The recognition of a printed word such as pot requires the analysis of letter identity (p, o, t) as well as of letter order (that the p goes before both the o and the t). In the absence of an analysis of letter order in the word recognition process, readers would be unable to detect the difference between anagram stimuli such as pot, opt, and top, which share the same letters, but in different configurations. Yet, despite the importance of letter order for disambiguating similar words, recent research has shown that letter position coding is surprisingly imprecise (see Davis, 2005; Grainger, 2008).

Some of the key phenomena in this respect concern the recognition of printed stimuli with transposed letters, such as waht. Despite the transposition of letters in this stimulus, research has shown that it is perceived as being very similar to its base word, what. Using a masked-priming technique, Forster, Davis, Schoknecht, and Carter (1987) established that identity primes (e.g., what–WHAT) and transposed-letter primes (e.g., waht–WHAT) yielded equivalent facilitation on word recognition, and further, that both of these types of primes yielded more facilitation than did replaced-letter primes (e.g., wrut–WHAT). This basic finding has now been observed across a number of the world’s languages (e.g., Lee & Taft, 2009; Perea & Lupker 2004; Schoonbaert & Grainger, 2004).

The transposed-letter benefit observed in masked priming (i.e., primes with transposed letters are more effective than those with replaced letters) has been interpreted as reflecting a degree of perceptual uncertainty in the coding of letter position (e.g., Davis, 2010; Gomez, Ratcliff, & Perea, 2008). Yet, there may be limits to readers’ tolerance for imprecision in letter order during word recognition. For example, research has suggested that tolerance for letter transpositions is reduced when the transpositions occur in external, as opposed to internal, positions (e.g., judeg–JUDGE vs. jugde–JUDGE; Johnson, Perea, & Rayner, 2007). The study of these limits of the transposed-letter benefit is important because it gives rise to a deeper understanding of the abstract representations at the foundation of reading.

One key controversy in this area concerns the impact of letter transpositions when they occur at a morpheme boundary (e.g., urneal for unreal). Christianson, Johnson, and Rayner (2005) investigated readers’ tolerance for these kinds of transpositions in a series of reading-aloud experiments using masked priming. In their third experiment, they compared the effects of letter transpositions on morphologically complex targets (e.g., BOASTER) and morphologically simple targets (e.g., BLUSTER, as blust is not a morpheme in English). The primes were identical to the targets (e.g., boaster, bluster), had a transposed letter at the morpheme boundary (e.g., boasetr, blusetr), or had a letter substitution (e.g., boasler, blusler). Simple-effects comparisons revealed different patterns of priming effects across the two sets of targets: While identity primes yielded facilitation in both conditions, transposed-letter primes facilitated recognition only for the morphologically simple targets, appearing to indicate that readers’ tolerance for letter transpositions is reduced if the transpositions arise across a morpheme boundary. However, the critical interaction between target type (morphologically complex vs. morphologically simple) and prime condition (identity, transposition, or substitution) did not approach significance, indicating an unacceptably high probability that the numerical trends reported may have arisen by chance. Furthermore, though similar patterns of data were evident in the other two experiments that they reported (both of which used compound targets—e.g., sunshine, silkworm), similar difficulties over statistical reliability were also present.

Duñabeitia, Perea, and Carreiras (2007) sought to address the limitations of Christianson et al. (2005) in a series of three experiments conducted in Basque and Spanish. In both of the first two experiments, the authors contrasted masked-priming effects on morphologically complex versus morphologically simple targets. The primes for morphologically complex targets had a transposed or substituted letter across the morpheme boundary (e.g., mesoenro–MESONERO vs. mesoasro–MESONERO). The primes for morphologically simple targets had a transposed or substituted letter in the same location as in primes for the morphologically complex targets (e.g., escobmro–ESCOMBRO vs. escohcro–ESCOMBRO). The results in both experiments showed an interaction between target type and prime type, such that transposed-letter primes yielded facilitation on target recognition (relative to substitution primes) only for morphologically simple targets. The transposed-letter benefit vanished when the targets were morphologically complex words. Both of these initial experiments involved between-target comparisons, in which any number of uncontrolled differences could have existed across the sets of targets. Thus, in a third, more compelling experiment, conducted in Spanish, the authors examined the transposed-letter priming effect on recognition of a single set of morphologically complex targets, when the letter transpositions fell within the stem or across the morpheme boundary. These results revealed a significant 21-ms benefit for transposed-letter primes (relative to letter substitution primes) when the transposition fell within the morpheme, but no benefit (–1 ms) when the transposition fell across the morpheme boundary, although this interaction between prime type and transposed-letter condition missed significance in the analysis by items (p < .11).

