1. Introduction
In his classic studies on spatial cueing, Posner (Reference Posner1980) demonstrated that symbolic cues can orient spatial attention, thereby affecting visual perception. For instance, an upward-pointing arrow presented in the center of a display facilitates detection of a square at the top of the display and hinders detection at the bottom. Like those symbolic cues (e.g., arrows), some words also have explicit spatial meanings, and those words also affect visual perception at their associated locations (i.e., linguistic cueing). For instance, the explicitly spatial word ‘up’ facilitates detection of a square at the top of a display, and impairs detection at the bottom (e.g., Gibson & Kingstone, Reference Gibson and Kingstone2006; Gibson & Sztybel, Reference Gibson and Sztybel2014; Hommel et al., Reference Hommel, Pratt, Colzato and Godijn2001; Logan, Reference Logan1995; Ostarek & Vigliocco, Reference Ostarek and Vigliocco2017; Pauszek & Gibson, Reference Pauszek and Gibson2018; Shaki & Fischer, Reference Shaki and Fischer2023a, Reference Shaki and Fischer2023b). In fact, words with merely implicit spatial associations, such as religious words (e.g., ‘god’ = high, ‘satan’ = low; Chasteen et al., Reference Chasteen, Burdzy and Pratt2010) and temporal words (e.g., ‘before’ = left, ‘after’ = right; Ouellet et al., Reference Ouellet, Santiago, Funes and Lupiánez2010), can also facilitate perception at their associated locations.
In some circumstances, however, the opposite may occur: Counter-intuitively, words with implicit spatial associations sometimes hinder identification of a visual target at their associated location. In the earliest demonstration of this spatial interference effect, Richardson et al. (Reference Richardson, Spivey, Barsalou and McRae2003) presented brief sentences with either a vertical association (e.g., ‘The eagle flies to the river’) or a horizontal association (e.g., ‘The miner pushes the cart’), followed by a visual target (■ or ●) on either the vertical axis (top or bottom of screen) or the horizontal axis (left or right). They found that vertically associated sentence cues slowed identification of targets along either end of the vertical axis. In a more fine-grained demonstration, Bergen et al. (Reference Bergen, Lindsay, Matlock and Narayanan2007) similarly embedded spatial cue words within brief sentences, and they showed that high-associated cues (e.g., ‘The mule climbed’) slowed identification of visual targets specifically at the top location, whereas low-associated cues (e.g., ‘The chair toppled’) slowed identification at the bottom location. Barsalou (Reference Barsalou2008) presented implicitly spatial cue words in isolation (e.g., ‘hat’), again demonstrating location-specific interference even without any semantic reference frame. Here, two experiments are reported that investigate the conditions under which this spatial interference effect may or may not replicate.
2. Why replicate spatial interference?
The spatial interference effect warrants replication for several reasons. (1) The effect is surprising, and surprising effects are relatively likely to be false positives (Forstmeier et al., Reference Forstmeier, Wagenmakers and Parker2017). (2) The spatial interference effect has been interpreted as an important source of evidence for grounded cognition, as explained below. (3) Consequently, those early demonstrations of spatial interference have had some impact on the field. Richardson et al. (Reference Richardson, Spivey, Barsalou and McRae2003), Bergen et al. (Reference Bergen, Lindsay, Matlock and Narayanan2007) and Estes et al. (Reference Estes, Verges and Barsalou2008) collectively have accrued 1265 citations on Google Scholar and 618 citations in Scopus (both retrieved 31 October 2024). (4) A series of replication failures in Italian has been reported (Petrova et al., Reference Petrova, Navarrete, Suitner, Sulpizio, Reynolds, Job and Peressotti2018). In summary, because the spatial interference effect is counterintuitive, it has had some theoretical impact, but it may be a false-positive effect, and at this point, the true state of the effect is unknown.
3. Defining the spatial interference effect
The original demonstrations of spatial interference and most of the replication failures shared several methodological commonalities. Those commonalities, which will be delineated below, will be taken as a working definition of the spatial interference effect.
3.1. Multiple-Cue Categories
Prior demonstrations of spatial interference have used cue words from various semantic categories (e.g., animals, clothing, vehicles, etc.). Studies in which the cue words are from a single category (e.g., house-related words such as ‘attic’ and ‘cellar’) do not elicit spatial interference (Gozli et al., Reference Gozli, Chasteen and Pratt2013).
3.2. Short SOA
Stimulus onset asynchrony (SOA) is the delay between cue and target onsets. Estes et al. (Reference Estes, Verges and Barsalou2008) presented cue words for 100 ms, followed by a blank delay of 50 ms and finally, the visual target (i.e., SOA = 150 ms). It has been shown that with SOAs longer than about 400 ms, facilitation may occur instead of interference (Goodhew et al., Reference Goodhew, McGaw and Kidd2014; Gozli et al., Reference Gozli, Chasteen and Pratt2013; Zhang et al., Reference Zhang, Luo, Zhang, Wang, Zhong and Li2013).
3.3. Nonsemantic Targets
The critical factor that makes the spatial interference effect so counterintuitive is the use of nonsemantic targets. If semantically related targets are used instead, the result is rather intuitive: Cue words with spatial associations (e.g., ‘bird’) facilitate recognition of the denoted object (i.e., an image of a bird) at its associated location (Estes et al., Reference Estes, Verges and Adelman2015). Tests of the spatial interference hypothesis, in contrast, use nonsemantic targets such as geometric shapes (e.g., ■ or ●) or alphanumeric characters (e.g., p or q).
3.4. Identification Task
Spatial interference occurs in the identification task, in which the visual target must be identified. For instance, Richardson et al. (Reference Richardson, Spivey, Barsalou and McRae2003) and Bergen et al. (Reference Bergen, Lindsay, Matlock and Narayanan2007) had participants press one or another button to identify whether the target was a square or a circle. A detection task, in which participants merely indicate the presence of a stimulus rather than identifying it, does not produce spatial interference (e.g., Dudschig et al., Reference Dudschig, Lachmair, de la Vega, De Filippis and Kaup2012; Gozli et al., Reference Gozli, Chasteen and Pratt2013).
In sum, tests of spatial interference use cue words from multiple semantic categories followed shortly by nonsemantic targets in an identification task. In the General Discussion, we consider why each of these factors affects spatial interference.
4. Theoretical explanation of spatial interference
The spatial interference effect is thought to arise from two separable and counteracting components: (i) facilitation of attentional orienting and (ii) interference of object recognition. Those two components, in turn, are thought to arise from distinct processes of (i) linguistically mediated visual search and (ii) perceptual simulation of the denoted object.
