Intention and performance when reading aloud: Context is everything

https://doi.org/10.1016/j.concog.2021.103211Get rights and content

Highlights

  • The “intention free” account of single word identification (the automatic perspective) is critically questioned.

  • The relation between intention and visual word identification is more nuanced than understood to date.

  • Using a Go/No-Go paradigm, we conducted two studies in which the proportion of GO trials varied.

  • Stimulus quality was the manipulated factor.

  • Results demonstrates when processing with prior intent occurs, and when it does not need prior intent.

  • We concluded that there is no fixed relation between intention and word identification.

Abstract

A widely held account asserts that single words are automatically identified in the absence of an intent to process them in the form of identifying a task set, and implementing it. We provide novel evidence that there is no fixed relation between intention and visual word identification. Subjects were randomly cued on a trial-by-trial basis as to whether to read aloud a single target word (Go) or not (No-go). When the Go-No Go probability was 50% (Experiment 1) the effect of stimulus quality (bright vs. dim targets) was the same size as in a separate block of 100% Go trials. In Experiment 2, where the Go-No Go probability was 80% in the cued condition, the stimulus quality effect was smaller than in the block of all Go trials. These results can be understood in terms of Go trial probability moderating whether subjects (i) hold off beginning to process the target until an intention in the form of a Task Set has been implemented, or (ii) begin to identify the target during the time taken to implement a Task Set. The additivity of stimulus quality and cueing conditions in Experiment 1 support the view that target processing only begins when a Task Set is in place, whereas the under-additivity of stimulus quality and cueing condition in Experiment 2 supports the interpretation that target identification can start during the time that a Task Set is being implemented. Taken together with other results, we conclude that there is no fixed relation between an intention and word identification; context is everything.

Introduction

It is widely believed that the identification of single, visually presented words is “automatic” in the sense that it requires no form of attention, is fast, unconscious, ballistic, and cannot be interfered with by other processes (e.g., Posner and Snyder, 1975, Neely and Kahan, 2001; Brown, Gore, & Carr, 2002, among many others). Although there is a large literature on the intersection of “automaticity” and visual word identification (e.g., see Besner et al., 2016 for a brief and selective review), there has been relatively little attention paid to the role of intent in the context of explicit word identification (implicit word identification, as in the case of many variants of Stroop’s paradigm, is not considered here). The purpose of the present research is not to determine whether or not visual word recognition is fully automatic. Instead, it is to determine whether an intention in the form of a task set must be in place before a word can begin to be identified, or whether such identification can begin during the time that the appropriate task set is identified, and implemented.

If various aspects of visual word identification (e.g., feature level activation, letter level activation, word level activation, phonological lexical activation, sub-lexical translation of orthography to phonology, and semantic activation) occur without any form of intention, as some have argued (e.g., Brown et al., 2002, among others), then what is the role of intention? One role is to prepare a task set so that subjects can carry out the experimenter’s instructions (e.g., read the word aloud, categorize it semantically, press a button, etc). However, in the vast majority of experiments the task set is typically held constant in that the same task is done on every trial. This makes it difficult to explore the relation between an intention (deciding what the task is, and implementing a task set to do it), and explicit word identification. One way to examine this issue is to make the task set vary unpredictably across trials rather than being held constant. We provide some background to the use of the present paradigm by first considering a paradigm in which, typically, the target task is fixed.

The logic underlying the present experiments is very similar to that used in the Psychological Refractory Period (PRP) paradigm. Hence, because that literature has deeper roots, we start there (see Pashler, 1984, Pashler, 1994 for a review).

Subjects in the PRP paradigm typically perform two tasks in response to sequentially presented stimuli (S1 and S2) with the instruction to respond to Task 1 before responding to Task 2. Under these conditions reaction time (RT) to S2 increases substantially (by several hundred ms) as the stimulus onset asynchrony (SOA) between S1 and S2 decreases. Pashler (1984) developed a seminal analysis of performance in this paradigm that has had a major impact on how researchers think about the resource demands of mental processing.

According to Pashler’s “cognitive slack” logic, the need for the same limited-capacity process at some point (hereafter referred to as “central” attention following Johnston, McCann, & Remington, 1995) by both tasks has straightforward consequences for Task 2 processes that occur before, during, or after this bottleneck. For example, when Task 1 and Task 2 overlap temporally (the shorter the SOA between the presentation of Tasks 1 and 2, the more temporal overlap there is), and participants are instructed to respond to Task 1 before Task 2, Task 1 typically gains access to central attention before Task 2 (see Fig. 1, Fig. 2). Task 2 processes that require central attention are therefore functionally postponed until that resource becomes available. If the effect of a factor manipulated in Task 2 occurs prior to the processing bottleneck (i.e., does not need limited capacity central processing), the effect of that factor will be absorbed into the slack created by other Task 2 processes waiting for central attention to become available. That is, a factor that affects processing prior to central attention will produce a smaller RT cost at the short SOA than the longer SOA because it makes no use of central attention. This pattern is commonly described as “under-additive” (of a factor in combination with decreasing SOA; also referred to as some or all of the effect of that factor having been “absorbed into slack”). In contrast, if a factor manipulated in Task 2 has additive effects with SOA on RT, then this factor affects a process that either (a) uses central attention or (b) occurs after central attention has been used by some other process.

