Holding a manual response sequence in memory can disrupt vocal responses that share semantic features with the manual response
- First Online:
- Cite this article as:
- Fournier, L.R., Wiediger, M.D., McMeans, R. et al. Psychological Research (2010) 74: 359. doi:10.1007/s00426-009-0256-9
- 69 Views
Holding an action plan in memory for later execution can delay execution of another action if the actions share a similar (compatible) feature. This compatibility interference (CI) occurs for actions that share the same response modality (e.g., manual response). We investigated whether CI can generalize to actions that utilize different response modalities (manual and vocal). In three experiments, participants planned and withheld a sequence of key-presses with the left- or right-hand based on the visual identity of the first stimulus, and then immediately executed a speeded, vocal response (‘left’ or ‘right’) to a second visual stimulus. The vocal response was based on discriminating stimulus color (Experiment 1), reading a written word (Experiment 2), or reporting the antonym of a written word (Experiment 3). Results showed that CI occurred when the manual response hand (e.g., left) was compatible with the identity of the vocal response (e.g., ‘left’) in Experiment 1 and 3, but not in Experiment 2. This suggests that partial overlap of semantic codes is sufficient to obtain CI unless the intervening action can be accessed automatically (Experiment 2). These findings are consistent with the code occupation hypothesis and the general framework of the theory of event coding (Behav Brain Sci 24:849–878, 2001a; Behav Brain Sci 24:910–937, 2001b).
Everyday actions such as turning on an appliance, starting a car, and programming a television remote require the production of action plans. An action plan is a set of muscle commands that are structured before a movement sequence begins that allows the entire sequence to be carried out (Keele, Cohen, & Ivry, 1990). According to Jeannerod (1997), action planning requires that the appropriate motor schemas are selected, are related to the proper internal and external cues, and are organized into the appropriate sequence. Sometimes we have to momentarily suspend the execution of one action plan in order to execute another. For example, as a pilot is about to execute a planned sequence of manual responses to change the course of an airplane, this action may be interrupted due to the detection of an engine malfunction which requires rapid, verbal communication to the co-pilot. We were interested in whether execution of a vocal response can be affected by an action plan held in memory that represents a manual response sequence. Recent research indicates that the execution of an action, at least in some cases, can be adversely affected by a planned action held in working memory.
For example, Stoet and Hommel (1999) showed that withholding a manual response for later execution can sometimes delay the execution of another manual action. In their study, two different visual stimuli were presented sequentially. Participants were instructed to plan and withhold a sequence of key-presses with either their right or left hand based on the identity of the first stimulus (Stimulus A). While withholding the action plan to Stimulus A, a second visual stimulus appeared (Stimulus B). An immediate key-press response with either the right or left hand was required based on the identity of Stimulus B. After executing a speeded response to Stimulus B, the planned action to Stimulus A was executed. Results showed that the response to Stimulus B was longer when it required the same (compatible) hand as Stimulus A as opposed to a different (not compatible) hand. This delay in executing an action to the intervening stimulus (Stimulus B) when it shares a feature with the action sequence currently being withheld is referred to as compatibility interference (CI).
To explain CI, Stoet and Hommel (1999, 2002) proposed the code occupation hypothesis (COH), which is a central assumption of the theory of event coding (Hommel, Müsseler, Aschersleben & Prinz, 2001a, b). The COH assumes that intentional actions are cognitively represented in terms of the proximal and distal effects that this action produces; and hence these action-effect codes contain both our internal actions and the anticipated or perceived effects in the environment (Hommel et al., 2001a; Müsseler & Hommel 1997a). Based on this assumption, action and perceptual features share a common representational domain (Prinz, 1990, 1997) which is consistent with the existence of mirror neurons (e.g., Gallese, Fadiga, Fogassi, & Rizzolatti, 1996) or visual-and-motor neurons (e.g., Sakata, Taira, Murata, & Mine, 1995; Taira, Mine, Georgopoulos, Mutara, & Sakata, 1990). An action feature can be a movement direction (e.g., a movement to the right) or the side of the body that the effectors used to carry out an action are located (e.g., a movement with the right-hand). A perceptual feature can be the orientation or movement direction of a stimulus (e.g., a right-pointing arrow, a stimulus moving to the right). To specifically account for CI, the COH assumes that all features (perceptual and action features) controlling the planned action are bound together into a common representation (action plan) that occupies the codes that represent it. This occupation of codes causes these codes to become temporarily less available for other actions or perceptions until the action plan is executed or abandoned.