4.1. Linguistic Orienting
Spatial interference may be partially understood in terms of linguistically mediated visual search (Estes et al., Reference Estes, Verges and Adelman2015; Gozli et al., Reference Gozli, Pratt, Martin and Chasteen2016). A wealth of evidence from the ‘visual world paradigm’, in which people hear spoken language while viewing object arrays, indicates that words elicit a visual search for semantically related objects (for review, see Huettig et al., Reference Huettig, Rommers and Meyer2011). For example, hearing the word ‘cake’ leads people to fixate on an image of a cake in a visual scene (e.g., Altmann & Kamide, Reference Altmann and Kamide1999). Moreover, the visual search for target objects is not random; rather, people systematically search for objects in the locations where they occur most often (i.e., contextual cueing; Chun & Jiang, Reference Chun and Jiang1998). The word ‘bird’, for instance, elicits a search for a bird-related image toward the top of a display. Thus, for words with implicit spatial associations, the visual search is biased toward the associated location (Estes et al., Reference Estes, Verges and Barsalou2008). This linguistic orienting is most simply shown via eye-tracking, where saccade launches are faster toward the word’s associated location (Dudschig et al., Reference Dudschig, Souman, Lachmair, de la Vega and Kaup2013; Dunn et al., Reference Dunn, Kamide and Scheepers2014). For instance, after hearing or reading ‘bird’, saccades are initiated faster upward than downward. This effect is also evident in ERP studies, where targets appearing in the cue’s associated location evoke a larger N1 response, which is linked to attentional shifts (Zhang et al., Reference Zhang, Luo, Zhang, Wang, Zhong and Li2013).
4.2. Perceptual Simulation
Spatial interference may also be partially understood in terms of grounded cognition (Barsalou, Reference Barsalou2008). Essentially, words evoke a perceptual simulation of the denoted object or event, which entails a reactivation of the neural patterns involved in prior experiences of that object or event (Barsalou, Reference Barsalou1999, Reference Barsalou2008, Reference Barsalou2016). For example, the word ‘bird’ may partially reactivate the neural pattern involved in the actual perception of a real bird, including its appearance, sound and so on. It may also reactivate a typical situation in which we experience birds, including typically co-occurring objects such as trees and contexts such as hiking in a forest. Thus, perceptual simulation is one mechanism by which situation models (van Dijk & Kintsch, Reference van Dijk and Kintsch1983; Zacks & Tversky, Reference Zacks and Tversky2001; Zwaan, Reference Zwaan2016) are constructed and updated during language comprehension.
Presumably due to perceptual simulation, words facilitate recognition of the denoted object. That is, ‘bird’ speeds recognition of bird-related images by pre-activating the perceptual representation of a bird. And conversely, objects are recognized more slowly when preceded by a semantically unrelated word, compared to a semantically related word or no word (e.g., Lupyan & Ward, Reference Lupyan and Ward2013). For instance, the word ‘bird’ hinders recognition of an apple. Proponents of grounded cognition attribute this interference effect to neural or perceptual competition (e.g., Bergen et al., Reference Bergen, Lindsay, Matlock and Narayanan2007; Estes et al., Reference Estes, Verges and Barsalou2008): Perceptual simulation of a word neurally competes with or perceptually masks the unrelated target object. That is, ‘bird’ pre-activates the perceptual representation of a bird, which interferes with the perceptual identification of an apple.
4.3. Location-Specific Perceptual Simulation
Neither linguistic orienting nor perceptual simulation alone can explain the spatial interference effect. To begin with, the speeded orienting toward a cue word’s associated location theoretically should facilitate perception at that location, not hinder it. Indeed, when the denoted object appears in its associated location, its recognition is facilitated (Estes et al., Reference Estes, Verges and Adelman2015; Gozli et al., Reference Gozli, Pratt, Martin and Chasteen2016). For instance, ‘bird’ speeds recognition of bird-related images at the top of a display. Critically, however, spatial interference occurs with semantically unrelated targets, such as when ‘bird’ precedes a square target. Nor can perceptual simulation of the cue word fully explain the spatial interference because the unrelated target is the same across varying locations (e.g., top, bottom), yet ‘bird’ differentially hinders recognition of squares at those different locations.
The spatial interference effect thus appears to rely on the particular combination of linguistic orienting and perceptual simulation. That is, spatial interference appears to result from (i) an attention shift to the cue word’s associated location and (ii) a perceptual simulation of the denoted object in that specific location. The cue word ‘bird’ shifts attention to the top of the display and activates the perceptual representation of a bird. Thus, when a bird image appears at the top of the display, recognition is facilitated. When that bird image instead appears at the bottom of the display, recognition is slightly delayed (Estes et al., Reference Estes, Verges and Adelman2015) because attention must shift down from the top to the bottom location.
Less intuitive is the case when an unrelated object follows the cue word (e.g., ‘bird’). Regardless of the target’s location, unrelated targets (e.g., a square) are recognized substantially more slowly than related targets (i.e., a bird; Estes et al., Reference Estes, Verges and Adelman2015). That is, the perceptual simulation of the cue word substantially delays recognition of the unrelated target (Lupyan & Ward, Reference Lupyan and Ward2013). If recognition of the unrelated target were simply a matter of overcoming the perceptual simulation of the cue word (e.g., awaiting its deactivation), then that target should presumably be recognized more quickly in the cue’s associated location because recognizing that target in the opposite location would additionally require an attention shift down from the top to the bottom location. But in fact, the opposite occurs: The unrelated target is recognized more slowly in the cue’s associated location (i.e., spatial interference). Why?
It appears that the perceptual simulation of the cue word is location specific. So when a square appears at the top of the display, the pre-activated perceptual representation of ‘bird’ neurally competes with or perceptually masks recognition of that target at that location. Only after that bird representation dissipates can the square be identified. When the square instead appears at the bottom of the display, however, it requires an attention shift down to that bottom location. And critically, that attention shift appears to disengage the visual system from the bird representation at the top of the visual field, allowing faster recognition of the square at that bottom location. That is, because the perceptual representation of ‘bird’ occurs at the top location, it creates stronger neuroperceptual competition at the associated location than at other, noncued locations. And thus, recognition of a square is faster at the bottom than at the top location because, evidently, shifting attention away from the perceptual representation of ‘bird’ is faster than waiting for that perceptual representation to dissipate.
In sum, spatial interference appears to arise from a location-specific perceptual simulation of the cue word, which competes with or masks the unrelated visual target at the cue’s associated location. This explanation, however, assumes that the spatial interference effect is indeed real and reliable. And the evidence of that is mixed.
5. Prior evidence of spatial interference
All known tests of the spatial interference hypothesis are summarized in Table 1.
