Basic processes in reading: Is visual word

Psychonomic Bulletin & Review 2005, 12 (1), 119-124

Basic processes in reading: Is visual word recognition obligatory? EVAN F. RISKO, JENNIFER A. STOLZ, and DEREK BESNER University of Waterloo, Waterloo, Ontario, Canada Visual word recognition is commonly argued to be automatic in the sense that it is obligatory and ballistic. The present experiments combined Stroop and visual search paradigms to provide a novel test of this claim. An array of three, five, or seven words including one colored target (a word in Experiments 1 and 2, a bar in Experiment 3) was presented to participants. An irrelevant color word also appeared in the display and was either integrated with or separated from the colored target. The participants classified the color of the single colored item in Experiments 1 and 3 and determined whether a target color was present or absent in Experiment 2. A Stroop effect was observed in Experiment 1 when the color word and the color target were integral, but not when the color word and the color target were separated. No Stroop effect was observed in Experiment 2. Visual word recognition is contingent on both the distribution of spatial attention and task demands.

A skilled reader is typically familiar with about 30,000 words and can recognize a visually presented word in less than half a second (Rayner & Pollatsek, 1989). A major approach to understanding this degree of fluency assumes that visual word recognition is automatic. As such, visual word recognition is argued to be obligatory and ballistic, in that it is triggered by the appearance of the stimulus in the visual field and runs to completion (i.e., activates semantics) independently of the observer’s intentions. For example, Brown, Gore, and Carr (2002) asserted that Visual word recognition is largely obligatory, in the sense that lexical processing is initiated by the presence of a word in the visual field. (p. 236)

Thus, a lexical object, present in the visual field, always activates its semantic representation in long-term memory. We therefore can expect that an obligatory process is (1) independent of spatial attention and (2) independent of the mental set of the participant, where mental set is defined as a state of preparedness, determined by context or the person’s experience. The Stroop effect is often cited as strong evidence for the obligatory nature of visual word recognition (Stroop, 1935; see also MacLeod, 1991, for a review). Participants are instructed to identify the display color of a letter string, and response latencies are longer when the print color and the color word are incongruent (e.g., the word red in blue), relative to when they are congruent

This work was supported by a Natural Sciences and Engineering Research Council of Canada summer fellowship to E.F.R. and by Operating Grants 0183905 and A0998 to J.A.S. and D.B. We thank A. Polanowski for discussion. Address correspondence to E. F. Risko, Psychology Department, University of Waterloo, Waterloo, ON, N2L 3G1 Canada (e-mail: [email protected]).

(e.g., the word blue in blue) or neutral (e.g., the word house in blue). This result is interpreted as support for the hypothesis that visual word recognition is outside the control of the observer. The present investigation combines Stroop and visual search paradigms to provide a novel test of this claim. Visual Search In a visual search task, participants determine whether a target item is present among a set of distractors. Search times vary from efficient (response time [RT] set size slope 0 msec/item) to very inefficient (RT set size slope 30 msec/item; see Wolfe, 1998). Targets producing efficient searches (e.g., featural singletons) are said to be processed in parallel across space, in that focused attention is not necessary to discriminate them from distractors. In contrast, targets producing inefficient searches (e.g., a conjunction of features) require that participants focus attention on each item in order to make the target /distractor discrimination (Treisman & Gormican, 1988). The efficiency of the search can, therefore, be interpreted as an index of the focused attention devoted to a distractor item. As Treisman and Gelade (1980) noted, if a display is searched in parallel, only those characteristics of the distractor items whose processing does not require spatial attention will influence responses to the target item. Therefore, if the meaning of a distractor word in a search task influences performance when parallel search is evident (i.e., flat slope), this would provide strong evidence for the obligatory nature of visual word recognition. In contrast, if that same word does not influence performance when search is efficient, one would conclude that this attribute of the distractor (i.e., semantics) was not processed in parallel or available (e.g., Starreveld, Theeuwes, & Mortier, 2004). The appropriate conclusion would, therefore, be that visual

119

Copyright 2005 Psychonomic Society, Inc.

