Consciousness and Cognition Consciousness and Cognition 14 (2005) 278–295 www.elsevier.com/locate/concog Automatic and controlled semantic processing: A masked prime-task effect B. Valde´sa,*, A. Catenab, P. Marı´-Beffaa a University of Wales, Bangor, Adeilad Brigantia, Penrallt Road, Gwynedd LL57 2AS, UK b University of Granada, Spain Received 5 February 2004 Available online 26 October 2004 Abstract A classical definition of automaticity establishes that automatic processing occurs without attention or consciousness, and cannot be controlled. Previous studies have demonstrated that semantic priming can be reduced if attention is directed to a low-level of analysis. This finding suggests that semantic processing is not automatic since it can be controlled. In this paper, we present two experiments that demonstrate that semantic processing may occur in the absence of attention and consciousness. A negative semantic priming effect was found when a low-level prime-task was required and when a masked lexical decision prime-task was performed (Experiment 1). This paper also discusses the limitations of the inhibitory mechanism involved in negative semantic priming effect. Ó 2004 Elsevier Inc. All rights reserved. Keywords: Automaticity; Consciousness; Attention; Semantic priming; Negative Priming; Word processing 1. Introduction When confronted with printed words, skilled readers have the subjective impression that read- ing is an automatic process that does not requires attentional control or effort. For decades this * Corresponding author. Fax: +44 1248 38 2599. E-mail address: [email protected] (B. Valde´s). 1053-8100/$ - see front matter Ó 2004 Elsevier Inc. All rights reserved. doi:10.1016/j.concog.2004.08.001 B. Valde´s et al. / Consciousness and Cognition 14 (2005) 278–295 279 idea has guided the research on word recognition (see Rayner, Foorman, Perfetti, Pesetsky, & Seidenberg, 2001, for a recent review) and has served to explain some experimental phenomena such as Stroop interference (Stroop, 1935; see also MacLeod, 1991, for a review) and semantic priming (Neely, 1991). There is evidence that words are processed up to semantic levels even though participants are not instructed to attend to them (Fuentes, Carmona, Agis, & Catena, 1994), or after attending to a non-semantic, physical dimension of the word (i.e., the ink colour) instead of the meaning (MacLeod, 1991, 1992). Processing of words is assumed to proceed with- out explicit intention but also without consciousness. Merely presenting words either under a sub- jective conscious threshold (Marcel, 1983) or under an objective one (Dehaene et al., 1998; Naccache & Dehaene, 2001) triggers a processing stream that cannot be prevented. The complete processing of words involves the processing of multiple dimensions or levels (McClelland & Rumelhart, 1981; Seidenberg & McClelland, 1989): a letter level, at which features are integrated to form the letters that compose a particular word; an orthographic level, at which letters are integrated to form orthographic patterns; a lexical level, at which the orthography of the word is activated; and a semantic level, at which the meaning of the word is accessed (Bentin, Mouchetant-Rostaing, Giard, Echallier, & Pernier, 1999). Event-related potentials and cerebral blood flow studies (Posner, Abdullaev, McCandliss, & Sereno, 1999; Posner & Petersen, 1990) have suggested that these levels are reached following a clear sequence. These stages of processing are commonly believed to occur in a bottom-up manner that proceeds automatically (Neely, 1991). Despite such automaticity, recent studies have suggested that both unattended and uncon- scious information can be controlled (Carr & Dagenbach, 1990; Dagenbach, Carr, & Wilhelmsen, 1989; Tzelgov, Henik, & Berger, 1992). To test the automaticity of word processing, several researchers have modified the traditional procedure of semantic priming by manipulating the level of representation at which attention is directed to during the prime display. The effect obtained with this procedure is known as Prime-Task Effect (Marı´-Beffa, Fuentes, Catena, & Houghton, 2000). In traditional semantic priming experiments, pairs of semantically related and unrelated words are presented in a sequen- tial manner. Participants are instructed to perform a task that demands explicit awareness of the meaning of the word (i.e., lexical decision, categorization, naming; see Neely, 1991 for a review). This task can be performed, or not, in the first word (prime) and the effect is measured in the reac- tion time to the second word (probe) task. It is assumed that the presentation of the first word produces the activation of its internal representation in memory, sending also activation to the representations