SLA Topics SLA Bibliography Vivian Cook
Orthographic Processes in L2 Users
(unpublished, ca 1997)
A common way of describing language is to divide it into general rules and individual instances, most obviously in the distinctions between grammar and the lexicon and between procedural and declarative memory. In orthography, this division is instantiated as the ‘dual process’ or ‘dual route’ model, alias the ‘standard’ model (Patterson & Morton, 1985). On one side are rules for finding sounds to correspond to letters, on the other instances of whole words held in a mental dictionary. The purpose of this paper is to argue that there are aspects of written language that cannot be captured within either of these two processes.
Dual process models
In dual process models of orthography, the phonological process, called ‘assembled phonology’ or the ‘sublexical route’, uses rules for letter/sound correspondences: the English reader sees <tack> and finds that <t> is corresponds to /t/, <a> to /¾/, and <ck> to /k/. The visual process, called ‘addressed phonology’ or the ‘lexical route’, involves looking up individual word-forms in a lexical store; the reader sees <colonel> and consults a mental list of whole words to establish that it corresponds to <k‘n‘l>.
knowledge of knowledge of
Fig. 1 The standard dual process model of reading aloud
Figure 1 sets out these relationships as two components, rules and instances, mediating between mental word-forms and physical processes. This figure is identical, apart from the wording of the labels, to figures portraying dual-route theories such as the ‘generic dual-route model’ given in Paap, Noel and Johansen (1992, p.294).
The figure shows the ‘routes’ from written text to speech sounds, that is to say, how written shapes are processed into sounds. It is thus a conversion model for changing one form of representation into another—written words into spoken sounds. Both Coltheart, Curtis, Atkins & Haller (1993) and Paap et al (1992) specifically call it a ‘model for reading aloud’; Kreiner (1992) calls it a model of spelling. The processes go from text to sounds, not in the reverse direction; however, Carney (1994) demonstrates amply that different rules are needed in English for each direction and indeed research such as Bradley & Bryant (1985) shows children handle the two directions differently in the early stages of learning to read.
The crucial assumption is that there are only two intervening components: letter/ sound correspondence rules and visual instances. The rules component has a highly complex structure, even if usually left unspecified; the instances component is a paired list of words and phonological forms with no structure apart from word frequency effects. New words or non-words can only be processed by the rules component since no entry exists for them in the lexicon of instances: a novel word such as <vurt> can be read as /v‘:t/ using this process; even an apparently non-English spelling such as dwadziesäcia can be attempted, as news-readers show daily with foreign names.
To avoid terminological confusion, the two components will be referred to here as the ‘rules’ and ‘instances’ components. The rules relating sounds to written forms will be called letter/sound correspondence rules through the written unit for correspondence may well consist of more than one letter, such as English <tch> in match or of syllables in Korean hangul or Japanese kana. The term ‘sounds’ does not imply that the form of phonological representation is necessarily based on the phoneme.
The balance between these two components may be disputed. To Perfetti, Zhang & Berent (1992, p.227) ‘the use of phonology is a general characteristic of reading that exists across writing systems’. For Kreiner (1992), however, the rules component is only used when the instances component fails. A writing system such as Chinese has minimal involvement with sound, even if phonology is more important than has been previously assumed (DeFrancis, 1989); if all the 5,000 odd Chinese characters known by a literate person are one-off instances, the rules component is virtually redundant. A two-component model has little explanatory power when either of its components is used minimally: a dual-route model minus one route is a single-route model (Paap et al, 1992). The routes through the model may be simultaneous rather than one excluding the another, a ‘horse-race’ as it is called by Paap et al (1992); most Japanese reading for instance employs several scripts simultaneously—kanji, katakana, hirakana and romaji—the balance varying from one text genre to another (Smith & Schmidt, 1996). Nevertheless the existence of two, and only two, components is not questioned by dual model theorists.