120

RISKO, STOLZ, AND BESNER

word recognition is not obligatory, since it depends on spatial attention. THE EXPERIMENTS The participants in the present experiments were presented with a display containing a variable number of words. A single word (the target) was colored, whereas the distractors appeared in white text on a dark background. A color word was included as either the color carrier (i.e., the integrated condition) or one of the distractors (i.e., the separated condition). If search is carried out in parallel, there are two possible outcomes: (1) All the words in the display are processed to the level of semantics, yielding a Stroop effect in both the integrated and the separated conditions, or (2) only the word acting as the color carrier, which is at the focus of attention, is processed to the level of semantics. This would produce a Stroop effect in the integrated, but not in the separated, condition. A further test of the obligatory nature of visual word recognition involved manipulating task demands across experiments. In Experiment 1, the participants were asked to classify the color of the target item. In Experiment 2, the participants were asked to detect the presence/absence of a specific color (e.g., is a red item present?). If activation of the distractor words is obligatory and ballistic, it is not immediately obvious why response mode should affect the propensity for an irrelevant word to interfere with performance. Finally, Experiment 3 replicated the separated conditions used in Experiment 1, except that a color bar was used as the color carrier, rather than a word. This was done to test the hypothesis that the presence of an irrelevant word as a color carrier in Experiments 1 and 2 increased the participants’ attentional load and restricted their ability to process other words in the display (e.g., Lavie, 1995). General Method Participants. Seventy-two University of Waterloo undergraduate students participated (24 in each experiment) in exchange for $4 each. All the participants reported normal or corrected-to-normal vision. Apparatus. The stimuli were presented on a 16-in. Viewsonic color monitor. Stimulus presentation and response collection were controlled by Micro Experimental Laboratory (MEL) software. Stimulus displays. The search display consisted of a 4 4 matrix yielding 16 locations (see Figure 1). The farthest distance between stimuli (12.4º of visual angle) was less than the distance reported in Brown et al. (2002), in which a Stroop effect was observed when the color word and the color carrier were separated by 12.9º of visual angle. Displays consisted of three, five, or seven words varying in length from three to six letters, displayed in the 16 possible locations. Words were presented in uppercase 72-point MEL system font. Horizontally, words subtended 1º, 1.33º, 1.6º, and 2.1º of visual angle for three-, four-, five-, and six-letter words, respectively. All the words subtended 0.5º of visual angle vertically. Words in the search display were either color words (i.e., green, red, blue, and yellow) or color-neutral words (e.g., house). Different color-neutral words were used as distractors on each trial and were repeated no

Figure 1. Dimensions of the search display used in the experiments. Distances between points are given in degrees of visual angle (where relevant, distances refer to the center-to-center distance). Fixation was in the center of the search display.

more than four times within an experimental session. Each display contained a single colored word (in Experiments 1 and 2; a color bar in Experiment 3); the display colors were green, red, blue, and yellow. Noncolored words appeared in white. The search display was presented on a black background. Design and counterbalancing. In Experiment 1, a 3 (set size: 3 vs. 5 vs. 7) 2 (integration: integrated vs. separated) 3 (congruency: congruent vs. incongruent vs. neutral) within-subjects design was used. In Experiment 2, a 2 (target presence: present vs. absent) 3 (set size: 3 vs. 5 vs. 7) 2 (integration: integrated vs. separated) 3 (congruency: congruent vs. incongruent vs. neutral) within-subjects design was used. Experiment 3 replicated the separated condition of Experiment 1, using a color bar as the color carrier. Therefore, a 3 (set size: 3 vs. 5 vs. 7) 3 (congruency: congruent vs. incongruent vs. neutral) within-subjects design was used. The levels of all the variables occurred randomly throughout the respective experiments. In all the experiments, congruent, incongruent, and neutral conditions appeared equally often and were fully crossed with all levels of the other variables in the experiment. In Experiments 1 and 3, all the colors and color words appeared equiprobably across all conditions. In Experiment 2, the target color was present on one half of the trials; one third of these trials were congruent, in which the color word corresponding to the target color was presented, one third of these trials were incongruent, in which the remaining three color words appeared equiprobably, and one third were neutral, with a noncolor word appearing as either the target or a distractor. When the target color was absent, all four color words appeared equiprobably, and three (four, excluding the target color) of the display colors appeared equiprobably. Target color was assigned between participants, and each color was used equally often across all participants. Procedure. The participants were tested individually, seated approximately 60 cm from the computer monitor. Each trial began with the presentation of a fixation symbol () in the center of the screen for 500 msec, which was followed by the experimental display. In Experiment 1, the participants were instructed to classify the color of a single colored word in the display by pressing one of four keys [c, v, n, or m], which was mapped to a color response (i.e., red, blue, green, or yellow). In Experiment 2, the participants were asked to detect the presence/ absence of a target color. Each participant was assigned a target color (e.g., is the color blue present /absent?) for the duration of the experiment. If the target color was present, the participants re-

