The present article, along with that of Rayner, White, et al. (2006)
, provides the first objective measures of the ease with which TL nonwords can be read within sentences. Text including TL nonwords increases reading times by as little as 9% for normally presented text. Given that these sentences effectively include nonwords, it is noteworthy that the language processing system is able to comprehend such sentences in such a relatively short time.
Although we did not use letter substitutions in the present experiments, it is instructive that when letters are substituted rather than transposed, readers take much longer to read sentences (Rayner & Kaiser, 1975
). In the Rayner and Kaiser experiment, when the letter substitutions were visually similar internal letters (so problem
was printed as pncblem
) reading time doubled; when ending letters were substituted (problnc
), reading time also doubled; and when beginning letters were substituted (qroblem
), reading time was 2.5 times longer than normal. When dissimilar letters were substituted for internal letters (prkylem
) or final letters (problky
), reading time tripled; when dissimilar letters were substituted for beginning letters (fyoblem
), reading time quadrupled. In all cases (except for visually similar internal letter substitutions), comprehension also suffered. The fact that letter transpositions are so much easier than letter substitutions demonstrates that the specific letters of a word are critical for identifying what the word is and that readers cannot rely exclusively on context for word recognition. In comparison to letter substitutions, letter transposition makes it much easier for readers to recover what the actual form of the word should be. Obviously, the availability of all of the letters provides veridical information; in addition, it may be easier to identify exactly which letters are incorrect in TL nonwords than in letter substitutions.5
Experiment 1 examined whether lexical difficulty impacted on the ease with which TL nonwords could be read. There were larger effects of transpositions on total reading time for infrequent compared to frequent words. In line with some previous studies (Andrews, 1996
; O’Connor & Forster, 1981
; Perea et al., 2005
), these findings suggest that lexical difficulty does impact on the ability of the word recognition system to process incorrect letter position information. To be clear, letter position encoding is a constituent process of lexical identification. Words that are difficult to process may have an increased dependence on correct letter position information, or else longer reading times for such words may provide more opportunity for processing of precise letter position information. Note that the effect of word frequency on processing of transposed letters reported here only reached significance for the measure of total time, which indicates that word frequency may impact on processing of TL nonwords largely in the later stages of word processing.
The present experiments also indicate that the position of some letters within words is more important than others. The results suggest that the positions of external letters are generally more important for word recognition than the position of internal letters. These findings support previous studies that suggest that external letters are more critical than internal letters for word recognition (Chambers, 1979
; Estes, Allmeyer, & Reder, 1976
). However, it is critical that of the external letters, the position of the word-initial letter is more important than the word-final letter. Furthermore, in the early stages of word processing, word-final letters may be equal in importance to word-internal letters, whereas in the later stages of processing the final letters of words are more critical for word recognition than the internal letters of words. Note that previous studies (e.g., Jordan, Thomas, et al., 2003
) that manipulated both external letters concurrently and recorded only overall reading times could not have identified these differences.
The position of the external letters of words could be more critical for word recognition for either visual or linguistic reasons. The external letters of words have less lateral masking (Bouma, 1973
) and therefore may be more easily identified such that they become very important in identifying letter strings (see account of the SERIOL model below). Alternatively, the external letters of words, perhaps along with word length information, may together be sufficient to allow fast recognition of some words (Clark & O’Regan, 1999
). Note that Johnson et al. (2007)
demonstrated that the position of the word-final letter is important in parafoveal word processing, whereas the results presented here suggest that the word-final letters are more important in the later stages of word processing. Perhaps when processing visually degraded words in the parafovea, lateral masking may have a greater influence on the clarity of letter information such that word-final letters may be easier to process than internal letters in the parafovea.
In addition to the effect of externality on letter position processing, the results of both Experiments 1 and 2 also show an effect of location such that word-beginning letters are more critical for word recognition than word-ending letters. As noted above, for later measures, the word-final letters may be more critical than internal-beginning letters, perhaps because word-final letters may be important in lexical identification. However, at least within the categories of internal and external letters, letter location is clearly important. Previous research has suggested that the beginning portions of words are more important than the final portions (Lima & Pollatsek, 1983
; Mewhort & Beal, 1977
). However, such effects might be explained by differences in processing only the external letters, whereas the results of Experiment 1 suggest that the position of internal letters is also more critical when these are toward the beginning, compared to the end, of words (though the effects are smaller than those for external letters). The results of Experiment 2 suggest that not only are word-beginning letters intrinsically important to lexical identification, but that parafoveal preprocessing is especially critical to processing of word-beginning letters during normal reading.6
One possible explanation for why beginning letters are more critical for word recognition than ending letters is because letters within words are processed serially rather than in parallel, at least for early word processing. The orthographic uniqueness point effects shown by Kwantes and Mewhort (1999a)
would support such a hypothesis. Kwantes and Mewhort (1999b)
proposed a model (LEX) that is based on the notion that in order to identify words, readers search their lexicon one letter at a time until the word can be uniquely identified. Letters at the beginning of a word, then, are especially important for narrowing down the possible set of lexical items. However, it is possible that the serial processing identified in these experiments was specific to the naming task, and the same effects may not occur during sentence reading. For example, Miller, Juhasz, and Rayner (2006)
were unable to replicate Kwantes and Mewhort’s (1999a)
orthographic uniqueness point effect when participants read words embedded in sentences. Alternatively, word-initial letters may highly constrain the number of lexical candidates such that these are more important for word recognition than word-ending letters.