An apparent alternative to the either/or dual process model is the concept of orthographic depth (Katz & Frost, 1992): ‘shallow’ languages rely more on the sounds component, such as Serbo-Croatian which has consistent speech-to-sound correspondences; ‘deep’ languages rely more on the instances component, such as Chinese. Processing a shallow orthography involves primarily ‘assembling phonology from a word’s component letters’ (Katz & Frost, 1992, p.71); processing a deep orthography means accessing a lexical ‘visual-orthographic’ store. The division is not so much a choice between two components as a continuum between phonological and visual processing. However, orthographic depth research is still concerned with the same two aspects as the dual model—letter/sound rules and visual instances. Rules exist only for conversion of letters to sounds: the instances component has no structure of its own. The only real alternative to the dual process model is the connectionist model put forward by Seidenberg & McClelland (1989) and Plaut, McClelland, Seidenberg & Pattison (to appear), which rejects the dual components as such in favour of a single system but nevertheless tries to account for the same two types of information.
The dual process model mostly acts as a universal model of variation, with languages being seen as having more or less of the same two components. But the dual process model is applied to the individual user as well as to the language as a whole. Surface dyslexia involves difficulty with the instances component, phonological dyslexia with the rules component (Castles & Coltheart, 1993); aphasia sometimes affects only one component (Funnell, 1983); deaf students find it more difficult to read tongue-twisters, showing that they too have a rules component (Hanson, Goodell & Perfetti, 1991).
The model is also relevant to an individual’s performance in a single language. Normal use of a language like Japanese or Korean involves the use of two scripts and two components, one using visual instances for the kanji characters, one sound correspondences for the kana syllabic symbols. Research into Japanese has, for example, compared the speed with which the character-based kanji and the sound-based kana can be recognised (Yamada, 1992). The same may apply within other languages that are less obviously mixed. In many instances the individual chooses whether to use a letter/sound rule or to look up the list of instances. I thought for many years that the English word dugout (as in dugout canoe) was an exotic word pronounced /du:gu:t/, not as having the regular correspondences /dŠgaœt/. Seidenberg (1992a) assigns high frequency English words to a store of visual instances.
Are both these components always available to individuals or is one cut off by the user’s earlier experience? In other words can all individuals use both routes simultaneously or are some able to use only one? The test case here is people who know more than one writing system. Is it possible to acquire knowledge equivalent to a native speaker’s in the components of a writing system if your first language system differs appreciably, say in orthographic depth? In experiments with Keki, a nonsense language, after 20 hours of training English participants reacted to the visual presentation of Keki words as quickly as to English words (Yang & Givón, 1993) and after 50 hours the ERP (event related potentials) in the participants’ brains were no different in Keki and English (McCandliss, Posner & Givón, to appear). It seems that a second language orthography can be acquired with efficiency in a similar fashion to that of a first language. We shall then be looking not only at L1 users of English but also at L2 users of English.
There have always been hints that these two components alone are insufficient to account for the knowledge of orthography. Several aspects peculiar to the writing system are ignored if it is treated only as a reflection of spoken sounds, collectively called by Nunberg (1990) ‘text-category indicators of written language’. Capitalization and word spaces for example are crucial to current Roman alphabet-based scripts. English spelling also has regular patterns such as the so-called ‘three letter rule’ that ‘grammatical’ words such as pronouns may have less than three letters but content words must have three or more letters: in/inn, to/two, I/eye, or/ore (Carney, 1994). In modern English certain morphemes preserve their spelling despite contextually predictable variations in their pronunciation, so that past tense <ed> corresponds to /d/, /t/ and /’d/ in earned, learned and insisted (Chomsky, 1972). Such features, while vital to word recognition, are neither letter/sound correspondence rules nor visual instances.
The main purpose of this paper is to establish that both L1 and L2 users have a systematic knowledge of orthographic regularities that is not encompassed within an division into letter/sound correspondences and visual instances. The aspect chosen to study here is the distribution of letters in the spelling of English words, called ‘orthographic regularity’ by Haynes and Carr (1990). Some spellings in English exhibit regularities concerning the permissible positions of letters. For example <k> occurs as a single consonant at the beginning and end of words (kind, book), <ck> occurs only at the end (back), though both correspond to /k/. Treiman (1993) tested whether children and adults are sensitive to such patterns as double <ff> occurring finally but not initially (beff/ffeb). Her results showed that correct scores rose from 56.4% for ‘kindergarteners’ through 62.3% for first graders and 83.2% for second graders to 94.5% for adults. She attributed the increasing success to experience of reading rather than to age per se. Not only do adults have a good knowledge of orthographic regularities but also children are acquiring them from the first days of learning to read. Such knowledge is readily acquired and known with a high degree of efficiency.