This logic is the normative interpretation of under-additivity and additivity of factor effects in combination with decreasing SOA in the PRP paradigm (Pashler, 1984, Pashler, 1994). Theoretical variants to this logic exist in which some processes share capacity between Tasks 1 and 2 rather than an all-or none bottleneck (Navon & Miller, 2002; Tombu & Jolicœur, 2002), but the central inference in which under-additivity implies capacity free processing still applies (i.e., no interference from Task 1 with regard to the factor that yields under-additivity with decreasing SOA).1 The diagram in Fig. 1 illustrates an under-additive interaction of a factor with decreasing SOA. Fig. 2 illustrates additive effects of a factor with decreasing SOA.

For present purposes the critical result is that the effect of stimulus quality (SQ) in a range of different tasks is under-additive with decreasing SOA (Pashler, 1984; Oriet & Jolicœur, 2003). The straight-forward inference is that the mental processing needed to deal with the effect of low SQ in Task 2 can proceed during the time that the subject is engaged with processing in Task 1 because that processing (for the effect of SQ) makes no demands on the central processing needed for Task 1.

For purposes of illustration, consider Pashler’s (1984) first experiment. Subjects either performed a visual search task by itself or when combined with a spatial localization task (this can be thought of as an SOA manipulation; doing the task by itself is akin to a long SOA, whereas doing both tasks is like a short SOA between Task 1 and Task 2). In the latter case subjects were instructed to respond to Task 1 before responding to Task 2.

The critical results were threefold. First, when having to respond to both tasks, RT to the visual search task was much slower than when this same task was presented in isolation as subjects were required to respond to Task 2 after they had responded to Task 1 (Task 2 slowing when the tasks overlap in time). Second, when subjects did both tasks the effect of stimulus quality in the visual search task was significantly smaller as compared to when subjects performed the visual search task in isolation. That is, the effect of stimulus quality was under-additive with one vs two tasks (i.e., SOA). Thirdly, the time to indicate whether a target was present or absent in the visual search task produced a large difference in RT, but this difference was the of the same magnitude when that task was done alone, or in combination with the other task (i.e., despite overall slowing when the two tasks were done compared to when only one task was done). This can be understood as evidence that a response selection stage in the visual search task was bottlenecked (postponed) by having to respond to the spatial localization task first because the spatial localization task consumed a resource that was needed for the component of the visual search task indexed by the presence/absent judgment.

Part of the elegance of Pashler’s experiment is that both under-additivity and additivity is seen in the same task and in the same experiment. In essence, the under-additivity (of stimulus quality) can be understood as reflecting the operation of an early process in the visual search task that makes no demands on an attention demanding process required by the spatial localization task, whereas the additivity (of indicating the presence vs absence of a target) implies that a later process in this visual search task is postponed (bottlenecked) because it needs resources being used in the processing of the spatial localization task.

Besner and Care (2003) modified the PRP paradigm and extended its logic so as to address the issue of where in the processing stream the intention to perform a specific task occurs. The effect of SQ is again of central concern. Besner and Care presented subjects with a target, and had them carry out one of two possible tasks on a trial. Which task was to be done was cued, for example, by a color. One colour instructed the subject to do one of the two tasks, and a different color to do the other task. Besner and Care randomly cued subjects to either read aloud a nonword (e.g., “frane”) or categorize it as appearing in upper or lower case via a key press. The task cue occurred either at the same time as the onset of the target (SOA = 0 ms) or 750 ms before it (SOA = -750 ms). In the latter case the reader knows what the task is before the target appears, allowing them time to do some preparation for the required task in advance of the target. The size of the SQ effect in the 750 ms SOA condition therefore provides a baseline against which the size of the SQ effect in the 0 ms SOA condition can be compared. In short, the target task varies across trials whereas it is typically fixed in PRP in that there is only one target task. And, in Besner and Care’s hands, and others, there is no explicit response to the task cue, whereas in PRP there is an overt response to Task 1.

