As White, Johnson, Liversedge, and Rayner (2008) reported, raeding wrods with jubmled lettres involves some cognitive cost in terms of reading time, but participants can read the words correctly. Indeed, it was more than 50 years ago that Bruner and O’Dowd (1958) found that nonwords created by letter transpositions (e.g., jugde) resembled their base words to a large degree (see also O’Connor & Forster, 1981, for early evidence). The transposed-letter effect has been obtained across a variety of paradigms, not only in other Indo-European languages (e.g., in Spanish, Perea & Lupker, 2004; in French, Schoonbaert & Grainger, 2004) but also in other families of languages (e.g., in Basque, Perea & Carreiras, 2006; in Japanese kana, Perea & Pérez, 2009; in Thai, Winskel et al. in press; in Korean Hangul, Lee & Taft, 2009).

The above-cited cross-linguistic evidence may be taken to indicate that letter position coding is an inherent characteristic of any orthographic system—and as such is considered in the current input-coding schemes of visual-word recognition (e.g., spatial-coding model, Davis, 2010; SERIOL model, Whitney, 2001; open-bigram model, Grainger & van Heuven, 2003; overlap model, Gomez, Ratcliff, & Perea, 2008; noisy Bayesian Reader model, Norris, Kinoshita, & van Casteren, 2010). However, in a previous study, Velan and Frost (2007) argued that letter transpositions may merely “reflect the specificities of the lexical structure of a given language” (p. 914) rather than being a universal phenomenon (see also Frost, in press). The evidence for their argument came from an experiment using rapid serial visual presentation (RSVP) of sentences, with each word presented for 200 ms. The experiment was conducted both with English sentences and with Hebrew sentences on Hebrew–English bilinguals; in half of the cases, the sentences were correctly written (e.g., “My sister accused me of lifting some skirts from her closet”), whereas in the other half, three words were misspelled via adjacent transpositions of internal consonants (e.g., “My sitser accused me of litfing some skitrs from her closet”). The participants’ task was to reproduce the whole (correctly spelled) sentence without replicating the actual transpositions. For the English sentences, the participants were able to produce the target words around 85 % of the time, regardless of whether several words were presented jumbled or intact (84 % vs. 86 %, respectively); similarly, the participants were able to reproduce around 81.5 % of the words in the sentences, regardless of whether the sentences had jumbled words or were presented intact (81 % vs. 82 %, respectively). That is, there was very little cost produced by the jumbled words in English, consistent with the accuracy scores in eye movement experiments that had employed transposed-letter words embedded in sentences (White et al., 2008). But the critical finding was that, for the Hebrew sentences, recollection of the target words was around 25 % lower when three of the words had letter transpositions in the consonantal root than when the sentences were spelled correctly (59 % vs. 84 % for words in the sentences with jumbled letters and the intact sentences, respectively); likewise, the percentage of recollection of the words in the sentences was also substantially higher when the sentences were presented intact than when the sentences had jumbled words (81 % vs. 62 %, respectively). That is, Hebrew readers had some difficulty reconstructing the actual transpositions in the Hebrew sentences. Velan and Frost (2007) concluded that “effects of letter transposition probably reflect the principles of defining lexical space and lexical organization, and do not emerge from the peripheral registering of letters in alphabetic orthographies” (p. 916). Their view is that lexical space in Semitic languages would be structured according to the morphological roots, whereas Indo-European languages would be structured according to full orthographic/phonological forms (see Frost, Kugler, Deutsch, & Forster, 2005; Perea, abu Mallouh, & Carreiras, 2010, for parallel evidence from another Semitic language, Arabic). Thus, the transposition of two letters of the consonantal root in Semitic languages would make it difficult to access the actual word.