of those words that are more closely associated with it. Thus, when an associated word appears as a probe stimulus, the previous activation of its representation due to the presence of the prime related word facilitates any response to it. In Prime-Task effect experiments the most common procedure is to compare the semantic prim- ing effect, when participants are instructed to perform a lower level task on prime display (i.e. let- ter search), with another condition in which a higher or deeper level task is required (e.g. naming, lexical decision, categorisation, etc.). In both cases, participants generally perform a lexical deci- sion task on the probe word. It has been shown that if the prime-task requires attention to be allo- cated to a lower level of analysis, like in a letter search task, then the semantic priming effect is reduced (Chiappe, Smith, & Besner, 1996; Henik, Friedrich, & Kellogg, 1983; Henik, Friedrich, Tzelgov, & Tramer, 1994; Kaye & Brown, 1985; Parkin, 1979; Smith, 1979; Smith, Theodor, & Franklin, 1983; Stolz & Besner, 1996). These results seem to challenge the idea of automaticity 280 B. Valde´s et al. / Consciousness and Cognition 14 (2005) 278–295 of word processing since they may suggest that the orientation of attention to a deep level of rep- resentation is necessary for the processing of meaning. Although there are some disagreements in the theoretical explanation of this effect, most authors assume that the allocation of attention to a low level feature impairs or interrupts the flow of processing to further stages. For example, Henik et al. (1994) suggested that dedicating all the processing resources to the low level features could exert this kind of attentional control. As a result, there are not enough resources left to activate semantic properties, and semantic priming is reduced or eliminated. Alternatively, the lack of semantic processing could also be modelled by blocking the feedback loops between lexical and semantic levels in a neural network, as attention is oriented to lower levels (Stolz & Besner, 1996). In addition, Duscherer and Holender (2002) claim that semantic processing requires the awareness of the meaning of the prime word. Thus, the reduction of semantic priming when atten- tion is oriented to low level features would be the logical consequence of deploying awareness away from the semantic level. In any case, all these theoretical accounts assume that semantic pro- cessing can be interrupted before it is completed if attention is not directed to a semantic level. However, using this basic paradigm, the results are not always consistent. One of the patterns of results most problematic, for the ‘‘lack of semantic processing’’ accounts, is that some studies have shown a negative semantic priming effect when a letter search task is performed on the prime display. If semantic processing was never completed, then no semantic effect at all, positive or neg- ative, would be expected. The presence of a negative semantic priming effect suggests that the con- trol mechanism responsible for the reduction of positive semantic priming may act once the semantic representations have been already processed, bringing down their level of activation, even below resting levels, in order to better control the goal of the task (Catena, Fuentes, & Tu- dela, 2002; Hoffman & MacMillan, 1985; Marı´-Beffa et al., 2000; Tipper, Weaver, & Houghton, 1994). Orienting attention to a low level of processing probably does not block the further pro- cessing of the semantic attributes of the prime. Rather, the prime word is automatically fully pro- cessed, but those dimensions that are not relevant for the task have to be inhibited to prevent them from reaching the control of the response (Marı´-Beffa et al., 2000). Interestingly, most models incorporating an inhibitory mechanism (see Tipper, 2001 for a re- view) assume that the development of inhibition requires certain time to be effective. Supporting this idea, Yee (1991) has found positive priming when prime-probe onset asynchrony (SOA) was 500 ms, but negative semantic priming when the SOA was increased to 600 ms. Indeed, some models (Houghton & Tipper, 1994; Houghton, Tipper, Weaver, & Shore, 1996) assume that inhi- bition cannot develop until the offset of the stimuli. Here, we addressed this issue in two experi- ments displaying words for a very short time (25 ms), and increasing the prime-response to probe interval in order to allow the development of inhibition of the prime words. A second line of evidence favouring the automaticity of semantic processing from words comes from masking experiments where the prime display is presented under the conscious threshold. In these experiments the prime stimuli were presented during a brief period of time.