Thus, the COH predicts that a response will be delayed if there is partial, feature code overlap with the action plan held in memory due to code occupation. For example, if an action sequence is planned that requires the right-hand, a “right” code will be occupied, delaying any additional actions that require the “right” code (e.g., the right-hand) until the planned action is performed or released which was the result obtained by Stoet and Hommel (1999; see also Mattson & Fournier, 2008; Wiediger & Fournier, 2008). In accordance with the COH, Stoet and Hommel found that CI occurred only if a sequence of actions was planned in advance of the stimulus that required an immediate action. In addition, they found CI when the effectors required for the two responses shared the same side of the body (e.g. left hand and left foot which share the left code), although the effect was not as large compared to when only one effector (e.g. left hand) was used for both responses. These findings suggest that it is not effector overlap, but an overlap in the feature codes (e.g., right or left) that make up the action plan that causes CI. For other evidence that CI is due to feature overlap, see research by Müsseler and colleagues which show a form of perceptual interference based on partial feature overlap between actions and perceptual information (e.g., Müsseler & Hommel, 1997a, b; Müsseler, Steininger, & Wühr, 2000; Müsseler & Wühr, 2002; Müsseler, Wühr, & Prinz, 2000; Wühr & Müsseler, 2001).
Although response delays due to CI have been attributed to partial feature overlap, as opposed to sensori-motor overlap, they have been observed only under conditions in which the withheld and intervening actions shared the same response modality—either manual or vocal. That is, CI has been shown for various manual motor responses (Stoet & Hommel, 1999; Mattson & Fournier, 2008; Wiediger & Fournier, 2008), and CI has also been shown for vocal motor responses where the withheld and intervening action shared phonetic features (Meyer & Gordon, 1985). In addition, partial feature overlap leading to CI has only been observed when the immediate and withheld actions shared the same hemisphere (e.g., left-hemisphere overlap in executing “right”-hand and “right”-foot actions; Stoet & Hommel, 1999). However, according to the COH, CI is due to partial feature overlap between events in terms of their cognitive representations, which can be defined as abstract, high-level, semantic codes (or meaning-based codes) that describe the internal or anticipated actions and perceptions of an event. Hence, CI should not be restricted to events that share the same response modality or same hemisphere. If partial overlap of semantic codes is sufficient to obtain CI as assumed by the COH, then CI should be observed between manual and vocal responses and should be observed regardless of whether these responses share the same hemisphere.
In contrast to this assumption, Hommel and Müsseler (2006) showed that partial feature overlap between a withheld action (e.g., manual “left”-hand response or vocal response “left”) and the perceptual characteristics of an intervening visual stimulus (e.g., a masked “left”-pointing arrow) hindered delayed-recall accuracy in identifying the intervening stimulus (e.g., a left-hand key-press indicating recall of a “left”-pointing arrow), but only when the response used to indicate the identity of the intervening stimulus was the same type of response (manual or vocal) as that required by the withheld action. Based on these findings, Hommel and Müsseler argued that perceptual delays leading to poorer accuracy in identifying the masked stimulus when it shared a feature with the withheld action was not due to semantic overlap alone. They suggested that the cognitive coding scheme proposed by the COH may be restricted to feature overlap between events that share the same response modality.
However, there is an alternative explanation of the Hommel and Müsseler (2006) results. Stoet and Hommel (1999) showed that CI effects were larger when more features partly overlapped between the withheld and intervening action events. For example, when the withheld action and the intervening action both required “left hand” responses (and hence shared two features “left” and “hand”), CI was greater (i.e., 33 vs. 12 ms) than when the withheld and intervening actions required left hand and left foot responses, respectively, (and hence shared only one feature “left”). It is possible that the recall accuracy measure used by Müsseler and Hommel was not sensitive enough to detect small delays in identifying the masked stimulus when the masked stimulus only shared one feature (e.g., “left”) as opposed to two features (e.g., “left” and “hand” or “vocalization”) with the prior action event.
