A dissociation between orthographic awareness and spelling production

In this study, two nonword spelling and two orthographic awareness experiments were used to examine people’s production and awareness of sound–spelling relationships. The results of the nonword spelling experiments suggest that, in general, people use phoneme–grapheme sized relationships when spelling nonwords. Alternatively, the results of the orthographic awareness experiments suggest that, under some circumstances, people can use larger sized sound–spelling relationships when judging how frequently subsyllabic relationships occur. Together the results suggest that there is a dissociation between sound–spelling production and sound–spelling awareness tasks, and the size of the sound–spelling relationships that people use varies under different tasks and task conditions. Production and awareness of spelling are two core components of spelling ability. They allow people to write and proofread text. Although intricately linked, the two components are not identical. This can be seen most obviously from young children’s spelling. When young children spell, they often use simple single-letter, single-sound transcriptions for words, even when more orthographically typical responses exist (e.g., Treiman, 1993). Similar comparisons with adults are more difficult because adults produce many fewer errors when proofreading and writing. This may be one reason why there is much less data in this area than similar areas, such as reading. However, for a better understanding of adult spelling ability, a systematic study of the relationship between spelling awareness and production is necessary. According to the dual route model of spelling (e.g., Ellis, 1982, 1984; Kreiner, 1992), one of the differences between nonword spelling and orthographic awareness is that nonword spelling requires the ability to segment larger syllabic sounds into smaller sound segments and then apply phonology to orthography  2002 Cambridge University Press 0142-7164/02 $9.50 Applied Psycholinguistics 23:1 44 Perry et al.: Spelling and orthographic awareness translation rules on each smaller segment. If people segment syllables when spelling nonwords, then the size of those segments is important, for at least two reasons. First, the size of the segments provides a constraint on the sound–spelling relationships that people use to spell nonwords. If, for example, people segment syllables into an onset-rime structure, then it would be expected that nonword spelling would be influenced by the frequency of the relationships based on those units. Alternatively, if people segment syllables into phonemes, then the relationships used would be based on phoneme–grapheme units. Second, the way in which syllables are segmented affects how the complexity of the sound–spelling domain should be measured. If people predominantly use phoneme–grapheme relationships when spelling, then factors determining the complexity of the sound–spelling relationship would need to be measured at the phoneme–grapheme level. Alternatively, if people use larger sound–spelling relationships (e.g., rime–body), then complexity would need to be measured at the larger level. In terms of small sized relationships, Kreiner (1992) found evidence suggesting that people are sensitive to phoneme–grapheme sized sound–spelling relationships when spelling. In Experiment 2 of his study, he measured the amount of time it took people to spell words aloud. The results indicated that people were faster at spelling words that had phoneme–grapheme relationships that could be spelled in only a few ways, compared to words that had phoneme– grapheme relationships that could be spelled in many ways. This finding suggests that the inconsistency of phoneme–grapheme relationships is an important factor in spelling aloud tasks. In terms of larger sized sound–spelling relationships, Nation (1997) found that children were more likely to correctly spell words with many rime neighbors than words with few rime neighbors. She argued that this meant that children are able to use relationships larger than the phoneme–grapheme size when spelling nonwords. Apart from data from spelling production tasks, data from orthographic awareness tasks (e.g., Cassar & Treiman, 1997; Siegel, Share, & Geva, 1995) have also been used to help understand processes that underlie spelling and orthographic development. Two types of orthographic awareness tasks that have been used are the orthographic choice task and the auditory–orthographic choice task. In the orthographic choice task, people are presented with a list of nonword pairs and asked to choose one item from each nonword pair based on a word likeness criterion. The auditory–orthographic choice task is identical to the orthographic choice task except that a nonword is presented auditorily before the presentation of each orthographic pair. An interesting aspect of the results from the orthographic and auditory–orthographic choice tasks is that they are not always the same as those found in spelling production tasks. Cassar and Treiman (1997), for instance, noted that children’s awareness of doubling constraints in medial and final letter positions differed from error patterns found in spelling production. In their orthographic choice experiment, the children exhibited a similar level of accuracy when judging whether a letter could be doubled in a medial or a