The latter study on Spanish word recognition seems to indicate that orthographic coding demands sufficient precision to identify morphemic units, perhaps in order to facilitate the morphemic segmentation processes thought to characterize the initial stages of visual word recognition (e.g., Rastle, Davis, & New, 2004; see Rastle & Davis, 2008, for a review). However, subsequent investigations in English have failed to replicate these findings (Beyersmann, Coltheart, & Castles, 2012; Beyersmann, McCormick, & Rastle, 2013; Rueckl & Rimzhim, 2011). In particular, Rueckl and Rimzhim reported five masked-priming experiments in which they demonstrated persuasively that (a) the processing of a target word is facilitated by the prior presentation of a masked prime with two letters transposed; (b) this facilitation is observed even in cases in which the transposed letters straddle a morpheme boundary; and (c) this facilitation is equivalent when the transposed letters arise within a stem or across a morpheme boundary.

The strength of the findings reported by Duñabeitia et al. (2007) against all of the subsequent literature investigating English has led to the suggestion that important linguistic differences between Spanish and English might account for the discrepant results (Beyersmann et al., 2012; Beyersmann et al., 2013). The work of Velan and Frost (2009) on Hebrew word recognition has persuasively shown that letter transpositions can have very different effects on the recognition of words in languages with qualitatively different morphological characteristics. While Spanish and English have a similar morphological structure, Beyersmann et al. (2012) argued that Spanish is characterized by far greater morphological diversity and productivity than English. They pointed out that unlike English, Spanish morphology is used to express diminutives, augmentatives, pejoratives, and gender-related information (Beyersmann et al., 2012). In light of these differences, it is not unlikely that precision in the orthographic coding of the morphemic boundary may be more important in Spanish than it is in English.

Accordingly, we conducted a masked-priming study in which we investigated the effects of within-morpheme and across-morpheme letter transpositions on word recognition in groups using the Spanish and English languages. In order to avoid the difficulties associated with between-target comparisons, we selected Spanish–English cognates that were identical (e.g., ANTISOCIAL–ANTISOCIAL) or near identical (e.g., INCORRECT–INCORRECTO) as targets. The use of cognates ensured that there could be no idiosyncratic differences in the target stimuli across languages, and critically, also allowed us to perform identical manipulations to the stimuli across language groups when creating the transposed-letter primes. This very tight control over the stimulus characteristics across languages was vital in addressing our central question of whether the languages differ in their sensitivity to letter transpositions within morphemes and across morpheme boundaries. Because of difficulties related to statistical robustness in some of the key previous work, we used a much larger sample than in those studies (cf. Christianson et al., 2005, Exp. 3, N = 27; Duñabeitia et al., 2007, Exp. 3, N = 32).

On the basis of previous findings (cf. Duñabeitia et al., 2007; Rueckl & Rimzhim, 2011), our prediction was that the transposed-letter benefit should be apparent for both language groups when the letter transposition arises within a morpheme, but should vanish for the Spanish-language group when the letter transposition arises across a morpheme boundary.

Method

Participants

The participants in Experiment 1a were 63 students from the University of Salamanca, all of whom were native Spanish speakers. The participants in Experiment 1b were 64 students from Royal Holloway, University of London, all of whom were native English speakers. None of the participants had any history of language or literacy impairment. The participants in Experiment 1a completed the experiment as part of a course requirement, while those in Experiment 1b were paid £5 for their participation.