Although CI effects have not been found between manual and vocal response modalities, dual task studies have shown that executing limb movements and speech simultaneously hinders performance relative to when these actions are performed alone (Pashler, 1990). Also, some studies have shown that dual task interference between movement and speech can be exacerbated when tasks are performed by the same hemisphere compared to different hemispheres (e.g., Kinsbourne & Cook, 1971). However, this hemispheric bias tends to disappear in choice reaction time tasks (Pashler & Christian, 1994; see review by Pashler & Obrien, 1993) which suggests that interference between manual responses and vocal responses can occur at a central bottleneck related to response selection as opposed to hemispheric overlap related to response execution (see also evidence by Mattson & Fournier, 2008).
Furthermore, there is evidence by Wühr (2006) that executing a manual and vocal responses simultaneously in a choice reaction time task can lead to greater dual task interference when there is semantic overlap between the manual and vocal responses than when there is no semantic overlap. He had participants vocally respond ‘left’ or ‘right’ as quickly and accurately as possible based on the color of a stimulus (red or green). Prior to the presentation of the colored stimulus a written message ‘left’ or ‘right’ appeared which instructed participants to hold down a key with either their left- or right-hand, and to continue this action during presentation and response to the colored stimulus. Results showed that ongoing execution of a manual key-press with the left- or right-hand delayed the vocal response of ‘left’ or ‘right’, respectively, (i.e., when the manual and vocal responses shared a semantic code—‘left’ or ‘right’). This finding is consistent with predictions of the COH. However, because a written message ‘left’ or ‘right’ preceded the manual response to verbally indicate which response hand to use, it is not clear whether the code representing the manual response or the written message interfered with the code-compatible vocal response—that is, phonetic overlap between the written message and vocal response could have led to the vocal response delays (e.g., Meyer and Gordon, 1985). Also, because the manual response was executed prior to and during the vocal response, these results may not generalize to situations in which the manual response is withheld in working memory during vocal response execution. Thus, it remains unclear whether CI will occur between planned actions held in working memory and intervening actions that share a common semantic code, but utilize different response modalities (e.g., manual and vocal).
Determining whether CI can generalize to actions that require different response modalities is important because it evaluates one of the key assumptions of the COH, that action delays attributed to CI are due to partial, cognitive code overlap (Hommel et al., 2001a). In particular, it addresses whether CI requires that the cognitive codes generated for two different actions must be associated with the same modality of action features (manual or vocal) or whether the cognitive codes generated can consist of more abstract, semantic representations of the action. If partial semantic overlap between the withheld and intervening actions is sufficient for CI, this would suggest that abstract, high-level semantic codes of a relevant action event get bound together, and hence these semantic codes become temporarily unavailable to the intervening event, resulting in a delay in executing the intervening action.
Also, because of its reliance on cognitive representations and the binding of these representations for action planning, the COH further assumes that CI is restricted to off-line action planning and to preparation of actions that have arbitrary response associations to stimulus events that require short-term or working memory (Hommel et al., 2001a, b). Thus, CI should not be observed for actions that are over-learned or automatic and hence do not impose a capacity demand on short-term or working memory (Hommel et al., 2001a, b). The only evidence to partly support this assumption is provided by Wiediger and Fournier (2008). They showed that a touch response to a stimulus location with the right- or left-hand was delayed when this action shared a feature with the withheld action (a sequence of key-presses with the right- or left-hand), but only when the selection of the response hand (left or right) for the touch response was arbitrarily determined (e.g., by stimulus color). Right-hand touch responses by right-handers to the location of a single transient stimulus occurring to the right-side of body center did not result in CI. These findings suggest that responses that require cognitive oversight or working memory to remember which response goes with which stimulus characteristics suffer if the code required to carry out this response is occupied by another action currently held in working memory. However, actions more naturally associated with the stimulus event can be carried out automatically without much cognitive oversight (e.g., a right-handers reach to a transient with the right-hand on the right-side of body center), and hence would not require access to codes associated with another action held in working memory. While these findings suggest that CI may be restricted to intervening actions that have arbitrary stimulus–response associations and hence impose a demand on working memory, there is no empirical evidence suggesting that the same conclusion holds when the withheld and intervening actions involve different response modalities.
The purpose of the present study was to determine whether execution delays attributed to CI can be based on semantic overlap alone and whether CI is reduced (or eliminated) for intervening events that can activate responses automatically, without relying on a stored representation of stimulus–response associations in short-term or working memory. The Stoet and Hommel (1999) paradigm described earlier was used in three different experiments to determine the conditions under which an intervening vocal response (‘left’ or ‘right’) could be delayed by a manual response sequence (with the ‘left’-hand or ‘right’-hand) held in memory when both responses shared a common semantic code (‘left’ or ‘right’).