Table 1. Prior tests of the spatial interference hypothesis
5.1. Successes
Bergen et al. (Reference Bergen, Lindsay, Matlock and Narayanan2007, Experiments 1 and 2) twice demonstrated spatial interference with brief sentences (e.g., ‘The mule climbed’). Estes et al. (Reference Estes, Verges and Barsalou2008) demonstrated the effect twice with word pairs (e.g., ‘cowboy hat’; Experiments 1 and 2) and once with single-word cues (e.g., ‘hat’; Experiment 3). Verges and Duffy (Reference Verges and Duffy2009) replicated that effect twice with noun cues (e.g., ‘bird’) and once with verb cues (e.g., ‘rise’). Gozli et al. (Reference Gozli, Chasteen and Pratt2013, Experiments 3, 4 and 6) replicated it a further three times with noun cues, and Estes et al. (Reference Estes, Verges and Adelman2015, Experiments 3 and 4) replicated it once more with concrete nouns (e.g., ‘bird’) and once with abstract nouns (e.g., ‘truth’). Finally, Petrova et al. (Reference Petrova, Navarrete, Suitner, Sulpizio, Reynolds, Job and Peressotti2018, Experiment 7) replicated the effect when they explicitly directed participants’ attention to the cue words’ spatial associations. Thus, spatial interference has been demonstrated 14 times.
5.2. Failures
Bergen et al. (Reference Bergen, Lindsay, Matlock and Narayanan2007, Experiments 3 and 4) twice failed to obtain spatial interference with metaphorical cues (e.g., ‘The market sank’). Petrova et al. (Reference Petrova, Sulpizio, Navarrete, Job, Suitner and Peressotti2013) failed to replicate the spatial interference effect in the absence of semantic context. In the largest test of spatial interference to date, Estes (Reference Estes2016), personal communication) also failed to obtain spatial interference. As part of the Reproducibility Project (Open Science Collaboration, 2015), Renkewitz and Müller (Reference Renkewitz and Müller2015) failed to replicate the effect. Most recently, Petrova et al. (Reference Petrova, Navarrete, Suitner, Sulpizio, Reynolds, Job and Peressotti2018) reported a series of 10 replication attempts, nine of which failed to replicate the spatial interference effect. Thus, 14 failures to replicate the spatial interference effect have been reported in the literature.
5.3. Weighing the Evidence
In total, there have been 28 known tests of the spatial interference hypothesis by five independent research groups (see Table 1). Yet, the evidential status of spatial interference remains equivocal: The effect has been successfully obtained 14 times, and 14 failures to replicate the effect have also been reported. So, is the effect real or not? On one hand, 14 successful demonstrations of spatial interference seem too many for them all to be false-positive Type I errors. Moreover, because four independent research groups have found the effect, it is also unlikely to be attributable to methodological idiosyncrasies. On the other hand, 14 known tests of spatial interference have failed to replicate the effect, and due to publication bias, there may well be more. Although nearly all of those replication attempts were substantially underpowered, it seems unlikely that they are all false-negative Type II errors. In order to achieve 80% power to reject a small effect, a replication study must have a sample that is about 2.5 times larger than the original sample (Simonsohn, Reference Simonsohn2015). In fact, the majority of the replication attempts actually had smaller samples than the original (see Table 1).
Thus, there appears to be valid evidence both for and against the veracity of a spatial interference effect. How can this apparent discrepancy be reconciled?
6. Orthographic transparency
Estes and Barsalou (Reference Estes and Barsalou2018) noted that prior tests of spatial interference conducted in English tended to produce a significant effect, whereas tests in other languages (i.e., Italian or German) tended to produce no effect. In an attempt to understand this pattern, they searched for relevant language properties on which (a) Italian and German are similar to one another and (b) both Italian and German differ from English. One salient factor that fits this description is orthographic transparency, which refers to the consistency of print-to-sound correspondences within a language. In orthographically ‘transparent’ languages, a given letter (or string of letters) tends to be pronounced the same across different words. In orthographically ‘opaque’ languages, in contrast, a given letter (or string of letters) may be pronounced in different ways across different words. In English, for instance, the letter ‘o’ has a soft pronunciation in ‘on’ but a hard pronunciation in ‘no’. Orthographic transparency is a matter of degree, and as it turns out, Italian and German both have more transparent orthography than English. More specifically, Italian orthography may be considered transparent, whereas German is semi-transparent, and English has opaque orthography (Schmalz et al., Reference Schmalz, Marinus, Coltheart and Castles2015).
Figure 1 illustrates the reliability of spatial interference as a function of orthographic transparency. In English, an opaque orthography, there have been 16 known tests of spatial interference. Thirteen of those (81%) produced significant effects. In more transparent orthographies such as Italian and German, there have been 12 known tests of spatial interference, 11 of which (92%) failed to replicate the effect. In a meta-analysis of these 28 tests of spatial interference, Estes and Barsalou (Reference Estes and Barsalou2018) found that orthographic transparency significantly moderated the effect. Specifically, in English, the overall effect was significant and of moderate size (19 ms, p < .001). In more transparent languages, however, there was no spatial interference effect (1 ms, p = .44). Thus, they identified orthographic transparency as a ‘hidden moderator’ of spatial interference. But two limitations of that observation are important to note here. First, this presumed moderator was identified post hoc and has not been directly tested. Second, this presumed moderation is merely descriptive. Spatial interference does indeed appear substantially more reliable in opaque languages, but why?
Figure 1. Prior studies in an orthographically opaque language (English) more often successfully demonstrated spatial interference, whereas studies in more orthographically transparent languages (Italian and German) more often failed to replicate the effect.
7. Semantic processing
Spatial congruence effects are sensitive to semantic processing (Lebois et al., Reference Lebois, Wilson-Mendenhall and Barsalou2015; Santiago et al., Reference Santiago, Ouellet, Román and Valenzuela2012). For instance, Meier and Robinson (Reference Meier and Robinson2004) showed that positive words (e.g., ‘love’) are evaluated more quickly when presented at the top of a display, whereas negative words (e.g., ‘hate’) are evaluated more quickly at the bottom. Subsequently, however, this valence–space congruence effect was shown to be affected by attention to the words’ meanings. Brookshire et al. (Reference Brookshire, Ivry and Casasanto2010) showed that this effect occurred when distractor trials required a semantic judgment, but not when they required a perceptual judgment. Santiago et al. (Reference Santiago, Ouellet, Román and Valenzuela2012) replicated the effect only when they oriented participants’ attention to either the meaning of the word or the word’s spatial location on the display. Lebois et al. (Reference Lebois, Wilson-Mendenhall and Barsalou2015) further showed that the effect occurred only when participants judged the words’ spatial associations. Thus, semantic processing appears to influence spatial congruence effects.
