of probe reaction times for the. three stimulus types in the combined condition, because this comparison informs us of the attention demands of word recognition processes. Probe times were submitted to a 3 (stimulus type) x 2 (block) x 3 (SOA) analysis of variance. The main effect of stimulus type was significant, F(2, 24) = 47.97, MS, = 7,360.31, p < .01. Multiple comparisons (Newman-Keuls) indicated that probe RTs were faster on ambiguous word trials than on either unambiguous word or pseudoword trials. Probe responses in the presence of unambiguous words were also faster than those with pseudowords. The main effect of SOA was significant also, F(2, 24) = 176.34, MS, = 3,136.27, p < .01. All pairwise comparisons of this effect were significant and revealed that probe responses were slowest at 90 ms and faster at 180 and 270 ms. These results indicate that word recognition processes become less demanding of resources across their timecourse. Discussion First, the results of the present experiment replicate those studies that have shown that multiple-meaning words are recognized more rapidly than are those with only one meaning (Jastrzembski, 1981; Rubenstein et al., 1970). Furthermore, this effect was obtained even when the stimuli were equated on such powerful variables as word frequency and experiential

Means and standard deviations are in milliseconds.

ATTENTION AND WORD RECOGNITION

605

between the execution of the two responses (Kahneman, 1973; Kantowitz, 1985). This interresponse interval (IRI) was calculated for each subject using theformulaIRI = RT1 - RT2 - SOA, where RT1 represents lexical decision time and RT2 represents probe response time. A 2 (stimulus type) x 3 (SOA) analysis of variance revealed a significant effect of stimulus type, F(2, 24) = 12.18, MSf = 626.88, p < .01. IRI increased in the order ambiguous, unambiguous, and pseudowords. This departure from a constant IRI argues against the hypothesis that the two responses are in fact executed as a single response. A second potential qualification of the results of this experiment concerns the possibility that the apparent attentional demands of word recognition are traceable to response competition (e.g., Kantowitz, 1985). Therefore, in Experiment 2, we examined word recognition in the presence of a recall task, in which the response requirements are very different from those of lexical decision.

z g H o

s ct
o a a.

s
90
180 270
STIMULUS ONSET ASYNCHRONY (IN MSEC)

Figure 1. Mean probe reaction times from the combined and isolated probe conditions in Experiment 1 as a function of stimulus onset asynchrony, stimuli (PW = pseudoword; U = unambiguous; A = ambiguous), and isolated secondary task (I = in isolation).

Experiment 2
The purpose of the second experiment was to provide converging evidence for the attentional demands of word recognition. The method used was derived from Logan (1978), in which subjects made their speeded responses while engaged in a concurrent short-term memory' task. Such a method should not be as susceptible as that used in Experiment 1 to an interpretation based on response competition. Subjects again made lexical decisions to ambiguous and unambiguous words and pseudowords. On each trial, subjects in one group were given a set of seven digits to be recalled after making their lexical decision response. The remaining subjects performed the lexical decision alone. If word recognition is an attention-demanding process, then performance should be slower when accompanied by the digit recall task than when performed in isolation.

familiarity (Gernsbacher, 1984). These results also support the contention that the early stages of word recognition, including the initial operations relevant to a determination of a word's meaning, are attention-demanding processes. The latter point is supported by the fact that the number of meanings possessed by a word influences a subject's ability to perform a concurrent task; that is, multiple-meaning words allow better performance on a secondary task than do words possessing only one meaning or no meanings. Finally, it is evident that the attentional demands of word recognition decrease during the timecourse of word recognition. As sensory information is extracted, the demand for resources diminishes, and these resources are increasingly available for deployment in the service of another task, and probe detection becomes faster. The results are in accord with Becker's (1976) finding that word recognition is attention demanding, and they show that a word's meaning, as well as its frequency, influences attentional requirements. If each of an ambiguous word's meanings is represented by a separate entry in the internal lexicon, then, at a given level of frequency, such a word should be identified more rapidly than a single-meaning word, through either more rapid accrual of activation or a higher priority in the search process. If the processes involved in extracting information from the stimulus are attention demanding, as they appear to be here, then these resources may be withdrawn sooner for allocation to a secondary task. There are two possible qualifications of the results of this experiment. The first concerns the possibility that the responses on the two tasks were being executed as a single molar response. If this were the case, then the ability of the probe RT to inform us about the attention requirements of lexical decision would be compromised. Such a response grouping would be indicated by a constant interval across conditions