In each experiment, two different visual stimuli were presented sequentially. A sequence of three manual key-presses with the left- or right-hand was planned and withheld based on the visual identity of the first stimulus, and a speeded vocal response (‘left’ or ‘right’) was made based on the visual identity of the second, intervening stimulus. In Experiment 1, the vocal response ‘left’ or ‘right’ was based on discriminating the color (e.g., red or green) of a stimulus. In Experiment 2, the vocal response ‘left’ or ‘right’ was based on identifying the stimulus word ‘left’ or ‘right’. In Experiment 3, the vocal response ‘left’ or ‘right’ was based on identifying the antonym of the stimulus word ‘left’ or ‘right’. Results showed that CI was obtained in Experiment 1 and 3, but not in Experiment 2. Also, CI was not influenced by hemispheric overlap between the manual and vocal responses. These findings suggest that partial overlap in terms of semantic coding is sufficient to obtain CI for events that require very different actions. However, if the intervening action can be accessed automatically, CI may not occur. These findings are consistent with the assumptions of the COH and the general framework of the Theory of Event Coding (Hommel et al., 2001a).
This experiment examined whether a vocal response could be delayed by a manual response sequence held in memory when the two actions shared a common semantic code (e.g., “left” or “right”), but otherwise contained distinct action and perceptual features. Participants saw an arrowhead and asterisk (Stimulus A) and planned a set of right- or left-handed key-presses based on the arrowhead direction (left or right) and location of the asterisk relative to the arrowhead (left or right). Then a second stimulus (Stimulus B) appeared which consisted of a red or green number sign (#). Participants made a speeded, vocal response (“left” or “right”) based on the color identity of the number sign. Afterward, the planned response to Stimulus A was executed. If partial overlap between semantic representations (e.g., ‘left’ or ‘right’) of the withheld and intervening responses can lead to CI, then the speeded vocal response to Stimulus B (e.g., ‘left’) should be delayed when it shares a semantic feature with the withheld action (e.g., ‘left’ hand), even though the action and perceptual features between these two response events are distinct. Also, if CI is due to semantic overlap and not to hemispheric overlap between the vocal response and the withheld manual action, we should observe CI for vocal responses (left-hemisphere) regardless of whether the withheld action required a left-hand response (right-hemisphere) or a right-hand response (left-hemisphere).
Twenty-six undergraduate psychology students from Washington State University participated. Each student received optional extra credit in a psychology course, and informed consent was obtained from all participants after the nature of the procedure was fully explained to them. All participants had at least 20/40 visual acuity (assessed using a Snellen chart). This study was approved by the Washington State University Institutional Review Board.
Stimuli were presented on a 17-in. CRT approximately 50 cm from the participant. E-prime software (version 1.2) was used for stimulus generation, presentation, and data collection. A microphone connected to an E-prime voice activated response device was used to record vocal response RT. The microphone was located on the table and centered in front of the participant. The microphone distance was adjusted so that it appeared at a distance of approximately one inch in front and one inch below the participant’s lips. A cassette player recorded vocal response accuracy. The cassette player was located to the left of the CRT, approximately 40 cm from the participant. Two keypads recorded manual responses from the index finger of the left- or right-hand. Each keypad was located on a table in front of the participant, 9 cm from the table edge. A distance of 21.5 cm separated the keypads: one keypad was located 10.75 cm to the left of the participant’s body midline and the other was located 10.75 cm to the right of the participant’s body midline. Participants were able to rest their palms on the table in front of each keypad while executing all keypad responses. Left-hand responses were executed on the keypad located to the left of the participants’ body midline, and right-hand responses were executed on the keypad located to the right of the participant’s body midline. Each keypad had three keys (each were 1 × 1 cm in size, separated by 0.2 cm) oriented in a vertical array. The key in the middle of each keypad was designated as the home key. Participants rested their left and right index fingers on the left and right home keys, respectively, before and during each trial. The participant’s hands and keypads were visible to the participant throughout the experiment.
Stimuli and responses
All stimuli appeared on a black background. A white fixation cross (0.67° of visual angle) appeared in the middle of the screen before and during each trial. Stimulus A appeared 0.57° of visual angle above the fixation cross and was followed latter by Stimulus B which appeared 0.57° of visual angle below the fixation cross.