Materials

The targets were 88 Spanish–English cognates matched very closely on log frequency (Spanish 0.70, English 0.69; t < 1), length (Spanish 9.18, English 8.99; t = 1.03), and N (Spanish 0.34, English 0.35; t < 1). In order to secure sufficient numbers of Spanish–English cognate targets, we included both prefixed (N = 44; e.g., COOPERACIÓN–COOPERATION) and suffixed (N = 44; e.g., ACCESIBLE–ACCESSIBLE) words. We had no theoretical or empirical reason to expect that the nature of affixation would modulate the effects of interest (see also Duñabeitia et al., 2007), and we did not attempt to match the prefixed and suffixed sets.

For each target, two transposed-letter (TL) nonword primes were created by transposing adjacent letters, either within the stem (TL-within) or across the morpheme boundary (TL-across). These transpositions involved identical letters across the Spanish and English primes (e.g., TL-across: atléitco, athleitc; TL-within: atéltico, atheltic) in 97 % of the cases. Two replaced-letter (RL) nonword primes were also created for each target by substituting other letters for the transposed letters, either within the stem (RL-within) or across the morpheme boundary (RL-across). RL primes were created by substituting vowels for vowels, consonants for consonants, ascending letters for ascending letters, and descending letters for descending letters, as far as possible in an identical manner across the languages. No transpositions involved the target’s initial or final letter, and TL-within transpositions never involved the stem’s initial or final letter. In all, 67 % of the transpositions in the TL-across condition and 74 % of the transpositions in the TL-within condition involved a consonant and a vowel. All prime stimuli are included in the Appendix.

Eighty-eight English–Spanish cognate nonwords were also designed to serve as the “No” responses. These nonwords all appeared morphologically complex and mirrored the word stimuli (half prefixed—e.g., ANTITOLM–ANTITOLMO—and half suffixed—e.g., FROMOLISM–FROMOLISMO).

The primes for the nonword targets were constructed in the same way as those for the word targets.

The assignment of primes to targets for each language version of the experiment was counterbalanced over four lists, such that each participant was exposed to all four priming conditions of the experiment but saw each target item only once.

Procedure and apparatus

Stimulus presentation and data recording were controlled by DMDX software (Forster & Forster, 2003). Each trial of the experiment consisted of a forward mask of hash marks presented for 500 ms, followed by a prime in lowercase, presented for four screen refreshes (because of a slight variation in refresh rates across the laboratories, this resulted in prime durations of 66 ms in Exp. 1a and 57 ms in Exp. 1b).Footnote 1 This lowercase prime was then masked by presentation of an uppercase target that remained on screen until participants decided whether the item was or was not a word in Spanish (Exp. 1a) or in English (Exp. 1b). The stimuli were presented on CRT monitors, and responses were collected using the keyboard. The target stimuli were presented in a different randomized order for each participant and were preceded by 13 practice trials constructed in a similar fashion to those in the main experiment.

Results

Reaction times (RTs) for correct responses were cleaned to remove outliers. Two of the target items were removed from Experiment 1b (COAUTHOR and COPILOT) because they yielded more than 40 % errors. Data points greater than 2,000 ms were also removed from each experiment. This led to the exclusion of 30 data points in Experiment 1a (0.54 % of the data) and 67 data points in Experiment 1b (1.2 % of the data). Data from all participants were retained in both experiments.

The data in both experiments were analyzed by subjects and by items in three-factor analyses of variance (ANOVAs). Condition (TL or RL) and Position (within morpheme or across morphemes) were treated as repeated factors in both analyses. Affix Type (prefixed or suffixed) was also included in the analysis, as a repeated factor in the by-subjects analysis and an unrepeated factor in the by-items analysis. List was treated as an unrepeated factor in both analyses. Our analyses focused on the effect of condition and its interaction with position across languages.

The latency analyses from Experiment 1a (Spanish) revealed a main effect of condition, in which responses to words preceded by TL primes were faster than those to words preceded by RL primes [F 1(1, 60) = 18.87, MSE = 5,855.82, p < .01; F 2(1, 80) = 24.71, MSE = 2,933.85, p < .01]. Critically, this main effect was not modulated by position [F 1(1, 60) < 1; F 2(1, 80) < 1], and no three-way interaction emerged between condition, position, and affix type [F 1(1, 60) < 1; F 2(1, 80) < 1]. The accuracy analyses from Experiment 1a revealed no significant effects of condition, nor any interactions between condition, position, and affix type (all ps > .10).