STROOP SEARCH

sponded by pressing a key corresponding to a present response, and if the target color was absent, the participants responded by pressing a key corresponding to an absent response. The “c” and “m” keys were used. In Experiment 3, the participants were instructed to classify the color of a single colored bar in the display by pressing response keys as described for Experiment 1. In all the experiments, response-to-key mappings were counterbalanced across participants, and the display remained in view until a response was made. A response initiated a 1,000-msec intertrial interval. All the participants performed 36 practice trials and 576 experimental trials.

Results RT analysis was conducted for correct responses. These data were first subjected to a recursive trimming procedure that removed outliers on the basis of a criterion cutoff set independently for each participant in each condition by reference to the sample size and the standard deviation in that condition (Van Selst & Jolicœur, 1994). This trimming procedure resulted in 2.7%, 2.9%, and 3.0% of the data being discarded from Experiments 1, 2, and 3, respectively. Overall percentages of errors were low in all the experiments (5.5%, 4.7%, and 5.7%, respectively). The results for Experiments 1 and 2 (collapsed across the presence/absence factor) are displayed

in Figure 2, and the results for Experiment 3 are presented in Table 1. Experiment 1. The main effect of set size was not significant [F(2,46) 1.14, MSe 2,916.24, p .3], consistent with parallel search across the display. The main effect of congruency was significant [F(2,46) 15.93, MSe 7,204.52, p .001] but was qualified by a reliable integration congruency interaction [F(2,46) 12.38, MSe 9,286.36, p .001]. When the target color and the color word were integrated, the participants responded significantly more quickly on congruent trials (684 msec) and reliably more slowly on incongruent trials (794 msec), relative to neutral trials [715 msec; t(23) 3.69, p .05, and t(23) 3.50, p .05, respectively]. When the target color and the color word were separated, responses on congruent trials (724 msec), incongruent trials (723 msec), and neutral trials (723 msec) were equivalent (ts 1). No other effect approached significance (all ps .25). The error data were consistent with the RT data. The main effect of integration was reliable [F(1,23) 19.35, MSe 11.68, p .05], and the main effect of congruency was marginal [F(2,46) 3.07, MSe 23.66, p .056]. These effects were qualified by a reliable integration congruency interaction [F(2,46) 3.64, MSe

RT (msec) RT (msec)

Color and color word are separated

900 875 850 825 800 775 750 725 700 675 650 625 600 575 550 525 500 475 450 425 400 375 350 325 300

RT

Classify

% Error

Congruent Incongruent Neutral

10 5 0

% Error

Experiment 1

Color and color word are integrated

Detect

10 5 0 3

5

7

3

5

% Error

Experiment 2

121

7

SET SIZE Figure 2. Response times (RTs, in milliseconds) and percentages of error as a function of integration, congruency, and set size for the classify instructions in Experiment 1 and for the detect instructions in Experiment 2 (collapsed across the presence/absence factor).

122


Table 1 Response Times (RTs, in Milliseconds) and Percentages of Errors (%E) for Classifying a Colored Bar, as a Function of Set Size and Irrelevant Word–Color Bar Congruency in Experiment 3 Irrelevant Word – Color Bar Congruency Congruent Incongruent Neutral Set Size RT %E RT %E RT %E 3 641 6.3 643 6.8 648 5.3 5 643 5.2 646 6.1 647 4.8 7 664 6.0 657 5.7 646 5.2