final position. Conversely, in a number of previous spelling studies (Stage & Wagner, 1992; Treiman, Berch, & Weatherston, 1993) children tended to have more difficulty spelling medial than final letters. Such a pattern suggests Applied Psycholinguistics 23:1 45 Perry et al.: Spelling and orthographic awareness the existence of a difference between the levels of children’s orthographic awareness and spelling production. Although a number of results from orthographic awareness and spelling tasks were reported individually, no study with adults systematically investigated the similarities and differences between the two types of tasks. Finding such differences would suggest that component processes involved in spelling may differ critically, depending on whether participants produce spellings or merely judge their legality. Furthermore, if people are aware of common sound–spelling relationships but do not use them when spelling, it suggests that processes related to more than simple relationship complexity cause spelling to be more difficult than reading. In this case, it suggests that there are certain difficulties associated with the production of spelling that are not associated with sound–spelling awareness. The main objective of the following experiments was to examine two interrelated factors: orthographic awareness and people’s production of sound–spelling relationships. Here we assume, based on child and adult spelling data (see Treiman, 1993, for a review of child spelling data, and Kreiner, 1992, for some adult spelling data), that smaller sound–spelling units tend to be easier to spell than larger ones and that the difficulty of task conditions can modulate the extent to which people produce orthographically typical answers compared to simply the most frequent sound–spelling correspondences (see Brown & Deavers, 1999, for a task condition manipulation in reading). We also assume that people’s orthographic awareness is not necessarily the same as their ability to produce sound–spelling relationships (or to read; see Siegel et al., 1995, for a dissociation of these two variables with dyslexics and normal readers). Note that orthographic awareness and spelling production are unlikely to be completely independent in literate adults, however, or people would produce nonword spellings with common sound–spelling relationships but orthographically atypical sequences (e.g., cwac). Thus, it is assumed that in spelling and orthographic awareness tasks the integration of knowledge involving both sound–spelling relationships and orthographic awareness must occur. Four experiments were used to allow these two factors to be investigated on a continuum. In the first experiment, participants were asked to spell a list of nonwords. In the second experiment, participants spelled the same nonwords, but the task conditions were changed by deliberately asking the participants to try and produce the statistically most common sound–spelling patterns they knew. The idea of the first two experiments was to examine the type of sound– spelling relationships people use when spelling nonwords, and to see whether those relationships can be strategically biased by changing task conditions. The third and fourth experiments examined people’s awareness of sound–spelling relationships using two orthographic awareness tasks. In the third experiment an auditory–orthographic choice task was used. The fourth experiment was identical to the third except that an orthographic choice task was used. In the third and fourth experiments, we assume that the correspondences that people use may be biased by sound–spelling relationships rather than only orthographic patterns (particularly when participants are asked to judge orthographic patterns based on sound criteria). Here, if people’s orthographic awareness and ability Applied Psycholinguistics 23:1 46 Perry et al.: Spelling and orthographic awareness to produce sound–spelling relationships never converges (i.e., if people’s orthographic awareness units are bigger than those that they used to spell), then we would expect that incorporating smaller units of spelling–sound knowledge when performing such tasks would reduce people’s potential use of larger orthographic units. Together, the idea is that neither spelling nor orthographic awareness tasks tap processes that are completely independent of each other, and thus that each task allows a different view on the processes discussed here. EXAMINING SOUND–SPELLING RELATIONSHIPS WITH NONWORDS Before we describe the experiments, it is useful to detail the methodology we used for examining the results from the nonword spelling experiments. Perhaps the biggest problem with examining the results of nonword spelling experiments is that the answers people can give for any given stimuli are often fairly unconstrained. That is, given a single nonword phonology, many answers can be given. Note that this problem essentially mirrors that found when examining nonword reading. In that task, people may give answers based on single grapheme–phoneme relationships (regularity), answers based on body–rime relationships (consistency), a combination of the two, or some other form (see Andrews & Scarratt, 1998, for a discussion). Thus, we need some way of examining nonword answers so that underlying processes can be examined. One potential method for analyzing the results is to examine the distribution of answers