Method
Subjects. Thirty-three undergraduate students participated for course credit. Data from 1 subject were discarded for exceeding 10% errors in lexical decisions. Sixteen subjects each were randomly assigned to the memory load and no-load conditions. All subjects were native English speakers, had normal or corrected-to-normal vision, and had not participated either in Experiment 1 or the preliminary research. Design and procedure. The design for this experiment was a 3 X 2 mixed factorial. Stimulus type (ambiguous words, unambiguous words, and pseudowords) was the within-subjects factor, and memory load (load or no-load) was the between-subjects factor. The stimuli for the lexical decision task were the same as in Experiment 1. The procedure for the no-load condition, therefore, was identical to the primary task alone condition of the first experiment. A plus sign was presented for 500 ms, followed by a blank interval of 1,000 ms. The target letter string was then presented for 500 ms, its initiation simultaneously starting a clock that was stopped by the subject's lexical-decision response. The procedure for the load condition was the same, except that prior to onset of the warning signal, subjects were read aloud a random string of seven 1-digit numbers at an approximate rate of one every 0.75 s. Five seconds

606

G. KELLAS, F. FERRARO, AND G. SIMPSON

following the final digit, the warning signal for the lexical decision was displayed, and the rest of the trial proceeded as in the no-load condition. Following the lexical-decision response, subjects were asked to recite, in the correct serial order, the seven digits presented on that trial. Subjects were instructed to make their lexical decisions as rapidly and accurately as possible. Feedback on lexical decision accuracy was given between blocks and midway through each block, as in Experiment 1.

Results and Discussion Digit recall accuracy was highly consistent across the three stimulus types (82.2%, 83.4%, and 81.6% for ambiguous words, unambiguous words, and pseudowords, respectively; F< 1). Lexical-decision responses were excluded from analysis if they were longer than 2,000 or shorter than 200 ms. These criteria resulted in the exclusion of 1.3% of the responses in the no-load condition and of 1.9% in the load condition. Lexical-decision times for each of the three stimulus types and the two memory load conditions are shown in Table 2. These data generally show that lexical decisions were slower with a memory load than without and that they were inversely related to the number of meanings. Initially, to confirm that the lexical decision data replicated those of the first experiment, an analysis was performed comparing the no-load condition of Experiment 2 with the primary task alone condition of Experiment 1. The effect of stimulus type was significant in this analysis, F(1, 30) = 129.34, MS, - 983.39, p < .01, with ambiguous words leading to faster responses than unambiguous words, which in turn showed faster responses than pseudowords. The effect of experiment and the Experiment x Stimulus Type interaction did not approach significance. A similar analysis performed on error rates also revealed an effect only of stimulus type, F(2, 30) = 29.04, MSC = 2.47, p < .01. More errors were made to pseudowords than to unambiguous words, and more were made to unambiguous than to ambiguous words. Again, there was no effect of experiment or an Experiment x Stimulus Type interaction. These analyses indicate that lexical decision performance was virtually identical in the two experiments. The lexical-decision data from Experiment 2 were submitted to a 3 (stimulus type) x 2 (load and no-load) analysis of variance. Results revealed significant main effects of stimulus type, F(2, 30) = 92.63, MS, = 2,023.06, p < .01, and memory load, F(l, 15) = 18.58, MSC = 27,846.40, p < .01. The