Languages vary in the extent to which they involve semantic processing during reading (Katz & Frost, Reference Katz and Frost1992; Schmalz et al., Reference Schmalz, Marinus, Coltheart and Castles2015). Word reading entails converting graphemes (letters) to phonemes (sounds), and in transparent orthographies such as Italian, the highly consistent mapping of letters to sounds allows words to be read directly, with relatively little activation of lexical–semantic representations (Burani et al., Reference Burani, Arduino and Barca2007; Kwok et al., Reference Kwok, Cuetos, Avdyli and Ellis2017; Peressotti & Job, Reference Peressotti and Job2003; Schmalz et al., Reference Schmalz, Marinus, Coltheart and Castles2015). That is, words can be read with relatively little activation of their meanings (i.e., nonsemantic reading). In fact, computational models that entirely lack a semantic system nonetheless can correctly read Italian words with up to 98% accuracy (Pagliuca & Monaghan, Reference Pagliuca and Monaghan2010). In contrast, in opaque orthographies such as English, due to the highly inconsistent grapheme–phoneme mappings, many words cannot be read correctly via phonological rules. Such ‘exception words’ with irregular pronunciation can be read correctly only by accessing the lexical–semantic system (i.e., semantic reading), and moreover, the high prevalence of exception words induces semantic processing in general, even when reading words with regular pronunciation.
Several lines of evidence confirm that semantic processing is more robust when reading in opaque orthographies (e.g., English) than in transparent orthographies (e.g., Italian). First, brain areas involved in phoneme processing are more strongly activated when reading in Italian, but brain areas involved in lexical–semantic processing are more strongly activated when reading in English (Paulesu et al., Reference Paulesu, McCrory, Fazio, Menoncello, Brunswick, Cappa, Cotelli, Cossu, Corte, Lorusso, Pesenti, Gallagher, Perani, Price, Frith and Frith2000). Second, semantic factors such as imageability and age of acquisition have more robust effects on reading in English (Balota et al., Reference Balota, Cortese, Sergent-Marshall, Spieler and Yap2004) than in Italian or other transparent orthographies (Barca et al., Reference Barca, Burani and Arduino2002; Bates et al., Reference Bates, Burani, D’Amico and Barca2001; Buchanan & Besner, Reference Buchanan and Besner1993; Burani et al., Reference Burani, Arduino and Barca2007; see also Bakhtiar & Weekes, Reference Bakhtiar and Weekes2015). Finally, semantic priming is more robust in English and other opaque orthographies (Hutchison et al., Reference Hutchison, Balota, Neely, Cortese, Cohen-Shikora, Tse and Buchanan2013) than in transparent orthographies such as Italian (Frost et al., Reference Frost, Katz and Bentin1987; Peressotti & Job, Reference Peressotti and Job2003; Tabossi & Laghi, Reference Tabossi and Laghi1992). Thus, a great deal of theoretical and empirical research indicates that semantic processing is stronger when reading in English than in Italian.
This is not to say that semantic processing never occurs when reading in Italian, of course. Rather, tasks that typically induce semantic processing, and manipulations that experimentally induce semantic processing, also elicit semantic processing in Italian. For instance, the lexical decision and picture-naming tasks elicit deeper semantic processing than the reading-aloud task. Accordingly, semantic effects occur in Italian with lexical decisions and picture naming to a greater extent than with reading aloud (e.g., Bates et al., Reference Bates, Burani, D’Amico and Barca2001; Burani et al., Reference Burani, Arduino and Barca2007). Moreover, when semantic processing is experimentally induced, such as by requiring semantic judgments (Peressotti & Job, Reference Peressotti and Job2003) or by including irregularly pronounced words (Tabossi & Laghi, Reference Tabossi and Laghi1992), then semantic priming also emerges in Italian. Although semantic processing can be observed in Italian – either by experimentally inducing it or by using tasks that naturally entail it – simple reading in Italian does not naturally elicit deep semantic processing (Burani et al., Reference Burani, Arduino and Barca2007; Kwok et al., Reference Kwok, Cuetos, Avdyli and Ellis2017; Peressotti & Job, Reference Peressotti and Job2003; Schmalz et al., Reference Schmalz, Marinus, Coltheart and Castles2015). And critically, the linguistic cueing paradigm that is used to test the spatial interference effect does not require semantic processing. Participants do not respond to the cue words in any way, the cue words do not predict the location of the subsequent target, and indeed, the task can be completed successfully without even reading the cue words. Thus, we suggest that prior tests of spatial interference in Italian may not have induced sufficiently deep semantic processing of the cue words.
Given that (i) semantic processing is necessary for spatial congruence effects, and (ii) semantic processing is more likely when reading in English than in Italian, it follows that the spatial interference effect should be more likely in English than in Italian. In other words, the lack of spatial interference in Italian may be attributable to insufficient semantic processing of the cue words.
8. The present research
Spatial interference has been demonstrated 13 times in English and has failed to replicate 10 times in Italian. Little would be learned by attempting to either replicate the effect again in English or fail to replicate the effect again in Italian. In the present research, we hypothesized that spatial interference could be obtained in Italian by bolstering semantic processing during the task. To this end, a subtle and natural method for increasing semantic processing during reading was used, that is, words with irregular stress were included.
Lexical stress refers to the prominence given to a certain syllable when pronouncing a polysyllabic word. Stress consists of a wide range of phonetic properties, such as loudness, vowel length and pitch. Within some languages, the same syllable may be stressed in most polysyllabic words. In Italian, for instance, about 70% of three-syllable words have stress on the penultimate syllable (Colombo & Zevin, Reference Colombo and Zevin2009; Spinelli et al., Reference Spinelli, Sulpizio and Burani2017). Such words, which have stress on the typical syllable within the language, are said to have regular stress (henceforth ‘regular words’). Others, in which a different syllable is stressed, have irregular stress (henceforth ‘irregular words’). Regular words can be read via sublexical processing, with relatively little semantic activation, based on the statistical–distributional knowledge that readers acquire about their language (Colombo & Zevin, Reference Colombo and Zevin2009; Sulpizio et al., Reference Sulpizio, Burani and Colombo2015). In contrast, irregular words more strongly activate the lexical–semantic representation in order to retrieve the correct pronunciation (Colombo, Reference Colombo1991; Colombo & Zevin, Reference Colombo and Zevin2009; Sulpizio et al., Reference Sulpizio, Burani and Colombo2015). That is, irregular stress can induce deeper semantic processing even in transparent orthographies. For example, age-of-acquisition (a semantic factor) affects the reading of irregular words but not of regular words (Wilson et al., Reference Wilson, Ellis and Burani2012).