Velan and Frost (2011) successfully replicated their 2007 RSVP experiment with a new set of English sentences and a new set of Hebrew sentences with root-derived words. Importantly, they also included a third set of Hebrew sentences in which the transposition occurred in morphologically simple words from a non-Semitic origin (e.g., agrtl [a vase]). In the latter case, there was only a small cost (around 4 %), which was similar to that found with English sentences. Velan and Frost (2011) explained this very small reading cost as indicating that “Hebrew words that are morphologically simple, do not contain a root, and do not have any internal structure, are lexically organized by orthographic neighbourhoods just like base-words in English” (p. 152). In this respect, it is important to note that root-derived words in Hebrew have a “well-defined set of conditional probabilities that rigidly determine few open slots for the consonants of the root only” (Velan & Frost, 2011, p. 153), which implies a “high saliency of the root morpheme” (see Frost, in press, for further discussion). Indeed, this saliency may apparently be picked up by learners of Hebrew as a second language (L2). Frost et al. (2005, Exp. 3) reported that native English speakers whose L2 was Hebrew showed the same pattern of masked priming effects as native speakers of Hebrew (i.e., identity but not form priming).

Here we examined whether the special status of root-derived words in Hebrew (or Arabic) can be generalized to another Semitic language, Maltese. There are two reasons why Maltese could be a potentially interesting test bed for the results found in other Semitic languages. First, unlike Hebrew and Arabic, Maltese represents both consonants and vowels in the orthography, which employs the Latin alphabet. Second, Maltese differs from Hebrew or Arabic in that it has a very productive non-Semitic (mostly Romance) morphology in addition to the Semitic component (see Mifsud, 1995a, for a thorough descriptive characterization).

As in other Semitic languages, many verbs and (some) nouns in Maltese can be represented as a root and a word pattern. For instance, the Maltese word miktub [“written”] would be formed by the Semitic root k.t.b (with the meaning of “marking, writing”) and the word pattern mi--u-. Experimental evidence suggests that, similarly to Hebrew and Arabic, the consonantal root in Maltese plays an important role in lexical access (see, e.g., Twist, 2006; Ussishkin & Twist, 2009). However, there are reasons to believe that Maltese root-and-pattern morphology is no longer truly productive. For instance, as Hoberman and Aronoff (2003) note, the most productive derivation rule in the verbal component of Maltese, accounting for many novel formations, is based on the Italian -are and operates on borrowings. Hence, from the borrowed (English) word monitor, via initial germination and suffixation of -ja, we obtain the verb immoniterja (“to monitor”). While this is far from the only case, it is illustrative of a general tendency in Maltese morphology to exhibit productive concatenative, rather than root-and-pattern, word formation processes.

Thus, despite the evident psychological importance of the root (Twist, 2006), words formed out of Semitic roots appear to constitute a fixed list, with no novel formations (Mifsud, 1995b). As was shown in a recent study, based on an exhaustive survey of standard lexicographic resources (Spagnol, 2011), roots in the verbal component occur on average in only two out of ten possible binyanim (range: 1 to 5). There is, furthermore, a marked asymmetry among the binyanim in their frequency of occurrence, with over 75 % of all verbs being in binyanim I, II, and V. This need not imply that Semitic words are infrequent—indeed, the materials used in our study include a number of very frequent formations, as we will show below. However, frequency does not imply productivity (e.g., Baayen, 2009).

Given the strong evidence for the hybrid nature of the Maltese morphological system (Fabri, 2009), as well as the differences in the writing system as compared to other Semitic languages, the question arises whether the consonantal root plays as crucial a role in reading as it does in Hebrew and Arabic. In the present experiment, we employed the same RSVP technique and procedure as had Velan and Frost (2007, 2011). The only difference was that instead of presenting sentences in English versus sentences in Hebrew, we presented sentences in English versus sentences in Maltese. The letter transpositions in the Maltese sentences always involved the transposition of two consonants of the root in words of Semitic origin—as in the Velan and Frost (2007, 2011) experiments. For comparison purposes, we employed the same set of 20 English sentences employed by Velan and Frost (2011); for these sentences, we expected to find little or no differences between the sentences correctly spelled and the sentences with two transposed-letter words.

If the pattern of data in Maltese sentences were similar to that from Hebrew sentences (i.e., a substantial reading cost in sentences with jumbled root-derived words), this would reinforce that view that Semitic morphology governs the way that words are processed, regardless of the presence or absence of vowel information in the orthography and/or the unproductivity of root-based morphological processes.