interaction of these factors was also significant, F(2, 30) = 5.59, MS. = 2,046.35, p < .01. Analysis of the simple effects within this interaction showed that although the effect of load was significant for each stimulus type, Fs(l, 90) = 12.00, 15.74, and 31.64 for ambiguous words, unambiguous words, and pseudowords, respectively, MSC = 9,1 27.47, all ps < .0 1 , the difference between the load conditions was inversely related to the number of meanings, with differences of 117, 134, and 190 ms, respectively, for ambiguous words, unambiguous words, and pseudowords. A parallel analysis performed on subjects' error percentages revealed a main effect of stimulus type, F(2, 30) = 35.36, MS, = 9.28, p < .01, and a significant Stimulus Type x Load interaction, F(2, 30) = 10.36, MS, = 10.68, p< .01. Further analysis of this interaction showed that more errors were made on pseudoword trials in the load condition than without a memory load, F( 1 , 90) = 24.58, MS, = 10.74, p < .01. Ambiguous and unambiguous words, however, did not differ as a function of load condition, These results indicate that maintaining a memory load has a large detrimental effect on word-recognition latency and confirm the conclusion of Experiment 1 that word recognition is an attention-demanding process. As attention is allocated to maintaining a set of digits for later recall, less is available for the lexical-decision task. Furthermore, the Stimulus Type x Memory Load interaction shows that although this effect of memory load is seen for all stimuli, its magnitude is related to the number of meanings of the lexical-decision stimulus, indicating again that the recognition of ambiguous words is less demanding than that of unambiguous and that the latter is in turn less demanding than the processing of pseudowords. ' In addition, the consistency of digit recall across the three stimulus types indicates that the differential effect of load on the recognition of ambiguous and unambiguous words and pseudowords was due to the attentional requirements of recognition of these letter-string types and not to the effects of stimulus type on the ability to maintain digits in memory. It is clear, then, that effects showing the attentional requirements of word recognition are not the result of response competition between the two tasks, as might have been claimed of Experiment 1.

General Discussion
The present data are in substantial agreement with those previous studies that have argued that the early perceptual operations associated with word recognition are attention demanding. Our results are quite similar to those of Becker (1976) in showing attentional effects when lexical decisions are performed. Even the encoding of single letters has been
1 The Stimulus Type x Memory Load interaction must be viewed with caution, however, because it is attributable largely to the disproportionately large difference between the load conditions for nonwords, in comparison with either of the word conditions. Nevertheless, the fact that the load difference is 17 ms larger for unambiguous than for ambiguous words is in a direction consistent with the finding in the first experiment of greater capacity demands for unambiguous words.

Table 2 Mean Lexical Decision Times, Standard Deviations, and Percentage of Errors (PE) as a Function of Condition and Stimuli From Experiment 2
Stimuli Ambiguous Condition Load No load Note. M 853 736 SD 93 73 PE 0.3 0.6 Unambiguous M 904 770 SD 96 77 PE 1.1 1.5 Pseudoword M 1,038 848 SD 140 76 PE 4.3 1.9

Means and standard deviations are in milliseconds.

ATTENTION AND WORD

RECOGNITION

607

shown to affect the performance of a secondary task (Ogden, Martin, & Paap, 1980; Paap & Ogden, 1981). The latter studies suggest the need for a separation of two of the criteria typically considered for "automatic" processes: that they are obligatory and that they operate without requiring attentional resources (see Humphreys, 1985, for a review of these issues). Similarly, it may be that the recognition of a word leads in an obligatory fashion to the activation of associated semantic, phonological, and orthographic codes (e.g., Seidenberg & Tanenhaus, 1979), and yet such recognition is still attention demanding. The present data also indicate that very early processes relevant to the eventual determination of a word's meaning are implicated in these attention demanding operations. This point is supported, first, by the finding in Experiment 1 that multiple-meaning words result in better performance on a secondary task than do unambiguous words even at the shortest SOA (90 ms). It is confirmed in Experiment 2 by the finding that at a constant level of performance on digit recall, the effect that this task has on lexical decision varies with the number of meanings associated with the letter string. An important question arises regarding the mechanism by which the number of meanings and attention are related. It appears that the present data are handled most parsimoniously by postulating a limited amount of processing resources that may be switched between tasks. When enough information has accumulated for a lexical decision, these resources may be applied to the secondary task. Such an attentionswitching view seems most compatible with the direct relation between the two tasks: Conditions that lead to faster lexical decisions also result in faster probe RTs. As ambiguous words are recognized more quickly, they allow attention to be withdrawn from the lexical decision task sooner than do unambiguous words. The SOA effect is similarly compatible with such a view. As the probe is presented later in the lexical decision task, more of the word recognition operations have been completed, and less additional time will be required before the point is reached at which an attention shift may occur. Such an explanation may be weakened somewhat by the variability in response order and in the IRI data. If subjects had consistently waited to respond to the probe until some point in the word recognition process, we might have expected that response order would be consistent as well and that IRI would be constant across conditions. Neither type of consistency was seen, however. Nevertheless, these aspects of the data are only suggestive, and in light of the relation between lexical decision and probe RT, an attention-switching model seems preferable to an alternative, capacity-sharing view as an account of the present data. In any case, establishing the superiority of either an attention-switching or a capacitysharing interpretation (a debate of long standing in the attention literature) has not been the intent of the present studies, and the problem of distinguishing these views does not compromise the principal conclusion drawn from the research, that the attentional consequences of lexical decision vary with the number of meanings of the stimulus. The effect of stimulus type on probe RT clearly indicates that the withdrawal of resources from the lexical decision task is facilitated for ambiguous words compared with unambiguous ones.