Like the Spanish data, the latency analyses from Experiment 1b (English) revealed a main effect of condition, in which responses to words preceded by TL primes were faster than those to words preceded by RL primes [F 1(1, 60) = 13.12, MSE = 3,507.56, p < .01; F 2(1, 78) = 6.82, MSE = 4,014.36, p < .02]. Critically, this main effect was not modulated by position [F 1(1, 60) < 1; F 2(1, 78) < 1], and no three-way interaction emerged between condition, position, and affix type [F 1(1, 60) = 3.59, p = .06; F 2(1, 78) = 2.21, p = .14]. The accuracy analyses from Experiment 1b revealed no significant effects of condition, nor any interactions between condition, position, and affix type (all ps > .18).

The RT and error data for both experiments are shown in Table 1. We display the critical Condition × Position contrast broken down by affix types for interest, although it is important to remember that we did not attempt to match the prefixed and suffixed targets.

Table 1 Reaction times (in milliseconds) and percentages of errors (in parentheses) for Experiments 1a and 1b, shown by condition (transposed-letter [TL] or replaced-letter [RL]), position (within or across morphemes), and affix type (prefixed or suffixed)
Full size table

Discussion

Recent studies of the transposed-letter benefit have provided evidence central to our understanding of the basic processes underlying the early stages of visual word recognition. The purpose of this study was to investigate the transposed-letter benefit in the context of the recognition of morphologically complex words. Specifically, we wished to investigate the impact of letter transpositions when they arise within morphemes or across morpheme boundaries. This is an issue that has seen considerable controversy in recent years, with some studies reporting a reduction in the transposed-letter benefit when letter transpositions arise across morpheme boundaries (e.g., Duñabeitia et al., 2007), and others reporting no difference in the magnitudes of the transposed-letter benefit as a function of the position of the transposed letter (e.g., Beyersmann et al., 2012; Beyersmann et al., 2013; Rueckl & Rimzhim, 2011). Critically, Duñabeitia et al. conducted their most compelling study in Spanish, while the other statistically reliable studies have used English stimuli, leading to speculation that there might be a cross-linguistic difference in orthographic coding (Beyersmann et al., 2012).

In order to assess this hypothesis, we created parallel Spanish and English experiments, using Spanish–English cognate stimuli matched very closely across languages (e.g., IDEALISTA–IDEALIST; ACCIDENTAL–ACCIDENTAL). These target stimuli were preceded by masked primes containing transposed letters within the stem morphemes or across morpheme boundaries, and the transposed-letter benefit was calculated relative to a replaced-letter priming condition in which letters were replaced within the stem morphemes or across morpheme boundaries. By using Spanish–English cognate stimuli that were identical or virtually identical, we ensured that any differences observed across our Spanish and English experiments could not be attributed to idiosyncratic aspects of the items chosen, and instead would need to be attributed to a genuine cross-linguistic difference in orthographic coding.

The results were unambiguous in showing transposed-letter benefits in both Spanish and English that were not modulated by the position of the transposed letter in the prime stimulus. For both language versions of the experiment, the sizes of the transposed-letter benefit were equivalent when the transposed letter occurred within the stem and when it occurred across the morpheme boundary. These results are inconsistent with the findings reported by Duñabeitia et al. (2007), but are consistent with all of the statistically reliable English studies (Beyersmann et al., 2012; Beyersmann et al., 2013; Rueckl & Rimzhim, 2011). Because our stimuli were so closely matched across language versions of the experiment, we can be assured that there is no difference between Spanish and English in the coding of orthographic information, as had seemed to be implied by the previous literature.

Our findings therefore add to the growing body of literature suggesting that the transposed-letter benefit is not affected by the position of the transposed letters relative to the morpheme boundary. The work of Duñabeitia et al. (2007) is a clear outlier in this body of literature. In light of our findings, we believe that the findings of Duñabeitia et al. reflected idiosyncratic properties of the stimuli or the participants, or a Type I error.