Alternatively, if the pattern of data in Maltese were similar to that found in English and with morphologically simple words in Hebrew (i.e., little or no reading cost in sentences with jumbled root-derived words in Maltese), this would raise at least three possibilities. One would be that the Semitic root in Maltese, despite evidence for its role in lexical access, simply does not have the saliency that it has for Hebrew readers (Velan & Frost, 2011). A second possibility, clearly related to the first, is that Maltese does not evince the same rigid letter position coding of Hebrew and other Semitic languages; that is, the root has limited utility in determining transitional probabilities in Maltese words. Finally, orthographic differences (the presence of vowels) may also play a role.

Method

Participants

A group of 20 students from the University of Malta took part in the experiment. They were recruited from linguistics courses on Maltese at the undergraduate and graduate levels. All of them reported having Maltese as their mother tongue and using Maltese on a daily basis. They also were highly fluent in English—which is the official second language (after Maltese) in Malta; all students at the University of Malta have their instruction in English, unless the instruction is part of a course related to a specific language (Maltese, French, Spanish, etc.). Most students reading for a degree in Maltese will also have a second subject, for which the instruction is likely to be in English. Apart from an early exposure to English on a day-to-day basis being likely for a large proportion of Maltese children, the language forms part of their formal education from the first year of primary education (age 5) in both state-run and independent (private) schools.

Materials

We created 20 sentences in Maltese such as the one shown below.

figure b

In each sentence, we transposed two adjacent middle consonants from the root letters of two of the words (e.g., libset → lisbet and iswed → iwsed; the roots are l.b.s and s.w.d, respectively). The average number of letters of the target words was 5.8 (range: 5–8), the mean word frequency of the root used in the target word (i.e., the number of times the root occurred in any morphological derivation) was 28.27 per million tokens (range: 0.14–480), and the mean number of orthographic neighbors of the target words was 3.8 (range: 1–13). Neighborhood density and frequency were calculated on the basis of a word list compiled from the 99.2-million-word corpus of the Maltese Language Resource Server (MLRS; available at http://mlrs.research.um.edu.mt). As in the Hebrew experiments, words with transposed letters never appeared consecutively. The individual sentences, as well as a spreadsheet containing frequency characteristics of the target words in Maltese, may be downloaded along with this article. In addition, we employed the 20 sentences in English from Experiment 1 of Velan and Frost (2011). For both the Maltese and the English sentences, we created two counterbalanced lists of 10 intact sentences and 10 sentences with jumbled words (i.e., if a given sentence was presented intact in List 1, it would be presented with jumbled words in List 2, and vice versa). Thus, each participant was presented with 20 Maltese sentences in one block (10 intact and 10 with jumbled words), and 20 English sentences in the other block (10 intact and 10 with jumbled words). There were two blocks in the experiment: half of the participants were initially tested with the Maltese sentences, and then with the English sentences, while the other half were tested in the opposite order.

Procedure

The participants were tested in groups of three or four in a quiet room. Presentation of the stimuli was controlled by Windows-based computers using DMDX (Forster & Forster, 2003). Each trial began when the participant pressed the spacebar. Then, each word of the sentence was presented for 200 ms at the center of the screen. As in a typical RSVP experiment (see Forster, 1970), the participants were instructed to write down the sentence after each trial, or at least the words that they could see in the order that they were presented. The participants wrote the sentences down on paper; they had been alerted that some of the sentences could involve jumbled words and that they had to produce the sentences with correctly spelled words. The instructions and examples in the Maltese block were given in Maltese, while the examples in the English block were given in English. The sentences within each block (English or Maltese) were presented in a randomized order for each participant.

Results

As in the Velan and Frost (2007, 2011) RSVP experiments, we computed, for each participant in each language, two dependent variables: on the one hand, we calculated the overall percentages of correctly reported words, both in intact sentences and in the sentences containing jumbled words; on the other hand, we computed the percentages of target words produced in each language (20 target words in each language), which had been presented either intact or jumbled in the RSVP sentences. The averages per participant in each language for intact sentences and for sentences with jumbled words are shown in Table 1.