Interestingly, when regular words occur in the context of many irregular words, then the regular words are also processed more semantically. For instance, semantic priming typically does not occur in reading aloud Italian words. When irregular words are added to the experimental list, however, semantic priming emerges for both the regular and irregular words (Tabossi & Laghi, Reference Tabossi and Laghi1992; see also Colombo & Tabossi, Reference Colombo and Tabossi1992). Thus, in the present research, we additionally included some irregular words among the cues. If the prior failures to obtain spatial interference in Italian (Estes, Reference Estes2016; Petrova et al., Reference Petrova, Sulpizio, Navarrete, Job, Suitner and Peressotti2013, Reference Petrova, Navarrete, Suitner, Sulpizio, Reynolds, Job and Peressotti2018) were due to insufficient semantic processing of the cue words, then the inclusion of irregular cues should evoke spatial interference in Italian.
9. Experiments 1 and 2
Procedurally, Experiments 1 and 2 were prototypical tests of the spatial interference hypothesis. Single cue words with high (e.g., ‘hat’) or low (e.g., ‘boot’) spatial associations were presented centrally on a computer display, followed shortly (SOA = 150 ms) by an unrelated visual target (■ or ●) appearing at either the top or bottom of the display. Participants’ task was simply to identify whether the target was a square or a circle. The experimental cue words, all of which had spatial associations and regular stress, were taken from Petrova et al. (Reference Petrova, Navarrete, Suitner, Sulpizio, Reynolds, Job and Peressotti2018). As in prior tests of spatial interference, there were congruent trials (i.e., those in which the target appeared in the cue’s associated location, that is, high cue with top target and low cue with bottom target) and incongruent trials (i.e., those in which the target appeared in the opposite location – high cue with bottom target, and low cue with top target). The proportion of spatially congruent trials was 50% in both experiments, so that the cue’s spatial association did not predict the target location. Thus, the experiments were close conceptual replications of Estes et al. (Reference Estes, Verges and Barsalou2008, Experiment 3).
Experiments 1 and 2 were identical, except that they included different filler cues. The irregular fillers in Experiment 1 had high or low spatial associations. That experiment provided strong conditions for obtaining spatial interference because (1) the presence of 50% irregular cues should induce semantic processing of all cues, and (2) the presence of spatial associations in 100% of cues should ensure that those spatial associations are activated during that semantic processing. The irregular fillers in Experiment 2 instead had no spatial associations, thus providing a more conservative test of spatial interference because only 50% of the cues had spatial associations. Thus, if the inclusion of irregular cues is sufficient for obtaining spatial interference in Italian, then a spatial interference effect of similar magnitudes should occur in Experiments 1 and 2. Alternatively, if a high proportion of spatially associated cues is necessary for obtaining spatial interference in Italian, then the spatial interference effect should be larger in Experiment 1 (100% spatial cues) than in Experiment 2 (50% spatial cues). That is, comparison of Experiments 1 and 2 will test whether the proportion of spatially associated cues moderates the effect.
Experiment 2 was preregistered (http://aspredicted.org/blind.php?x=4bb6zh), and all data and code for both experiments are available at the Open Science Foundation (available at https://osf.io/fbm7d/). Given their high similarity, we report Experiments 1 and 2 together.
9.1. Methods
9.1.1. Sampling
Simonsohn (Reference Simonsohn2015) recommended that replication samples should be about 2.5 times larger than the original sample. Given that the present experiments were close conceptual replications of Estes et al. (Reference Estes, Verges and Barsalou2008, Experiment 3), where N = 27, we sought a target N of about 68 participants in each of these two replication studies.
9.1.2. Participants
Students at an Italian university participated in exchange for course credit or a small reimbursement. All participants were native speakers of Italian, and all participated in only one experiment reported herein. Sixty-eight students (43 females, M = 21.76 years, SD = 1.24, range = 19–26) participated in Experiment 1, but three participants whose overall error rate was 20% or more were excluded, leaving 65 valid participants. Seventy students (44 females, M = 21.34 years, SD = 1.31, range = 19–24) participated in Experiment 2, and no participant committed more than 20% errors, so all were included in analyses. In total, then, there were 135 participants included in the analyses.
9.1.3. Stimuli
See the Supplementary Material for the full set of stimuli. Experimental cues were 24 regular words with a high (n = 12) or low (n = 12) spatial association, all taken from Petrova et al.’s (Reference Petrova, Navarrete, Suitner, Sulpizio, Reynolds, Job and Peressotti2018) Experiments 1–4, and selected from a spatial rating pretest (see the Appendix). To note, two of them (chioma, funivia) were added after the pretest. The spatial ratings of the high-association cues (M = 6.22, SD = 0.44, Range = 5.48–6.70) did not overlap with the low-association cues (M = 1.58, SD = 0.20, Range = 1.33–2.00). The same experimental cues were used in both Experiments 1 and 2.
Each experiment also included 24 irregular filler cues, selected from the spatial rating pretest (see the Appendix). In Experiment 1, 12 of the filler cues had a high spatial association (M = 5.81, SD = 0.62, Range = 4.88–6.77), and 12 had a low association (M = 1.85, SD = 0.15, Range = 1.55–2.11). In Experiment 2, all 24 filler cues had neutral associations (M = 3.55, SD = 0.32, Range = 3.03–4.22).
9.1.4. Apparatus
Stimulus presentation, response times (RTs) and accuracy were controlled and recorded by E-Prime 2 (Psychology Software Tools, Inc., Sharpsburg, PA). Participants completed the experiment on a Lenovo notebook running Windows 10 with a 15.6-inch monitor and a display resolution of 1366 × 768 pixels.
9.1.5. Procedure
This research complied with APA ethical standards for the treatment of participants, and it was approved by the ethics committee of the host university. Participants were tested individually in a sound-attenuated, uniformly lit room. They were seated approximately 60 cm from the monitor. Participants initiated each trial by pressing the spacebar, which triggered a central fixation cross that appeared for 250 ms, followed by the cue word, which appeared centrally for 100 ms. After a 50 ms delay, a target object (either a circle or a square) subtending approximately 5° of visual angle appeared at the top or bottom of the screen. Thus, as in Estes et al. (Reference Estes, Verges and Barsalou2008) and Petrova et al. (Reference Petrova, Navarrete, Suitner, Sulpizio, Reynolds, Job and Peressotti2018, Experiments 5–9), the SOA was 150 ms. Cues were presented in black on a white background in Courier New 18-point font. The ‘top’ and ‘bottom’ locations were centered horizontally approximately 9° vertically from the center of the display. Circle and square targets were also used by Petrova et al. (Reference Petrova, Navarrete, Suitner, Sulpizio, Reynolds, Job and Peressotti2018, Experiments 1 and 2). Cue Association (high, low), Target Location (top, bottom) and Target Object (circle, square) were fully crossed and balanced, such that each target object was equally likely to appear at each target location within each cue condition. This counterbalancing yielded eight stimulus lists of 96 trials each, with each participant assigned randomly to one of those eight lists.