17.17, p .05]. Percentages of error for congruent, incongruent, and neutral trials were 4.9%, 7.7%, and 5.9%, respectively, for integrated trials and 4.7%, 4.8%, and 4.7%, respectively, for separated trials. Experiment 2. None of the main effects (i.e., target presence, set size, or congruency) were significant (all Fs 1). The congruency integration interaction was not significant [F(2,46) 1.04, MSe 1,263.80, p .35]. The set size congruency interaction was marginal [F(4,92) 2.20, MSe 1,061.01, p .1], and there was a significant target presence integration interaction [F(1,23) 5.64, MSe 988.23, p .05], in which the participants appeared to respond more quickly on integrated trials (453 msec), relative to separated trials (461 msec), when the target was present, but not when the target color was absent (461 and 459 msec for integrated and separated, respectively). No other effect was reliable (all ps .13). The error analysis yielded a main effect of target presence [F(1,23) 8.35, MSe 100.20, p .05] that was qualified by a marginal target presence integration interaction [F(1,23) 3.71, MSe 10.97, p .05]. The participants made more errors on integrated trials (6.2%), relative to separated trials (5.4%), when the target was present, yet made similar percentages of errors on integrated (3.8%) and separated (3.9%) trials when the target was absent. Experiment 3. In the RT analysis, there were no significant main effects, nor was there a significant congruency set size interaction (all Fs 2.3, ps .12). The error analysis indicated a marginal main effect of congruency [F(2,46) 2.92, MSe 78.20, p .065]. No other effects approached significance (all Fs 1.4, ps .27). DISCUSSION In the classify task (Experiment 1), a large Stroop effect (incongruent congruent 110 msec) was observed when the color word was integrated with the color

carrier. In contrast, no Stroop effect (1 msec) was observed when the color word and the color carrier were separated. In the detect task (Experiment 2), no Stroop effect was observed in either the integrated or the separated condition (0 and 1 msec, respectively). The results of Experiment 3, demonstrating no Stroop effect when the color bar and the color word were separated, suggest that the failure to observe a Stroop effect in the separated condition in Experiment 1 did not result from a capacity limitation on visual word recognition (see, e.g., Brown et al., 2002). Instead, taken together, the results of these experiments suggest that visual word recognition is dependent on both spatial attention and the mental set of the participant. As such, these results are inconsistent with the standard account that visual word recognition is obligatory. The mere appearance of a lexical string in the visual field is insufficient to trigger processing that runs to completion. Spatial attention and visual word recognition. In contrast to the results reported here, other researchers have observed Stroop interference from a spatially separated color word (e.g., Brown, 1996; Brown et al., 2002; Kahneman & Chajczyk, 1983). Why did we not observe such an effect? An important factor that needs to be considered when evaluating these seemingly disparate patterns of results is spatial uncertainty. Specifically, the location of the target stimulus in the present study was highly uncertain, as compared with previous studies. In those studies, the target was typically presented in the same location throughout the experiment, whereas the color word(s) were presented in the periphery. Assuming that target localization requires resources and that processing of distractor information is contingent upon having adequate resources (see Lavie, 1995), one would, therefore, predict that an increase in spatial uncertainty would decrease the likelihood that a peripheral word will be processed. Some evidence for this claim can be found in Brown et al. (2002). Brown et al. had participants name the color of a bar and presented irrelevant color-neutral word(s) in the periphery. In their Experiments 1–3, the target appeared randomly in one of three possible target locations. The Stroop effects observed in these experiments were smaller than those observed in Experiments 4 and 5, in which the color bar appeared in the same location on every trial (see Table 2). These trends suggest that spatial uncertainty and the need to locate the target plays a significant role in modulating interference from an irrelevant word (for a related argument, see Stolz & McCann, 2000). The degree of spatial uncertainty in the present experiments was considerably larger than that in any of the pre-

Table 2 Stroop Effects as Reported in Brown, Gore, and Carr (2002) Spatial Uncertainty No Spatial Uncertainty Near distractors 70 msec (Experiment 1) 105 msec (Experiment 5) Far distractors 55 msec (Experiment 2), 58 msec (Experiment 3) 79 msec (Experiment 4)