Stimulus B was a red or green number sign (#) that subtended 0.67° of visual angle. Stimulus B required a speeded vocal response “left” or “right” based on its color identity. For half of the participants, a red number sign required the vocal response “right” and a green number sign required the vocal response “left”. The other half of the participants had the opposite stimulus–response assignment.
The RT to Stimulus B was measured from the onset of Stimulus B to the onset of the vocal response to Stimulus B (detected by the voice activation device). Vocal response accuracy for Stimulus B was recorded by an audio cassette player. Accuracy data for each trial was later coded into the data file by the experimenter. If a vocal word response other than the correct word response was detected, an error was recorded for Stimulus B. Also, if any sound occurred on a single trial which may have activated the voice recording device before the vocal word response (e.g., clear throat, “ah”), that trial was eliminated from data analysis; this occurred on less than 0.16% of the total trials.
Semantic code compatibility was manipulated between Stimulus B and Stimulus A. The vocal response required for Stimulus B (e.g., “right”) either matched the response hand (e.g., “right”) required for Stimulus A (compatible) or did not match the response hand (e.g., “left”) required for Stimulus A (not compatible). Importantly, Stimulus B and Stimulus A did not share any perceptual features or motor movements. Semantic code compatibility (compatible, not compatible) between Stimulus B and Stimulus A occurred randomly and with equal probability of occurrence in each block of trials. Also, the four possible symbols for Stimulus A were equally paired with the two possible colored stimuli for Stimulus B.
Participants completed one, 90-min session consisting of one block of 32 practice trials followed by 12 blocks of 32 experimental trials. Participants vocally announced the start of each new block of trials into the microphone when prompted by a message indicating the number of the new set of blocks. At the end of block 6, participants took a short break and the experimenter turned over the audio cassette tape in the cassette player.
A repeated measures analysis of variance (ANOVA) with the factor of semantic code compatibility (compatible, not compatible) and response hand (left, right) corresponding to Stimulus A was conducted separately on correct vocal RTs and percent errors for Stimulus B, as well as percent errors for Stimulus A. The vocal RT and error analyses for Stimulus B were restricted to trials in which manual responses to Stimulus A were accurate.
Results and discussion
Percent error (within-subject standard error of the mean) for Stimulus A by hand and compatibility condition for Experiments 1–3
Demonstrating CI for two different events that overlap only in terms of the semantic codes representing these events suggests that CI is attributed to partial overlap of semantic codes representing different response events. That is, semantic overlap alone, without perceptual, motor, or hemispheric overlap between the withheld and intervening action is sufficient to obtain CI. These results also suggest that code generation and code CI occurs somewhere between late perception and early response as assumed by the COH (e.g., Hommel et al., 2001a; see also evidence by Wiediger & Fournier, 2008); that is, after developing a semantic representation of the response and prior to translating these semantic codes into a response.
According to the COH, CI should not be observed under conditions in which the intervening response can be executed automatically even if this action shares a similar code with an action plan held in short-term or working memory. The COH assumes that actions that are generated automatically are comprised of feature codes that are already integrated in long-term memory and hence occupation of a shared code in short-term or working memory by another event should not delay execution of an automatic response (Hommel et al., 2001a). To test this, Experiment 2 examined whether the CI effect found in Experiment 1 would be eliminated if the intervening, vocal response was elicited automatically (i.e., without relying on a stored representation of stimulus–response associations in short-term or working memory). The method used was identical to Experiment 1 except that the intervening stimulus (Stimulus B) consisted of the word LEFT or RIGHT which can automatically elicit a vocal response ‘left’ or ‘right’, respectively, (Kornblum & Lee, 1995; Gathrecole & Baddeley, 1993; Miyake, Emerson, Padilla, & Ahn, 2004). If the CI effect is eliminated, it would provide confirmatory evidence that CI does not generalize to automatic responses.
Twenty-six different undergraduate psychology students from Washington State University participated. Participant compensation and visual acuity was the same as that in Experiment 1.