Participants were instructed to identify the target object as quickly as possible, without making errors, by pressing the appropriate key (‘C’ or ‘M’, as in Petrova et al., Experiments 4, 6, 7 and 8) on a QWERTY keyboard. Half of the participants responded by pressing the ‘C’ key with their left index finger when a circle appeared on the monitor, and the ‘M’ key with their right index finger when a square appeared on the monitor. The other half were assigned to the opposite mapping.
The experiment consisted of 16 practice trials on which new nonspatial cue words were presented, and two experimental blocks of 48 trials each. Each cue was presented twice, once in each block. Trials were randomly presented within each block. Blocks were separated by a self-paced break, and the order of blocks was counterbalanced across participants. The task took about 7 minutes to complete.
9.1.6. Data Analysis
RTs from trials with incorrect responses were excluded from analyses. Outlying RTs, defined as those more than 2.5 SDs from the participant’s mean, were also excluded from analyses (Experiment 1: 2.84% of trials; Experiment 2: 2.99%). After completion of the experiments, we discovered that we had inadvertently included two irregular words (i.e., ‘aereo’ and ‘sommergibile’) among our experimental cues. We therefore report results with those two cues removed from all analyses.
We combined the data from Experiments 1 and 2 and analyzed them via linear mixed-effects models (LMM) using IBM SPSS Statistics 25. Following Barr et al. (Reference Barr, Levy, Scheepers and Tily2013), we first attempted to fit a maximal random-effects model, with unstructured covariance and random slopes for Congruence across both subjects and items. Because that maximal model failed to converge, we then used a ‘step-down’ strategy to identify the maximal model supported by the data (Barr et al., Reference Barr, Levy, Scheepers and Tily2013; Bates et al., Reference Bates, Kliegl, Vasishth and Baayen2015a; Matuschek et al., Reference Matuschek, Kliegl, Vasishth, Baayen and Bates2017). The maximal convergent model for both error rates and RTs had Congruence and Experiment as fixed effects and random intercepts for subjects only (Matuschek et al., Reference Matuschek, Kliegl, Vasishth, Baayen and Bates2017). The models were specified as:
lmer(Target.ERR ~ Congruence * Experiment + (1 | Subj_n), data = data);
lmer(Target.RT ~ Congruence * Experiment + (1 | Subj_n), data = data).
We dummy coded both Congruence (congruent = 0, incongruent = 1) and Experiment (Experiment 1 = 0 and Experiment 2 = 1).
9.2. Results
Error rates were generally low in both Experiment 1 (overall M = 2.76%, SE = .38) and Experiment 2 (M = 3.40%, SE = .46), and they exhibited no significant effect of Congruence (p = .788), Experiment (p = .141) or their interaction (p = .481). We therefore do not consider error rates in any further analyses.
Response time results are summarized in Table 2, and the spatial interference effect is illustrated in Figure 2. The main effect of Congruence was significant, F(1, 5514.42) = 6.98, p = .008. As predicted, cue words slowed identification of targets in their associated location (i.e., spatial interference). The effect of the Experiment was not significant, p = .453, and the interaction also failed to approach significance, p = .503, thus providing no evidence that the spatial interference effect was moderated by the proportion of spatially associated cues. See the Supplemental Analyses for full details of individual parameter estimates.
Table 2. Mean response times (RTs) and error rates (ERs; with standard deviations in parentheses) as a function of Condition (Incongruent, Congruent) in Experiments 1 and 2 and in the Combined Analysis of Experiments 1 and 2
* p < .05;
** p < .01.
Figure 2. The spatial interference effect in Experiments 1 and 2, and in a combined analysis of Experiments 1 and 2. Bars indicate ±1 SE, corrected for within-participant designs (Loftus & Masson, Reference Loftus and Masson1994).
For thoroughness and transparency, we also conducted several supplemental analyses that were intended to either facilitate comparison to prior tests of spatial interference (i.e., t-tests, ANOVAs, and Bayesian hypothesis tests) or investigate the robustness of the effect (i.e., inclusion of filler trials, counterbalancing checks and alternative outlier detection methods). The outcomes largely align with the results of the linear mixed models reported above. See the Supplementary Material for further details.
10. General discussion
These results replicate the spatial interference effect. Notably, the effect was shown here in Italian. Inducing semantic processing of the experimental cues via a rather subtle manipulation of the regularity of the filler cues was sufficient to reveal spatial interference in an orthographically transparent language, where many prior attempts have failed (see Table 1). The standardized effect size was moderate, with spatial congruence accounting for 6.6% of the variance in target identification times. The raw effect size was 14 ms, which is comparable to the meta-analytic effect size observed in English under otherwise comparable conditions (i.e., 17 ms; Estes & Barsalou, Reference Estes and Barsalou2018). These results provide the first demonstration that spatial interference can be obtained reliably in an orthographically transparent language.
10.1. Re-Weighing the Evidence
The spatial interference effect has been the subject of some controversy, having been demonstrated 14 times by several independent research groups and also having failed to replicate at least 14 times by several other independent research groups (see Table 1). The present research, by adding two successful replications of the effect, does not tip the balance of evidence in favor of the effect’s reliability simply as a matter of score keeping: 16 for the defense to 14 for the challengers. Such simple counting is not how bodies of evidence are evaluated. There are other, more important factors such as methodological fidelity (i.e., ‘closeness’ of the replication attempt), strength of the manipulation, sensitivity of the measurement and statistical power (primarily affected by sample size) to detect the hypothesized effect. Aside from their small samples, prior tests of the spatial interference hypothesis generally were methodologically sound. Therefore, there is little point in trying to identify ‘better’ or ‘worse’ replication attempts among this literature.
The present research does tip the balance of evidence in favor of the effect’s reliability, but for a reason other than simple counting: This research provides the first test of a previously hidden moderator. Estes and Barsalou (Reference Estes and Barsalou2018) noted, post hoc, that most tests of spatial interference in orthographically opaque languages (e.g., English) were successes, whereas most tests in more transparent orthographies (i.e., Italian or German) were failures (see Table 1). However, because orthographic transparency is a property of languages, it cannot be manipulated experimentally, rendering direct tests of this hypothesized moderator impossible. In the present research, this methodological limitation was circumvented by inducing participants to process an orthographically transparent language as if it were an opaque language (i.e., by inducing deeper semantic processing). In this way, a reliable demonstration of spatial interference in a transparent language was obtained. Thus, the present research explains why some prior tests successfully obtained spatial interference and others failed to do so. Consequently, this research strongly supports the reliability of the effect by providing a systematic explanation of the conditions under which it does or does not occur, as described next.