STROOP SEARCH vious experiments, thereby promoting smaller Stroop effects. Mental set and visual word recognition. The joint results in Experiments 1 and 2 also suggest a role for the nature of the task in determining whether an irrelevant word interferes with performance. When participants performed the detect task, no Stroop effect was observed in either the separated or the integrated condition. This is in stark contrast to the large Stroop effect (110 msec) found in the integrated condition for the classify task in Experiment 1. One difference between the classify and the detect tasks is the number of response alternatives. In the classify task, the participants made a four-alternative discrimination (4AD), whereas in the detect task, the participants made a two-alternative discrimination (2AD). RTs are typically longer in 4AD tasks than in 2AD tasks (Jolicœur, 1999); therefore, one might argue that the use of a 2AD detect task did not allow time for the irrelevant word to interfere. This argument can be falsified by considering a study conducted by Bauer and Besner (1997), who compared Stroop effects generated from a 2AD classify task versus a 2AD detect task. They obtained a significant Stroop effect in the classify task, but not in the detect task, suggesting that the nature of the task, rather than the number of responses, influenced performance in the present experiments. What, then, should we make of the task-influenced results reported here? Translation accounts of Stroop interference focus on the internal representations required for a response and offer an interesting framework through which to understand the present results (Durgin, 2003; Virzi & Egeth, 1985). In the standard Stroop task, responses are typically verbal. Words are argued to have privileged access to response mechanisms, whereas visual stimuli (e.g., display colors) have to be translated into their corresponding verbal representations in order to influence response mechanisms. This requisite translation results in the color–word asymmetry (i.e., words interfere with color naming, but colors do not interfere with word naming) typically found in Stroop experiments (Virzi & Egeth, 1985). If no translation were required, one would not expect to find this asymmetry (Durgin, 2003). The detect task may present such a situation, in that participants are hypothesized to match an internal visual representation of the target color with the visual representation of the display color, all without need for translation. If this is the case, the elimination of Stroop interference in the detect task would be due to the absence of response translation. In addition to the lack of response translation, another important factor in the elimination of Stroop interference in the detect task might be the lack of irrelevant stimulus–response overlap. Irrelevant stimulus–response overlap has been shown to reduce, but not eliminate, Stroop interference in the classify version of the task (Klein, 1964). A multimechanism account, positing roles for both the elimination of translation and irrelevant

123

stimulus–response overlap, would be the most prudent account of the present results. Do words automatically activate memory representations? The present results are incompatible with the hypothesis that an irrelevant word presented in the visual field automatically activates its representation in long-term memory. Instead, our results, in conjunction with previous work (e.g., Kahneman & Henik, 1981), suggest that the processing of an irrelevant word is contingent on various experimental factors. When conditions are favorable for the processing of irrelevant information (e.g., spatial certainty, low attentional load, irrelevant stimulus–response overlap, and the need for translation), visual word recognition may appear obligatory and ballistic. When examined under a different set of conditions, however, (e.g., spatial uncertainty, high attentional load, no irrelevant stimulus–response overlap, and no requisite translation), the limits of visual word recognition are readily apparent. Similar arguments have been put forward with regard to failures of selective attention and the early versus late selection debate. Yantis and Johnston (1990) argued that Failures of selective attention may be attributed to suboptimal conditions for eliciting such selection. When conditions are more conducive to effective selection, an otherwise latent ability to selectively attend to spatial locations in a sharply focused manner is revealed. (p. 146)

Yantis and Johnston further stated that failures of selective attention may be a product of task parameters and may not reflect the participant’s innate inability to selectively attend. The characterization of these suboptimal conditions is currently unclear (Miller, 1991); however, we have delineated some here (e.g., spatial certainty, attentional load, irrelevant stimulus–response overlap, and translation). Kahneman and Henik (1981) suggested three factors that determined whether a representation in long-term memory was activated: (1) the quality of the sensory information; (2) the priming of the relevant memory node by set or expectancy; and (3) the availability of an added enabling or facilitating input-attention. (p. 208)

The present results, therefore, fit well into the theoretical frameworks offered by Yantis and Johnston and by Kahneman and Henik. An alternative account? Brown (1996) viewed selection for action as the crucial factor in determining whether spatial attention is required for visual word recognition. According to his account, if the word is selected for action (i.e., is directly relevant to the response required by the participant), visual word recognition is delayed until the word is spatially attended. In contrast, if the word is irrelevant to the task, visual word recognition proceeds in the absence of spatial attention. The results of Experiment 1 suggest that Brown’s claim needs modification. Specifically, in Experiment 1, words were not selected for action, since display colors were the relevant targets. Yet when the irrelevant words were not at the focus of at-