We may also question the stage or stages of the lexicaldecision process that demand attention. On the one hand, it may be that activation of lexical candidates (or the perceptual analyses necessary for such activation) is attention demanding (Paap & Ogden, 1981). In this case, a multiplicity of meanings allows the encoding operations to be completed earlier, which in turn allows an earlier shift of attention to the probe task. On the other hand, it may be the decision process itself that is sensitive to the number of meanings, so that ambiguous words speed the postlexical access decision operations and thereby allow attention to be switched sooner. These alternatives could possibly be tested by examining dual-task performance with a task that does not require a decision (such as word naming) or by testing probe RT when no immediate response to the word is required at all. If a decrease in probe RT across SOA still occurred, we could infer that it is the early processes of lexical access that are attention demanding. The effect of ambiguity, both on recognition speed and on attentional demands, seems to require a model of the lexicon that incorporates multiple entries for ambiguous words within a lexical network, each linked to a separate entry in a conceptual network. Many studies of context effects on ambiguous word processing, both within psychology (e.g., Seidenberg, Tanenhaus, Leiman, & Bienkowski, 1982) and artificial intelligence (e.g., Cottrell & Small, 1983), have favored single lexical entries for ambiguous words, linked to multiple conceptual or word-sense nodes in a semantic network. (Seidenberg et al., 1982, argued for a distinction based on form class: Meanings from a single syntactic category, such as nouns, share a representation, whereas meanings from different form classes are separate.) Most research that has directly compared ambiguous and unambiguous words in isolation, however, has concluded in favor of multiple entries (e.g., Forster & Bednall, 1976; Jastrzembski, 1981; Rubenstein et al., 1971). With word frequency, bigram frequency, length, and familiarity equated, it is difficult to see how single lexical representations could result in the advantage for ambiguous words, unless we propose that multiple conceptual representations drive word recognition. Such a proposal seems unparsimonious, however, if we are able to account for the effect without appealing to processing levels higher than that of the word. The notion of multiple entries receives further support from a recent priming study by Masson and Freedman (1985), showing that the usual facilitation that arises for repeated words does not obtain for ambiguous words when different meanings are primed on separate trials (e.g., runner-race followed later by ethnic-race). This result is contrary to the very strong repetition effects usually seen in lexical decisions (e.g., Scarborough, Cortese, & Scarborough, 1977) and suggests that separate entries were being processed on the two trials. The model that we propose for these data draws heavily upon characteristics of the interactive activation model of word perception (McClelland & Rumelhart, 1981; Rumelhart & McClelland, 1982). In this conception, word recognition is achieved through activation spreading through a hierarchy of interconnected networks. At the lowest level, visual features of letters (lines, angles, etc.) are activated by the perceptual analysis of the presented stimulus. These features activate