Besner and Care assumed that at least part of the SQ effect reflects early processing (e.g., at the featural and letter levels) and argued that if the reader could mitigate the detrimental effect of low SQ during the time that they were interpreting the task cue at the 0 ms SOA, then there should be an under-additive SQ × SOA interaction with decreasing SOA such that the SQ effect is smaller than in the −750 SOA baseline condition. This should be so because in the baseline condition the task is known to the subject before the target appears and thus there would be no task-set preparation time to absorb the lengthening of RTs produced by low SQ.

This task-set procedure thus provides a test of whether a given process can be carried out in the absence of an intention in the form of a task set being in place. If the process(es) affected by SQ do not first require a task set to be in place, then there should be an under-additive interaction with decreasing SOA (we already know that the process(es) that can deal with low SQ is not capacity demanding given previous PRP research with various other tasks as noted above in Pashler (1984), and Oriet and Jolicœur (2003).

The results of Besner and Care (2003) provide no support for the claim that the processes responsible for dealing with the effects of SQ can begin in the absence of a task set being in place: SOA and SQ had additive effects in both the read aloud and case identification tasks. Besner and Care therefore concluded that, at least for their two tasks, processing of the cue and the instantiation of a task set (an intention) preceded any processing of the target that is affected by SQ. Kahan, Hengen and Mathis (2011, Experiment 1), replicated Besner and Care’s findings of additive effects of SQ and SOA when the targets were again all nonwords and the nature of the cue along with the two tasks were different, implying some generality.

An important qualification to the additive effects of SQ and SOA that both Besner and Care, 2003, Kahan et al., 2011 obtained with nonwords is that when Paulitzki et al., 2009, Kahan et al., 2011 did further experiments in which the targets were all words, and the same tasks that had previously been used, SQ and SOA now produced the previously defined under-additive interaction. In summary, there is unambiguous support for the claim that the nature of the target (words vs. nonwords) influences when SOA moderates the effects of SQ in this Task-Set paradigm, and when it does not. When only nonwords are used as targets, SQ and SOA have additive effects, whereas when only words are used, SQ has an under-additive effect with decreasing SOA.2

The present experiments sought to further investigate possible constraints as to when absorption into slack is observed. The effect of SQ is again a central manipulation of interest. The task(s) consisted of a Go vs No-Go procedure, one that has not been used before in the context of the stimuli addressed here (but see Besner and Risko (2005) for experiments in which this procedure was applied to a spatial localization task and included a SQ manipulation).

To summarize the main elements of Experiment 1, there were two blocks of trials. In the first block 50% of the trials were Go trials and 50% were No-Go trials (critically, 50% Go trials were used because that was the ratio of one task to another in all of the task set experiments noted earlier, and, as we will see in due course, probability turns out to be important). Subjects read aloud a single high frequency word on a random half of the trials and withheld a response on the remaining trials. Whether subjects read the word aloud or not was cued by a brief tone appearing 25 ms before the word target was presented. In the second, baseline block, subjects read the word aloud on every trial (100% Go trials). We assume that the appropriate set is in place prior to encountering the target word in the baseline block because the task set is to read the target word aloud on every trial. We thus consider the 100%-Go block to be functionally equivalent to a long SOA as used in previous work. The target word was equally often bright/dim (the SQ manipulation) on a randomly determined basis.

If absorption into slack is determined only by the use of words as targets, as seen in the previous results reported by Paulitzki et al. (2009) and Kahan et al. (2011), there should, again, be an under-additive Go/No-Go vs. baseline × SQ interaction (a smaller effect of SQ in the cued condition than in the baseline condition). In contrast, if the nature of the tasks moderates whether additivity or under-additivity is obtained, then the effects of SQ and condition (50% Go/50% No-Go vs 100% Go [baseline]) may be additive.

To anticipate the results of Experiment 1, there was a large main effect of SQ in the 50% Go/50% No-Go condition, and there was a large main effect of Cueing condition. Critically, the effect of SQ in the Cued condition was of the same magnitude as that seen in the 100% Go (baseline) condition in which the subjects read the word aloud on every trial. These results demonstrate that the use of words as targets is not a sufficient condition for absorption into slack to be seen.

The results of Experiment 2 provide further clarification in that they show that the results of Experiment 1 were likely due to a bottleneck produced by an unconscious strategy. Experiment 2 was identical to Experiment 1 except that the Go probability in the cued block was 80%. This experiment yielded an under-additive interaction of SQ and cued vs baseline conditions, showing that it is neither the nature of the task on its own, nor the nature of the targets on its own, but rather the combination of task type, target type, and Go probability that determines whether absorption into slack is observed or not. We take these results up further in the General Discussion.

Section snippets

Subjects

Twenty University of Waterloo undergraduates were recruited either through the SONA research experience group who participated in exchange for course credit, or the PSYC pool listing of volunteer subjects who each received $10.00 financial remuneration. All subjects had English as a first language and had normal or corrected to normal vision.