124


tention (i.e., in the separated condition), they did not interfere with color identification, as would be predicted by Brown’s account. It remains likely, however, that selection for action is one of many important factors in determining whether an irrelevant word is processed. CONCLUSION The results of the present experiments are inconsistent with the widespread claim that visual word recognition is obligatory and ballistic. Instead, they point to the need for theorists to consider the roles of spatial attention, response demands, and attentional load, among other factors, when discussing the retrieval of word knowledge from long-term memory. REFERENCES Bauer, B., & Besner, D. (1997). Processing in the Stroop task: Mental set as a determinant of performance. Canadian Journal of Experimental Psychology, 51, 61-68. Brown, T. L. (1996). Attentional selection and word processing in Stroop and word search tasks: The role of selection for action. American Journal of Psychology, 109, 265-286. Brown, T. L., Gore, C. L., & Carr, T. H. (2002). Visual attention and word recognition in Stroop color naming: Is word recognition “automatic”? Journal of Experimental Psychology: General, 131, 220-240. Durgin, F. H. (2003). Translation and competition among internal representations in a reverse Stroop effect. Perception & Psychophysics, 65, 367-378. Jolicœur, P. (1999). Concurrent response-selection demands modulate the attentional blink. Journal of Experimental Psychology: Human Perception & Performance, 25, 1097-1113. Kahneman, D., & Chajczyk, D. (1983). Tests of the automaticity of reading: Dilution of Stroop effects by color-irrelevant stimuli. Journal of Experimental Psychology: Human Perception & Performance, 9, 497-509. Kahneman, D., & Henik, A. (1981). Perceptual organization and attention. In M. Kubovy & J. R. Pomerantz (Eds.), Perceptual organization (pp. 181-211). Hillsdale, NJ: Erlbaum. Klein, G. S. (1964). Semantic power measured through the interference of words with color-naming. American Journal of Psychology, 77, 576-588. Lavie, N. (1995). Perceptual load as a necessary condition for selective attention. Journal of Experimental Psychology: Human Perception & Performance, 21, 451-468.

MacLeod, C. M. (1991). Half a century of research on the Stroop task: An integrative review. Psychological Bulletin, 109, 163-203. Miller, J. (1991). The flanker compatibility effect as a function of visual angle, attentional focus, visual transients, and perceptual load: A search for boundary conditions. Perception & Psychophysics, 49, 270-288. Rayner, K., & Pollatsek, A. (1989). The psychology of reading. Englewood Cliffs, NJ: Prentice-Hall. Sharma, D., & McKenna, F. P. (1998). Differential components of the manual and vocal Stroop tasks. Memory & Cognition, 26, 1033-1040. Starreveld, P. A., Theeuwes, J., & Mortier, K. (2004). Response selection in visual search: The influence of response compatibility of nontargets. Journal of Experimental Psychology: Human Perception & Performance, 30, 56-78. Stolz, J. A., & McCann, R. S. (2000). Visual word recognition: Reattending to the role of spatial attention. Journal of Experimental Psychology: Human Perception & Performance, 26, 1320-1331. Stroop, J. R. (1935). Studies of interference in serial and verbal reactions. Journal of Experimental Psychology, 18, 643-662. Treisman, A. M., & Gelade, G. (1980). A feature integration theory of attention. Cognitive Psychology, 12, 97-136. Treisman, A. M., & Gormican, S. (1988). Feature analysis in early vision: Evidence from search asymmetries. Psychological Review, 95, 15-48. Van Selst, M., & Jolicœur, P. (1994). A solution to the effect of sample size on outlier elimination. Quarterly Journal of Experimental Psychology, 47A, 631-650. Virzi, R. A., & Egeth, H. E. (1985). Toward a translational model of Stroop interference. Memory & Cognition, 13, 304-319. Wolfe, J. M. (1998). What can 1 million trials tell us about visual search? Psychological Science, 9, 33-39. Yantis, S., & Johnston, J. C. (1990). On the locus of visual selection: Evidence from focused attention tasks. Journal of Experimental Psychology: Human Perception & Performance, 16, 135-149. NOTE 1. Stroop effects tend to be larger with vocal responses, and some readers might suppose that manual responses are not sensitive enough to detect a Stroop effect. The size of the Stroop effect in the integrated condition in Experiment 1 (110 msec) suggests that the response type employed in the present investigation was sensitive enough, however. Interestingly, Sharma and McKenna (1998) found Stroop effects of approximately the same magnitude with a 4AD manual and a 4AD vocal Stroop task, if one ignores the letter condition.

(Manuscript received September 22, 2003; revision accepted for publication May 7, 2004.)