608

G. KELLAS, F. FERRARO, AND G. SIMPSON

nodes at the letter level, which in turn activate entries at the word level. As information accumulates for a particular lexical entry, activation spreads back to the letter level to facilitate recognition of letters relevant to that word's identification. Entries within a level also interact with one another, so that activated entries tend to inhibit others at that level. To account for the present results, we amend the model to include separate entries at the lexical level for each meaning of an ambiguous item and to propose that each of these entries independently inhibits competing entries. The net amount of inhibition for competing items is therefore greater when an ambiguous word is presented, and the difference in activation between the presented word and its competitors is greater for ambiguous than for unambiguous words. Consequently, evidence converges at the letter level from two highly activated lexical entries, greatly facilitating recognition of the appropriate letters, relative to others. It is this facilitation of recognition at the letter level that leads to the earlier withdrawal of processing resources for the performance of a secondary task. Certainly, an interactive activation model is not the only possible account for the present data. For example, Becker's (1976) verification model can provide an interpretation of these results, as discussed previously. The interactive-activation account is proposed simply because it is most in line with much of the recent work in word recognition (e.g., Rumelhart & McClelland, 1982; Tanenhaus, Dell, & Carlson, 1987), including models of the effects of context on ambiguity resolution (Cottrell Si Small, 1983). Indeed, the reason that the present research does not strongly favor a particular model of word recognition is that, for the most part, such models have minimized or ignored the role of attention in the wordrecognition process. The restrictions that the present data carry are that a model recognize the attention-demanding nature of word-recognition processes and that it allow for faster and less demanding recognition of ambiguous words in isolation, regardless of frequency and familiarity.

References Becker, C. A. (1976). Allocation of attention during visual word recognition. Journal of Experimental Psychology: Human Perception and Performance, 2, 556-566. Brown, T. L., Carr, T. H., & Chaderjian, M. (1987). Orthography, familiarity, and meaningfulness reconsidered: Attentional strategics may affect the lexical sensitivity of visual code formation. Journal of Experimental Psychology: Human Perception and Performance, 13, 127-139. Cottrell, G. W., & Small, S. L. (1983). A connectionist scheme for modelling word sense disambiguation. Cognition and Brain Theory, 6,89-120. Forster, K. I., & Bednall, E. S. (1976). Terminating and exhaustive search in lexical access. Memory & Cognition, 4, 53-61. Gernsbacher, M. A. (1984). Resolving 20 years of inconsistent interactions between lexical familiarity and orthography, concreteness, and polysemy. Journal of Experimental Psychology: General, 113, 256-281. Henik, A., Friedrich, F. J., & Kellogg, W. A. (1983). The dependence of semantic relatcdness effects upon prime processing. Memory & Cognition, 11, 366-373. Humphreys, G. W. (1985). Attention, automaticity, and autonomy in visual word processing. In D. Besner, T. Waller, & G. Mac-

Kinnon (Eds.), Reading research: Advances in theory and practice (Vol. 5, pp. 253-310). New York: Academic Press. Jastrzembski, J. E. (1981). Multiple meanings, number of related meanings, frequency of occurrence, and the lexicon. Cognitive Psychology-, 13, 278-305. Jastrzembski, J. E., & Stanners, R. F. (1975). Multiple word meanings and lexical search speed. Journal of Verbal Learning and Verbal Behavior, 14, 534-537. Juola, J. F., Ward, N. J., & McNamara, T. (1982). Visual search and rapid serial presentations of letter strings, words, and text. Journal of Experimental Psychology: General, 111, 208-227. Kahncman, D. (1973). Attention and effort. Englewood Cliffs, NJ: Prentice-Hall. Kantowitz, B. H. (1985). Channels and stages in human information processing: A limited analysis of theory and methodology. Journal of Mathematical Psychology, 29, 135-174. Kellas, G., Baumeister, A. A., & Wilcox, S. J. (1969). Interactive effects of preparatory intervals, stimulus intensity, and experimental design on reaction time. Journal of Experimental Psychology, 50,311-316. Kerr, B. (1973). Processing demands during mental operations. Memory & Cognition, 1, 401-412. Kucera, H., & Francis, W. N. (1967). Computational analysis of present-day American English. Providence, RI: Brown University Press. Logan, G. D. (1978). Attention in character-classification tasks: Evidence for the automaticity of component stages. Journal of Experimental Psychology: General, 107, 32-63. Massaro, D. W., Taylor, G. A., Venezky, R. L., Jastrzembski. J. E., & Lucas, P. A. (1980). Letter and word perception: The role of orthographic structure and visual processing in reading. Amsterdam: North-Holland. Masson, M. E. J., & Freedman, L. (1985, November). Fluency in the identification of repeated words. Paper presented at the annual Psychonomic Society meeting, Boston, MA. McClelland, J. L., & Rumelhart, D. E. (1981). An interactive activation model of context effects in letter perception: Part 1. An account of basic findings. Psychological Review, 88, 375-407. Morton, J. (1969). Interaction of information in word recognition. Psychological Review, 76, 165-178. Morton, J. (1979). Word recognition. In J. Morton & J. C. Marshall (Eds.), Psyche/linguistics.'2. Structure and processes (pp. 107-156). Cambridge, MA: MIT Press. Nelson, D. A., McEvoy, C. L., Walling, J. R., & Wheeler, J. W. (1980). The University of South Florida homograph norms. Behavior Research Methods and Instrumentation, 12, 16-37. Ogden, W. C., Martin, D. W., & Paap, K. R. (1980). Processing demands of encoding: What does secondary task performance reflect? Journal of Experimental Psychology: Human Perception and Performance, 6, 355-367. Paap, K. R., & Ogden, W. C. (1981). Letter encoding is an obligatory but capacity-demanding operation. Journal of Experimental Psychology: Human Perception and Performance, 7, 518-527. Poulton, E. C. (1982). Influential companions: Effects of one strategy on another in the within-subjects designs of cognitive psychology. Psychological Bulletin, 91, 673-690. Roget's H new thesaurus. (1976). New York: Houghton. Rubenstein, H., Garfield, L., & Millikan, J. A. (1970). Homographic entries in the mental lexicon. Journal of Verbal Learning and Verbal Behavior, 9, 487-494. Rubenstein, H., Lewis, S. S., & Rubenstein, M. A. (1971). Homographic entries in the lexicon: Effects of systematicity and relative frequency of meanings. Journal of Verbal Learning and Verbal Behavior, 10, 57-62. Rumelhart, D. E., & McClelland, J. L. (1982). An interactive activa-