Apparatus, stimuli, and procedures
The apparatus, stimuli, procedures, and analyses were identical to Experiment 1, except for the following. Stimulus B was the printed word “LEFT” or “RIGHT” (in upper case) which subtended 2.86° and 3.44° horizontally, respectively. Stimulus B required that the participants read and vocalize the word “left” or “right” as quickly and accurately as possible. Thus, unlike Experiment 1, participants did not have to translate the stimulus into an arbitrarily assigned response. The vocal response was intrinsically associated with the stimulus; and the written word should automatically activate its corresponding vocal response (e.g., see Gathrecole & Baddeley, 1993; Kornblum & Lee, 1995; Miyake et al., 2004). Note also that no more than 2.4% of the trials were eliminated due to a sound occurring (e.g., clear throat, “ah”) which may have activated the voice recording device before the vocal word response.
Results and discussion
The average error rate for Stimulus A was 4.7%, see Table 1. No significant effects for error rate were found based on code compatibility [F < 1], response hand [F(1, 25) = 3.62, MSE = 0.061, p = 0.07] or the interaction between code compatibility and response hand [F(1, 25) = 3.97, MSE = 0.170, p = 0.06]. The trend toward significance showed right-hand responses to be 2.6% more accurate than left-hand responses for the not compatible condition only. The overall low error rate suggests, similar to Experiment 1, that participants had little difficulty recalling the action plan to Stimulus A held in memory.
Figure 3 (middle panel) shows the mean correct vocal RT and percent errors for Stimulus B when the semantic codes between Stimulus A and B were compatible and when they were not compatible for each response hand (corresponding to Stimulus A). As shown in Fig. 3 (middle panel), there is no evidence of CI; vocal RTs to Stimulus B were similar when Stimulus B was semantically compatible and semantically not compatible with Stimulus A (F < 1). In addition, there was no significant effect of response hand [F(1, 25) = 1.25, MSE = 937.86, p > 0.27] nor was there a significant interaction between code compatibility and response hand (F < 1) found for vocal RTs. Furthermore, the average error rate was 3.35%, and no significant effects were found for error rate based on code compatibility [F(1, 25) = 2.55, MSE = 0.039, p > 0.12], response hand [F < 1] or the interaction between code compatibility and response hand [F < 1]. These findings suggest that semantic code compatibility does not lead to CI when the intervening action can be evoked automatically.
Comparisons between Experiment 2 and Experiment 1
To confirm that compatibility effects differed between Experiment 2 and Experiment 1, a mixed design ANOVA with the between-subjects factor of experiment (1, 2) and within-subjects factor of semantic compatibility (compatible, not compatible) and hand (left, right) was conducted separately on correct vocal RTs and accuracy for Stimulus B. The only effect involving the factor of experiment that approached significance was the interaction between experiment and semantic compatibility for vocal RT [F(1, 50) = 3.71, MSE = 550.65, p = 0.06]. This finding suggests that CI occurred in Experiment 1 where the color of the stimulus was arbitrarily mapped to the vocal response of “left” or “right”, but not in Experiment 2 where the word stimulus “left” or “right” could automatically elicit its associated vocal response of “left” or “right”, respectively. No other effects involving the factor of experiment approached significance (ps > 0.10).
These results suggest that when the withheld and intervening actions share semantic codes, CI occurs for the intervening action if the selection of this action imposes a demand on working memory (Experiment 1). However, CI does not appear to occur when the action corresponding to the intervening stimulus can be elicited automatically (Experiment 2). This finding is consistent with the COH which assumes that actions that do not impose a demand on short-term or working memory are immune to CI. According to the COH, the codes associated with the automatic, intervening response should already be bound in long-term memory, and hence executing an automatic response should not be delayed if it shares a code with an action held in working memory (Hommel et al., 2001a). These results also suggest that the CI effect found in Experiment 1 was not due to phonetic overlap between the semantic codes used to represent the withheld and intervening actions. Phonetic overlap for the withheld and intervening actions was identical between Experiment 2 and Experiment 1, yet CI was observed only in Experiment 1.
It is possible, however, that the failure to find a CI effect in Experiment 2 was due to a floor effect in RT. Even though the mixed design ANOVA failed to show a significant main effect of Experiment, the trend in data suggest that responses were overall much faster in Experiment 2. Thus, it is possible that our RT measure was not sensitive enough to detect differences between the compatible and not compatible conditions in Experiment 2. Experiment 3 addressed this possibility. In addition, Experiment 3 more directly examined whether CI is restricted to actions that impose a demand on short-term or working memory. Similar to the methods of Experiment 2, the intervening stimulus (Stimulus B) was the word “LEFT” or “RIGHT”. However, instead of vocalizing the word as quickly as possible, observers were to vocalize the antonym of the word. If CI is linked to accessing and vocalizing the antonym of the word stimulus (which imposes a demand on working memory), as opposed to the semantics of the word stimulus (which can be accessed directly from long-term memory), this would indicate that CI is restricted to actions that impose a demand on working memory.