Stimuli

Our method for combining items and subjects in one analysis involved each subject seeing a unique stimulus set that was sampled from 480 monosyllabic, high

Results

Incorrect responses (0.5%) and spoiled trials (3.3%) were first removed. The remaining RT data were subjected to an outlier analysis in which RTs falling 3 standard deviations above or below the mean correct RT for each subject in each of the 4 conditions considered on its own were discarded. This analysis resulted in the removal of 1.6% of the correct RT data. The resulting mean RT and percentage error means can be seen in Table 1.

Experiment 2

In previous experiments with the task set paradigm, the use of words as targets was associated with an under-additive interaction between SQ and decreasing SOA (Paulitzki et al., 2009; Kahan et al, 2011). Nevertheless, the present Experiment 1 yielded no evidence of an under-additive interaction despite the use of words as targets. Experiment 2 assessed the hypothesis that the 50/50, Go/No-Go probability was driving the failure to see under-additivity in Experiment 1. Experiment 2 therefore

General discussion

Absorption into slack logic pioneered by Pashler, 1984, Pashler, 1994 was adapted for use in Task set experiments by Besner and Care (2003). Paulitzki et al. (2009, Experiment 1) presented a word target and an auditory task-set cue that indicated whether the subject was to pronounce the word target aloud or to press a key indicating whether it was in upper or lower case. The SOA between the task-set cue and word target was either 0 ms or 750 ms. The logic was that if early word-target

Conclusions

Rather than the claim that word identification is “automatic” in the sense that words are always identified without the need for an intention to create and implement a task set appropriate to the task at hand first be in place, some form of context-dependent account is called for. For psycholinguistic processing, this account, minimally, would need to include considerations regarding both the nature of the stimulus (e.g., words vs. nonwords) and the nature of the tasks that compose the task

CRediT authorship contribution statement

Derek Besner: Conceptualization, Funding acquisition, Methodology, Writing – original draft, Writing – review & editing. David McLean: Writing – original draft, Writing – review & editing, Data curation, Formal analysis, Software, Project administration. Torin Young: Validation, Formal analysis, Investigation, Visualization. Evan Risko: Conceptualization, Validation, Writing – review & editing.

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Acknowledgment

This work was supported by a Discovery Grant from the Natural Sciences and Engineering Research Council of Canada to DB. Data collection conforms to the Helsinki Accord. The data will be made available on Open Science Framework servers upon publication. We thank the seven independent reviewers for their comments, but even more so Jim Neely.

References (29)

  • D. Navon et al.

    Queuing or sharing? A critical evaluation of the single-bottleneck notion

    Cognitive Psychology

    (2002)
  • S. O’Malley et al.

    Lexical processing while deciding what task to perform: Reading aloud in the context of the task set paradigm

    Consciousness and Cognition

    (2011)
  • J.R. Paulitzki et al.

    On the role of set when reading aloud: A dissociation between prelexical and lexical processing

    Consciousness and Cognition

    (2009)
  • D.A. Balota et al.

    Attentional control of lexical processing during word recognition and reading

  • D.A. Balota et al.

    Moving beyond the mean in studies of mental chronometry: The power of response time distributional analyses

    Current Directions in Psychological Science

    (2011)
  • B. Baluch et al.

    Visual word recognition: Evidence for strategic control of lexical and non-lexical routines in oral reading

    Journal of Experimental Psychology: Learning, Memory, and Cognition

    (1991)
  • D. Besner

    The myth of ballistic processing: Evidence from Stroop’s paradigm

    Psychonomic Bulletin & Review

    (2001)
  • D. Besner et al.

    A paradigm for exploring what the mind does while deciding what it should do

    Canadian Journal of Experimental Psychology

    (2003)
  • D. Besner et al.

    When under-additivity of factor effects in the Psychological Refractory Period paradigm implies a bottleneck: Evidence from psycholinguistics

    Quarterly Journal of Experimental Psychology

    (2009)
  • D. Besner et al.

    Stimulus-response compatible orienting and the effect of anaction not taken: Perception delayed is automaticity denied

    Psychonomic Bulletin & Review

    (2005)
  • D. Besner et al.

    Varieties of attention: Their roles in visual word identification

    Current Directions in Psychological Science

    (2016)
  • T.L. Brown et al.

    Visual attention and word recognition in Stroop color naming: Is word recognition“ automatic?”

    Journal of Experimental Psychology: General

    (2002)
  • H. Elchlepp et al.

    A change of task prolongs early processes: Evidence from ERPs in lexical tasks

    Journal of Experimental Psychology: General

    (2015)
  • J.C. Johnston et al.

    Chronometric evidence for two types of attention

    Psychological Science

    (1995)
  • Cited by (1)

    View full text