ATTENTION AND WORD RECOGNITION

609

tion model of context effects in letter perception: Part 2. The contextual enhancement effect and some tests and extensions of the model. Psychological Review, 89, 60-94. Scarborough, D. L., Cortese, C, & Scarborough, II. S. (1977). Frequency and repetition effects in lexical memory. Journal of Experimental Psychology: Human Perception and Performance, 3, 1-17. Seidertberg, M. S., & Tanenhaus, M. K. (1979). Orthographic effects in rhyme monitoring. Journal of Experimental Psychology: Learning, Memory, and Cognition, 5, 546-554. Seidenberg, M. S., Tanenhaus, M. K., Leiman, J. L., & Bienkowski, M. (1982). Automatic access of the meaning of ambiguous words in context: Some limitations of knowledge-based processing. Cognitive Psychology, 14, 489-537. Simpson, G. B. (1979). Meaning dominance and semantic context in the processing of lexical ambiguity. Unpublished doctoral dissertation, University of Kansas, Lawrence. Simpson, G. B. (1984). Lexical ambiguity and its role on models of word recognition. Psychological Bulletin, 96, 316-340.

Simpson, G. B., & Burgess, C. (1985). Activation and selection processes in the recognition of ambiguous words. Journal of Experimental Psychology: Human Perception and Performance, 11, 28-39. Tanenhaus, M. K., Dell, G. S., & Carlson, G. (1987). Context effects in lexical processing: A connectionist approach to modularity. Jn J. Garfield (Ed.), Modularity in know/edge representation and natural language understanding (pp. 83-108). Cambridge, MA: MIT Press. Webster's seventh new collegiate dictionary (1976). New York: Merriam-Webster.

Whaley, C. D. (1978). Word-nonword classification time. Journal of Verbal Learningand Verbal Behavior, 17, 143-154.

Received April 27, 1987 Revision received December 11, 1987 Accepted December 30, 1987

Rayner Appointed Editor ofJEP: Learning, Memory, and Cognition, 1990-1995


The Publications and Communications Board of the American Psychological Association announces the appointment of Keith Rayner. University of Massachusetts, as editor of the Journal of Experimental Psychology: Learning, Memory, and Cognition for a 6-year term beginning in 1990. As of January I, 1989, manuscripts should be directed to Keith Rayner Department of Psychology Tobin Hall University of Massachusetts Amherst, Massachusetts 01003 Manuscript submission patterns for JKP: Learning, Memory, and Cognition make the precise date of completion of the 1989 volume uncertain. The current editor, Henry Roediger, will receive and consider manuscripts until December 31, 1988. Should the 1989 volume be completed before that date, manuscripts will be redirected to Rayner for consideration in the 1990 volume.

Вам также может понравиться