Table 1 Percent report of all words (top) and of target words (bottom) in Maltese and in English, with intact sentences and with sentences with jumbled words
Full size table

Percent report of all words

The percent reports of all words were very similar for intact sentences and for sentences with jumbled words, both in English (88.7 % vs. 85.6 %, respectively; F < 1) and in Maltese (88.5 % vs. 88.8 %, respectively; F < 1). Note that the percent reports of all words for the intact sentences (i.e., the “baselines”) were also similar for English and Maltese sentences, F < 1.

Percent report of target words

The participants were able to reproduce the target words, in a similar proportion of cases, when these words were presented intact and when they were presented jumbled, both in English (78.0 % vs. 76.5 %, respectively; F < 1) and in Maltese (81.0 % vs. 81.5 %, respectively; F < 1). There were no significant differences between the percent reports of target words for the intact sentences in English and Maltese (F < 1).Footnote 1

Discussion

Unlike other Semitic languages, Maltese uses the Latin script, does not omit vowel information in print, and also has productive non-Semitic morphology. This allowed us to examine, in a writing system similar to that of Indo-European languages, whether Maltese–English bilinguals have difficulty producing the target words in RSVP in Maltese sentences when two root letters from a Semitic Maltese word were transposed. The findings were very clear: The participants were able to reproduce these words, regardless of whether or not the words were misspelled. This pattern of data closely resembled that of English sentences, thus successfully replicating the results of Velan and Frost (2007, 2011) in English. Therefore, unlike the Hebrew data reported by Velan and Frost (2007, 2011), in which there was “a dramatic drop in performance” when the sentences included root-derived jumbled words, here we found no signs of a parallel effect with Maltese root-derived words (see Table 1). Thus, readers of Maltese, a Semitic language, do not have much trouble reading transposed words of Semitic origin—or rather, do not have trouble reconstructing sentences containing jumbled root-derived words (e.g., libset → lisbet).

In the introduction, we suggested that such a finding might have three possible explanations. One is that differences in the orthographic systems of Maltese and other Semitic languages—especially the presence of vowels—might help to explain the results. However, the mere omission of (short) vowels in Hebrew orthography cannot be the only cause of the difficulties in reading transposed text, because—similarly to the present experiment in Maltese—Velan and Frost (2011) demonstrated that Hebrew readers can easily reconstruct transposed-letter words in an RSVP task when the words are morphologically simple. This suggests that part of the explanation must lie in the differences between the morphological systems of the languages in question. Here, we considered two related factors.

One possibility is that the lack of productivity of root-and-pattern morphology in Maltese may result in lower saliency of the root for Maltese readers. Support for this claim comes from the observation that Maltese has a highly productive non-Semitic concatenative morphology (e.g., Mifsud, 1995a, 1995b; Spagnol, 2011), in addition to the (diachronically prior) root-and-pattern morphology characteristic of Semitic languages. However, this is unlikely to be the whole story, given previous experimental work that has suggested that the root plays a role in lexical access (e.g., Twist, 2006; Ussishkin & Twist, 2009).

Perhaps more importantly, the centrality of the root in Hebrew and Arabic has been cited as a crucial factor in determining transitional probabilities between letters. To explain the difficulty of reproducing jumbled root-derived words, Frost (in press) indicated that, because of the distribution of transitional probabilities between letter sequences, the Hebrew reading system focuses “on a specific subset of letters and is rigid regarding their position”—the consonantal root, in particular.

We believe that the apparent divergences between the RSVP data with Hebrew versus Maltese sentences with morphologically complex target words are due to a combination of the characteristics of the Hebrew writing system with aspects of the Semitic morphology of Hebrew versus Maltese. In particular, the presence of a more flexible process of letter position coding with root-derived words in Maltese than in Hebrew (i.e., two Semitic languages) provides some empirical support for the claim that the flatter the distribution of transitional probabilities of letter sequences, the more the orthographic code will be flexible in letter position (see Frost, in press). Future computational and empirical studies should examine in detail the transitional probabilities in different languages and how they affect the process of letter position coding across languages.

In sum, we believe that further research in Maltese, a Semitic language in which both consonants and vowels are written down (as in Indo-European languages) using the standard Latin alphabet, and which, in addition, has a mixed morphology, opens up a window of opportunity to examine the intricacies of Semitic morphology and its similarities and differences with respect to Indo-European morphology.