Word recognition models that demand that letters be processed in relation to specific positions simply can not explain why TL nonwords can be processed so easily (Grainger & Jacobs, 1996
; McClelland & Rumelhart, 1981
; Paap et al., 1982
). In contrast, more recent models that enable more flexible letter position coding may be able to account for the ease with which TL nonwords can be identified (Davis, 1999
; Davis & Bowers, 2004
; Gómez et al., 2008
; Whitney, 2001
; Whitney & Berndt, 1999
; Whitney & Lavidor, 2005
). It is important to note that all of these models are based on the assumption that words are processed only foveally. Some accounts have included the variable of which letter is fixated, which has implications for the visual degradation of other letters in the word (Clark & O’Regan, 1999
). However, models of word recognition still have not attempted to explain how initial parafoveal processing prior to fixation is subsequently integrated with foveal processing once the word is fixated. The differences in the effects of transposition between the available and unavailable preview conditions in Experiment 2 highlight how critical such factors are in word recognition processes (see also Rayner, Liversedge, & White, 2006
). Ultimately, a truly comprehensive model of reading would integrate models of word recognition with models of eye movement control (Grainger, 2003
; see also Liversedge & Blythe, 2007
Critically, the present experiments show that in natural reading, whereas some effects have an early influence on behavior (e.g., parafoveal processing of external letter positions), others tend to influence later measures of behavior (e.g., modulation of transposition effects by word frequency). However, current models of word recognition and letter coding simply do not predict such a detailed time course of effects. Furthermore, although TL nonwords provide an opportunity to investigate letter position coding, ultimately the process of recognizing them as words may also entail reevaluation of word identity in relation to sentential context. Clearly, such processes are not necessarily captured by models of word recognition or letter coding. Nevertheless, in the final part of this article we consider whether models that allow flexible letter coding can account for the effects of externality and location on letter position processing reported here. (See Davis and Bowers, 2006
, for further evaluation of these models.)
In the overlap model (Gómez et al., 2008
; Perea & Lupker, 2004
; Ratcliff, 1981
), each letter is activated to the greatest degree within its correct letter position, but activation also spills over into neighboring letter positions as a Gaussian function. However, while the model does give special status to word-initial letters, it treats word-final letters the same as word-internal letters. As the current results indicate, although word-initial letters are most important in visual word recognition, word-final letters are more crucial for word recognition processes than word-internal letters, especially for later measures of processing. In addition, it is not clear whether the overlap model discriminates between the levels of importance of different internal letter positions. Again, the current results indicate that in reading sentences, these letter positions have different levels of importance and should not be treated equally.
The SOLAR model (Davis, 1999
; Davis & Bowers, 2004
) relies on the spatial coding of letter nodes in visual word recognition. According to this model, words with transpositions are similar to their base words because they contain the same letter nodes. The SOLAR model also has difficulty accounting for some of the current findings. In the model, the strength of the activation differs as a function of letter position, with activation levels decreasing systematically from left to right. Thus, the model can account for both the privileged role of word-initial letters and the overall difference between beginning and ending letters. However, the SOLAR model predicts that word-internal letters are more important in visual word recognition than word-final letters, which was not supported by findings for the later measures reported here.7
The SERIOL model (Whitney, 2001
; Whitney & Berndt, 1999
; Whitney & Lavidor, 2005
) codes letter nodes with varying activation levels based on their position within the letter string. Furthermore, ordinal letter pairs are activated as bigrams (e.g., the word judge
activates 10 bigrams: JU, JD, JG, JE, UD, UG, UE, DG, DE, GE
). However, as explained in the introduction, bigrams with less separation are more highly activated. Thus, encoding of letter identity is based on relative letter position rather than absolute letter position. Transposed-letter words are similar to their base words because they differ in only one bigram (e.g., the TL nonword jugde
has all of the bigrams of its base word judge
except for DG
). In contrast to the overlap model and the SOLAR model, the SERIOL model may be able to account for more of the findings presented here. Moving from left to right through the letter string, letters receive progressively less excitatory input, creating a positional gradient across the word. This feature of the model, then, accurately predicts that word-initial letters will receive the most activation. It also accounts for the stronger activation of word-initial letters found near the beginning of the word than near the end of the word. However, the model also specifies that lateral inhibition of adjacent letters can reduce the excitatory input of a letter node. Thus, word-final letters (which do not experience as much lateral inhibition as word-internal letters due to the following space) would have a stronger advantage than word-internal letters.
To summarize, we report two experiments that investigated reading of sentences including TL nonwords. The results of Experiment 1 indicate that letter position processing is more critical for words that are lexically difficult to process. Furthermore, the position of letters within words influences their importance for word recognition and also the time (or stage) at which letters at particular positions become critical for word processing. Both Experiments 1 and 2 show that the position of word-initial letters is especially critical for word processing. In addition, the results of Experiment 2 suggest that processing of words parafoveally prior to fixation not only plays a critical role in processing of word-beginning letters, but also has important implications for the time course and nature of the word recognition process.