Twenty-eight different undergraduate psychology students from Washington State University participated. Participant compensation and visual acuity was the same as that described in Experiment 1.
Apparatus, stimuli, and procedures
The apparatus, stimuli, procedures, and analyses were identical to Experiment 2; only the vocal response required for Stimulus B differed. Participants were required to vocalize the antonym of the printed word “LEFT” or “RIGHT” as quickly and as accurately as possible. Thus, if the word presented was “LEFT”, the participants had to quickly vocalize the word “right”; if the word presented was “RIGHT”, the participants had to quickly vocalize the word “left”. Note also that no more than 1.3% of the trials were eliminated due to a sound occurring (e.g., clear throat, “ah”) which may have activated the voice recording device before the vocal word response.
Results and discussion
The average error rate for Stimulus A was 6.20%. No significant effects for error rate were found based on code compatibility (F < 1), response hand (F < 1) or the interaction between code compatibility and response hand [F(1, 27) = 3.10, MSE = 0.265, p = 0.09], see Table 1. The low error rate suggests, similar to Experiment 1 and 2, that participants had little difficulty recalling the action plan to Stimulus A held in memory.
Figure 3 (right panel) shows the mean correct vocal RT and percent errors for Stimulus B when the semantic codes between Stimulus A and the antonym of Stimulus B were compatible or not compatible for each response hand (corresponding to Stimulus A). CI was obtained for vocal responses to Stimulus B (i.e., vocalizing the antonym of the written word). Vocal responses to Stimulus B were delayed by approximately 9 ms when the vocalized antonym was semantically compatible as opposed to not compatible with Stimulus A, F(1, 27) = 4.40, MSE = 455.38, p < 0.05. Also, the average error rate for vocal responses to Stimulus B was 2.3%, which did not differ between the code compatibility conditions, F < 1. No other effects were significant for vocal RT or error rate.
These findings suggest that when the intervening vocal response shared a semantic code with the action plan held in memory, the vocal response was delayed. Because the vocal response was determined by the antonym of the action word “LEFT” or “RIGHT” and was not determined by simply reading the word (as in Experiment 2), the CI effect appears to be restricted to actions that rely on working memory—where the previously planned action is being kept active. There was no evidence that reading the word was delayed due to semantic code overlap with the planned action held in memory. If this had been the case, we would have found the reverse RT data pattern: longer RTs for the not compatible (i.e., antonym not compatible; but word compatible) relative to the compatible (i.e., antonym compatible; but word not compatible) conditions.
These findings are consistent with the lack of a CI effect found in Experiment 2, when the intervening action to Stimulus B required reading and vocalizing a written word. Mean vocal RTs in this experiment were much larger than those obtained in Experiment 2 and yet there was no evidence that CI occurred based on semantic code overlap between the planned action held in memory and that corresponding to the reading of the word. Moreover, because CI was obtained based on semantic code overlap between the planned action held in memory and the vocalization of the antonym, this suggests that this experiment was sensitive enough to detect CI.
Consistent with the COH, partial overlap of semantic codes between an action plan held in memory and an intervening action is sufficient to delay execution of the intervening action; the two actions do not have to share the same response modality or same hemisphere. This study showed that holding a manual response sequence (key-presses with the ‘left’-hand or ‘right’-hand) in memory could delay execution of a vocal response (‘left’ or ‘right’) when both action events shared a compatible semantic code (‘left’ or ‘right’) compared to when they did not. This CI effect was not contingent on whether the withheld manual action required a right- or left-hand response, and hence was not dependent on whether the vocal response and the withheld manual action shared the same hemisphere. Also consistent with the COH, this CI effect was observed only for intervening actions that imposed a demand on working memory (e.g., Kornblum & Lee, 1995), not for intervening actions that could be executed automatically. CI due to semantic code overlap occurred when the intervening vocal responses ‘left’ or ‘right’ were arbitrarily mapped to the color identity (red or green) of a stimulus (Experiment 1) or to a word (antonym of ‘LEFT’ or ‘RIGHT’; Experiment 3), but not when the vocal responses ‘left’ or ‘right’ could be evoked automatically by the written words ‘LEFT’ or ‘RIGHT’, respectively, (Experiment 2).