10.2. Theoretical Implications
These results suggest that prior failures to obtain spatial interference in more orthographically transparent languages such as Italian and German (i.e., Estes, Reference Estes2016; Petrova et al., Reference Petrova, Sulpizio, Navarrete, Job, Suitner and Peressotti2013, Reference Petrova, Navarrete, Suitner, Sulpizio, Reynolds, Job and Peressotti2018; Renkewitz & Müller, Reference Renkewitz and Müller2015) were likely due to insufficient semantic processing. Because transparent languages have highly consistent spelling-to-sound correspondences, words can be read with relatively little semantic processing (Bates et al., Reference Bates, Burani, D’Amico and Barca2001; Buchanan & Besner, Reference Buchanan and Besner1993; Burani et al., Reference Burani, Arduino and Barca2007; Frost et al., Reference Frost, Katz and Bentin1987; Katz & Frost, Reference Katz and Frost1992; Kwok et al., Reference Kwok, Cuetos, Avdyli and Ellis2017; Pagliuca & Monaghan, Reference Pagliuca and Monaghan2010; Peressotti & Job, Reference Peressotti and Job2003; Schmalz et al., Reference Schmalz, Marinus, Coltheart and Castles2015). And because semantic processing appears to be necessary for spatial congruence effects to occur (Brookshire et al., Reference Brookshire, Ivry and Casasanto2010; Lebois et al., Reference Lebois, Wilson-Mendenhall and Barsalou2015; Santiago et al., Reference Santiago, Ouellet, Román and Valenzuela2012; Shaki & Fischer, Reference Shaki and Fischer2023a, Reference Shaki and Fischer2023b), these more orthographically transparent languages typically fail to elicit spatial interference.
In contrast, the inconsistency of spelling-to-sound correspondences in opaque orthographies such as English elicits a stronger reliance on semantic processing during word reading (Frost et al., Reference Frost, Katz and Bentin1987; Katz & Frost, Reference Katz and Frost1992; Kwok et al., Reference Kwok, Cuetos, Avdyli and Ellis2017; Schmalz et al., Reference Schmalz, Marinus, Coltheart and Castles2015). Critically, however, in transparent languages, words with irregular stress also induce deeper semantic processing during reading (Tabossi & Laghi, Reference Tabossi and Laghi1992; Wilson et al., Reference Wilson, Ellis and Burani2012). The present experiments demonstrate that, when such irregular words are included among the cue words, spatial interference is also observed in an orthographically transparent language. This observation suggests that prior failures to replicate spatial interference in more transparent languages may well have been attributable to insufficient semantic processing.
Consistent with prior research from other paradigms (Brookshire et al., Reference Brookshire, Ivry and Casasanto2010; Lebois et al., Reference Lebois, Wilson-Mendenhall and Barsalou2015; Santiago et al., Reference Santiago, Ouellet, Román and Valenzuela2012), the present results further suggest that relatively deep semantic processing is necessary for spatial interference. This conclusion also provides a unifying explanation of the previously observed moderators of the spatial interference effect. First, spatial interference does not occur when only a single cue category is used (Gozli et al., Reference Gozli, Chasteen and Pratt2013). Presumably, with only one cue category, the perceptual simulation of that cued category becomes strongly activated within the first few trials of the experiment. After those first few trials, processing of the given scenario or event no longer requires as many neural and/or perceptual resources. Consequently, the cues produce less neural and perceptual competition with the target stimulus, thereby eliminating the spatial interference effect (Ostarek & Vigliocco, Reference Ostarek and Vigliocco2017). Second, spatial interference does not occur with long SOAs (Gozli et al., Reference Gozli, Chasteen and Pratt2013). At short SOAs, the cue word is thought to evoke an attention shift toward the associated location and a perceptual simulation of the denoted object or event. That location-specific simulation, in turn, is thought to perceptually or neurally compete with (or ‘mask’) identification of the visual target in that location. At longer SOAs, however, the perceptual simulation begins to dissipate, leaving visual attention in the associated location without perceptual competition. Third, spatial interference occurs in identification tasks but not in detection tasks (Gozli et al., Reference Gozli, Chasteen and Pratt2013). This is because spatial interference arises from the semantic incongruence of the cue and target, but the detection task does not require it, and hence may not always evoke deep semantic processing of the target. The common denominator among all these known moderators of spatial interference, including orthographic transparency (Estes & Barsalou, Reference Estes and Barsalou2018), is more or less semantic processing. Collectively, these moderations can be summarized as follows: If the cues are processed semantically and the targets are unrelated to those cues, spatial interference tends to occur. If the cues are not processed at a sufficiently deep semantic level, then interference does not occur.
10.3. Future Directions
Six promising directions for further research that may be theoretically informative of spatial interference could be identified. One striking aspect of the spatial interference effect is that it occurs despite the fact that the cue words are entirely unrelated to the target identification task. The task can be completed error free without even reading the cue words, and hence, reading the cues is purely incidental to task performance. In fact, the cue words and their spatial associations do not predict the location or identity of the subsequent target, so reading the cues could not possibly improve performance. Some studies, however, required semantic judgment of the cue words (e.g., Amer et al., Reference Amer, Gozli and Pratt2017; Gozli et al., Reference Gozli, Chasteen and Pratt2013; for related tasks, see also Brookshire et al., Reference Brookshire, Ivry and Casasanto2010; Lebois et al., Reference Lebois, Wilson-Mendenhall and Barsalou2015; Peressotti & Job, Reference Peressotti and Job2003; Petrova et al., Reference Petrova, Navarrete, Suitner, Sulpizio, Reynolds, Job and Peressotti2018; Santiago et al., Reference Santiago, Ouellet, Román and Valenzuela2012; Shaki & Fischer, Reference Shaki and Fischer2023a). It may be informative to examine whether the presence or absence of such semantic judgment moderates the magnitude of spatial interference, especially in orthographically transparent languages, where the cue words may otherwise be processed semantically only weakly.
A second avenue for further research is to re-examine the boundary or generality of spatial interference. In the introduction, based on prior tests of the hypothesis, spatial interference was operationally defined as occurring with nonsemantic targets such as geometric shapes (e.g., circle and square) and alphanumeric characters (e.g., p and q). It remains an open question, however, whether the targets must actually be nonsemantic, or whether they need only be semantically unrelated. For instance, Estes et al. (Reference Estes, Verges and Adelman2015) included targets that were semantically unrelated to the cues, such as the word ‘bird’ followed by an image of a wrench at either the top or bottom of the display. They found significant spatial interference with those semantically unrelated targets. On the other hand, Ostarek and Vigliocco (Reference Ostarek and Vigliocco2017) similarly tested semantically unrelated targets, but there, the targets also had their own spatial associations. For instance, ‘sky’ preceded an image of a hat (which also has a high association). And in that case, there was no spatial interference, although it may simply have been overshadowed by the target’s own spatial association. Thus, it is currently not entirely clear whether spatial interference requires nonsemantic targets or semantically unrelated targets.