based on different sound segmentations that can be used. Let us assume that the phoneme is the smallest form of segmentation. If the largest unit into which people can segment sound is the phoneme, then we would expect the simplest form of phoneme–grapheme spellings to be given. Take the nonword /jaIt/ as an example. When broken into its individual phonemes, it is /j/, /aI/, /t/. If the most frequent set of phoneme–grapheme correspondences is applied to these, then the relationships are /j/ → j, /aI/ → i.e, /t/ → t. If the letters are assembled into a contiguous string, they form the nonword jite. Thus, based on phoneme–grapheme frequency, this is the most common spelling that would be expected. Of course, on the basis of individual phoneme–grapheme relationships, some words are also spelled with the /aI/ → igh relationship. If the phonemes people spell nonwords with are susceptible to probabilistic influence, then some of the answers would also be likely to be written as jight. It would still be expected that jite would be more common than jight, however. The idea here is that if people spell nonwords using the most common graphemes, it suggests that phoneme–grapheme sized sound–spelling relationships are important. Let us now assume that people can process larger units of sound than the phoneme and that these larger units also influence people’s spelling. The rime appears to be a relevant unit in people’s speech perception (e.g., Treiman & Danis, 1988), and therefore seems a reasonable choice as a possible larger unit that people use when spelling. Here /jaIt/ would be broken into /j/ and /aIt/. The most common orthography that corresponds to /aIt/ is -ight (e.g., fight). If people use these larger relationships, then the distribution of answers that they give would be expected to differ compared to the distribution if only phoneme– grapheme relationships are used. Here, it would be expected that jight would be Applied Psycholinguistics 23:1 47 Perry et al.: Spelling and orthographic awareness given as an answer, even though the /aI/ → igh relationship is a relatively uncommon spelling pattern, in terms of phoneme–grapheme relationships. In this case, because the -ight body is more common than -ite, it may influence the nonword spellings that people use. Thus, if the distribution of answers was such that jight was commonly given, then it would suggest that not only phoneme– grapheme relationships play a role in people’s nonword spelling, but that the rime plays an important role, too. In summary, the idea behind the analysis of the nonwords reported here is to group the answers based on statistical categories derived a priori from database statistics. Examining the distribution of responses based on these groups may then allow some insight into the process that people use when spelling. EXPERIMENT 1: FAST SPELLING The objective of this experiment was to examine the type of subsyllabic relationships that people use when spelling nonwords. One method of examining these relationships is to choose nonwords that have alternative spellings that can be manipulated on a dimension of interest, as discussed above. Using such a technique, seven groups were chosen to examine the subsyllabic units that people use in nonword spelling. They can be broken into two main subsets; grain size and morphological subsets. The purpose of the two subsets is the same: that is, to examine the size of the subsyllabic units that people use to spell nonwords (i.e., which units are smaller than the syllable people use when spelling nonwords). However, because the relationships between the individual groups within each subset is slightly different, we describe and analyze the results from the two subsets separately. In terms of the items of the grain size subset, four different groups were used. The first three groups all used a high–low phoneme–grapheme contingency (PGC) manipulation. That is, the nonwords were chosen such that two expected spelling groups for the nonwords would be given for any given stimuli. The expected spelling groups were chosen from previously derived database statistics, with the high PGC group always containing the most common vowel spelling unless that spelling would have resulted in an orthographic body that did not exist. When that happened, the next most common vowel was selected. Further constraints on these stimuli choices are documented here. Where possible, the phoneme–grapheme relationships were taken from Barry and Seymour (1988) and Hanna, Hanna, Hodges, and Rudorf (1966). Relationships that were not in those databases were simply counted from the monosyllabic CELEX database. For example, the phoneme examined in the nonword /t1E*p/ was /E*/. Therefore, the two most common spellings expected were chope (most common) and choap (second most common). The idea is that if people are sensitive to phoneme–grapheme frequency, then it would be expected that more spellings would be given where higher frequency relationships are used than where lower frequency relationships are used. That is, chope would be given more often than choap. Note that the PGC counts we used were sensitive to whether a vowel phoneme occurred at the end of a word. This was done by considering the same vowel different when it occurred at the end of a word and when it occurred in Applied Psycholinguistics 23:1 48 Perry et al.: Spelling and orthographic awareness Table 1. Summary of design