According to the COH, action planning involves initial activation of all features relevant to the action event followed by a process in which these activated codes become integrated into a common representation (action plan) that occupies the codes that represent it. These codes are occupied by the action plan (held in working memory) until the action plan is executed or abandoned. This occupation of codes renders these codes temporarily unavailable for the construction of a new action plan. If the construction of a new action plan requires a code that is already occupied in working memory (e.g., the semantic code ‘left’), execution of the new action plan will be delayed relative to other newly constructed action plans that do not require a code that is currently occupied in working memory. Also, actions that can be executed automatically are assumed to be based on retrieval of an action plan, where action event codes are already integrated, from long-term memory. Thus, execution of automatic responses (e.g., vocal response associated with word reading) would not require access to occupied codes in working memory. Because semantic code compatibility alone was sufficient to obtain CI for actions that impose a demand on working memory, this suggests that the cognitive codes representing an action plan can consist of abstract, semantic descriptions of the different action events.
In addition, because CI was obtained without perceptual, motor, or hemispheric overlap between the withheld and intervening actions, this suggests that the CI effect occurs during response selection. This finding and interpretation is consistent with the assumption of the COH that CI occurs somewhere between late perception and early response. It is also consistent with the findings of Mattson and Fournier (2008). Using a response competition paradigm, they showed that CI occurred for selected actions associated with a target stimulus, but not for actions activated by distractors. They concluded that CI must occur during a central, response selection stage because response activation alone was not sufficient to evoke CI.
It is possible that participants used verbal descriptors to represent the relevant semantic features corresponding to the action events in order to aid and facilitate response selection and retrieval (e.g., Miyake et al., 2004). Detailed debriefing in recent research conducted in our lab suggests that approximately 85% of participants used a verbal strategy (e.g., “left-up”) and 10% a spatial-motor strategy (e.g., mental image of the motor response) to remember the withheld manual response to Stimulus A which was similar to the version of Stimulus A used in this study (i.e., right- or left-pointing arrowhead with an asterisk above or below the arrowhead). Integration of verbal descriptors into an action plan may serve as a reliable retrieval cue for actions that must be executed after a short delay. However, recent research conducted in our laboratory (Fournier, Kirkwood, Mattson & Herzog, in preparation) showed that the size of the CI effect was not reduced when an articulatory suppression task was carried out before and during the presentation and action planning of Stimulus A. This suggests that CI is not contingent on the use of articulatory rehearsal or verbal descriptors to encode the action plan. Also, the size of the CI effect was not influenced by whether the withheld and/or intervening actions were represented verbally or visually (Fournier et al., in preparation) which further suggests that semantic overlap is sufficient to obtain CI.
Taken together, our results suggest that feature binding leading to code occupation is not restricted to low-level action-related codes associated with a particular action modality. Instead, feature binding leading to code occupation can consist of semantic codes representing action events. This conclusion contrasts that made by Hommel and Müsseler (2006). They found CI effects between a withheld and intervening event only when the two events required the same type of response (manual or vocal). However, as argued in “Introduction”, their failure to find CI across different response modalities was likely due to the poor sensitivity of the accuracy measure used to quantify CI, particularly since the CI effect sizes we found were small—12 ms in Experiment 1 and 9 ms in Experiment 3. In fact, the CI effect sizes obtained in Experiment 1 and 3 were about half of that obtained in other studies where the withheld and intervening actions required the same response modality and limb (i.e., left- or right-hand; Stoet & Hommel, 1999; Wiediger & Fournier, 2008); however, they are comparable to the effect size obtained by Stoet and Hommel (1999) where feature overlap was based on the side of the body (left or right) that two different limbs (hand and foot) were located. Thus, while it is possible that motor overlap can increase CI, by increasing the number of codes in common between the withheld and intervening action (e.g., Stoet & Hommel, 1999; Hommel et al., 2001a), it is not necessary to obtain CI (see also Wiediger & Fournier, 2008).
In summary, execution of a vocal response can be delayed due to code occupation by an action plan held in memory that represents a manual response sequence. That is, if the two actions share a similar semantic code and if the stimulus event requiring a vocal response imposes a demand on working memory to retrieve the appropriate response.