Third, the salience of the cues’ spatial associations might be an important direction for additional research. Petrova et al. (Reference Petrova, Navarrete, Suitner, Sulpizio, Reynolds, Job and Peressotti2018), in their Experiment 7, explicitly informed participants that the cue words had spatial associations. Those ‘biased’ instructions, which render spatial associations highly salient, produced the only prior demonstration of spatial interference in an orthographically shallow language (see Table 1). However, when Petrova et al. conducted an exact replication in their Experiment 8, they obtained the exact opposite result, finding instead a tendency toward spatial facilitation (p = .061). In the present research, the salience of spatial associations was manipulated by varying the proportion of cue words that had spatial associations across experiments, but no difference in spatial interference across those experiments emerged. Given this empirical ambiguity, these results collectively suggest that the salience of spatial associations may indeed be relevant to the spatial interference effect, but its effect (if any) appears to be complex.
A fourth direction for theoretical advance is to more thoroughly examine the spatial interference effect in semi-transparent languages such as German. As far as we are aware, only a single test of spatial interference has been conducted in German: Renkewitz and Müller (Reference Renkewitz and Müller2015) failed to obtain spatial interference in German, but their study might be considered underpowered. Semi-transparent languages are theoretically interesting to study because they tend to entail a moderate amount of semantic processing during language comprehension. Would a semi-transparent language such as German produce a spatial interference effect midway between Italian (a transparent language) and English (an opaque language)? Or does the mere presence of some moderate amount of orthographic complexity in the language induce deep semantic processing, such that spatial interference effects are equally large in German and English? Large-scale, cross-language tests of linguistic cueing are needed to address this question.
It would also be theoretically informative to test for spatial interference with the auditory presentation of linguistic cues. Twenty-four of the 28 prior tests of spatial interference (see Table 1) used visual presentation of written cues, as in the present experiments. Given that orthographic transparency is a property of written language – i.e., the consistency of spelling-to-sound correspondence – we see no reason why orthographic transparency would moderate spatial interference with auditory presentation of linguistic cues. So, would spatial interference occur in transparent languages (e.g., Italian) with auditory presentation of cues? By demonstrating that spatial interference can also occur in an orthographically shallow language, the present experiments reveal that it is not orthographic depth per se that moderates spatial interference. Rather, the true hidden moderator is semantic depth, or the extent to which the linguistic cues elicit semantic processing (see also Shaki & Fischer, Reference Shaki and Fischer2023a). And critically, spoken language also elicits varying degrees of semantic processing (Sanford & Sturt, Reference Sanford and Sturt2002). The present experiments thus suggest that spatial interference from auditory cues likely depends on the semantic depth of cue processing both within and across languages. Indeed, of all prior studies of the spatial interference effect (see Table 1), only Bergen et al. (Reference Bergen, Lindsay, Matlock and Narayanan2007) presented the linguistic cues auditorily and they obtained spatial interference twice with literal cues (e.g., ‘The mule climbed’; Experiments 1 and 2), but failed to obtain spatial interference twice with metaphorical cues (e.g., ‘The market sank’; Experiments 3 and 4). More research with auditory cue presentation is needed to disentangle the potential roles of orthographic and semantic depth in spatial interference.
Finally, these results raise implications for linguistic cueing effects more generally, beyond spatial interference. If orthographically transparent languages do not typically induce deep enough semantic processing of cue words to elicit the spatial interference effect, as we argue, then presumably such languages may also fail to elicit the more common spatial congruence effect, whereby cue words instead facilitate perception at the associated location (e.g., Hommel et al., Reference Hommel, Pratt, Colzato and Godijn2001). Aside from the tests of spatial interference in Italian that we reviewed extensively above, our literature search revealed only one other investigation of linguistic cueing in an orthographically transparent language: Ouellet et al. (Reference Ouellet, Santiago, Funes and Lupiánez2010) centrally presented time-related words such as ‘before’ and ‘after’ in Spanish, and then tested perception of visual targets on the left or right of the display. Across three experiments, they obtained spatial congruence effects, such that past-related cues facilitated perception at the left location and future-related words sped perception on the right. At face value, this finding seems to contradict the implication that transparent languages do not typically elicit linguistic cueing effects. Crucially, however, Ouellet et al. explicitly required participants to semantically process the cue words during all three of their experiments. Thus, Ouellet et al. did obtain a linguistic cueing effect in a transparent language, but as in the present experiments, it occurred with relatively deep semantic processing of the cues. As for why Ouellet et al. found a congruence effect instead of interference, we note that their experiments did not have the conditions under which spatial interference tends to occur. Specifically, Ouellet et al. used long SOAs, which are known to elicit facilitation instead of interference (Goodhew et al., Reference Goodhew, McGaw and Kidd2014; Gozli et al., Reference Gozli, Chasteen and Pratt2013). As explained in our introduction, the spatial interference effect tends to occur only with short SOAs (see ‘Defining the Spatial Interference Effect’), before the perceptual simulation of the cue word has dissipated (see ‘Theoretical Explanation of Spatial Interference’). Thus, more research is warranted to investigate more fully the conditions under which linguistic cueing effects in general (i.e., both congruence and incongruence effects) may occur in orthographically transparent languages.
10.4. Concluding Remark
The spatial interference effect has attracted relatively many replication attempts. Given the counterintuitive nature of this effect, such replication attempts are not merely justified but necessary for the integrity of the field. Surprising effects should be subjected to replication attempts because surprising effects are relatively likely to be false-positive, Type I errors (Forstmeier et al., Reference Forstmeier, Wagenmakers and Parker2017). Our knowledge of the underlying process(es) may become deeper and broader only if other, independent researchers continue testing for spatial interference via direct and conceptual replications. Spatial interference is a positive example of how counterintuitive effects – and their subsequent replication failures – can advance theoretical understanding of the phenomenon further than mere confirmations of the effect.
Data availability statement
Experiment 2 was preregistered (available at http://aspredicted.org/blind.php?x=4bb6zh) and all data and code are available at: https://osf.io/fbm7d/.
Funding statement
This research did not receive any specific grant from funding agencies in the public, commercial or not-for-profit sectors.
Competing interests
The authors declare none.
Open access