These orthographic regularities are unlikely to be explicitly taught in the way that “i before e except after c” is instilled into English speakers. Instead they are picked up incidentally in the process of learning to read, perhaps even more so than letter/sound correspondences, which are often taught explicitly. Nor could these regularities be deduced purely from letter/ sound correspondences: initial /f/ (fin) does not differ significantly from final /f/ (off) in pronunciation despite the different spelling; the variations in Voice Onset Time or position of /k/ are not linked to the use of <ck> and <k>. These visual patterns of letter combinations are not related to sounds, and are on a par with the purely orthographic differences between hog/Hogg and in/inn.
The experiment reported here
aimed to demonstrate that L1 and L2 users show a knowledge not only of the sound
and visual components but also of orthographic regularities. The research
questions to be addressed are:
—do both L1 and L2 users of English show a knowledge of visual instances, letter/ sound correspondences, and orthographic regularities?
—do L2 users score less than L1 users on all three types of orthographic knowledge?
The methodology is to display pairs of words on a computer monitor; the participants have to choose between the word on the left or the word on the right. The two measures are accuracy and speed of response. The overall aim is to establish that the participants have processes available both for individual words and for letter/sound correspondences through two standard experimental tasks and then to add a third, novel, task to demonstrate their knowledge of orthographic regularities.
The aim was to take a standard paradigm testing the existence of the two components and to amplify it with a test for orthographic regularities. The intention was to find a test at which the subjects would succeed so that their speed of response would relate to things that they knew well rather than those they were doubtful about. The work of Olson, Kleigl, Davidson and Foltz (1985) provided a useful starting point. The first two tasks on ‘instances’ and ‘sounds’ used here were based directly on Olson et al (1985), who used them to show differences both between normal and disabled American children and between the tasks. The first two tasks were essentially identical to those in Olson et al (1985) apart from being slightly shorter. The 48 word pairs in each test were cut to 46 to eliminate certain items that did not work in pilot studies, to be discussed below. Each task had 40 pairs of test words, preceded by six examples. A complete list of the words is given in the appendix.
1. Instances test
The ‘instances’ test corresponds to the ‘orthographic condition’ in Olson et al (1985), apart from the elimination of answer/anser and toward/toard. It tests whether the subjects know the correct spelling for a range of words. One word in each pair has the correct spelling, the other a plausible mis-spelling, such as room/rume. The question displayed on the screen is “Which word is spelled correctly?” It is a test primarily of ‘instances’, i.e. knowledge of individual words, though it is necessarily impossible to exclude some elements of phonological processing (Manis, Seidenberg, Doi, McBride-Chang & Petersen, to appear).
2. Sounds test
The test of letter/sound correspondences is based on the ‘phonological condition’ in Olson et al (1985). It tests whether the subjects can identify words that are disguised under alternative spellings, thus testing letter/sound correspondence rules. Both words in each pair are spelled wrongly but one word would correspond to the sounds of a real English word if said aloud, the other would not. An example is fense/felce where fense corresponds to /fens/, a ‘real’ word, but felce corresponds to the non-existent /fels/. Olson et al (1985) chose words that were common in the vocabulary of ‘the average second grader’. Some did not seem to work for an adult British group. Changes were omission of thair/theer and neer/nerr and alteration of ait/afe to aip/afe, plaice/plice to plais/plice, fead/feam to feer/feem and craul/crail to crorl/crail. The question displayed on the screen is “Which word sounds like an English word?” The participants were told that they could, if they wished, say the word aloud.
3. Orthographic regularities
The ‘orthographic regularities’ task was invented for this experiment, inspired by the Orthographic Constraints task in Treiman (1993), in which subjects had to circle which one of a pair of words ‘looks like it could be a real word’. Treiman’s test used 16 pairs of non-words, one member of each pair conforming to English orthography, the other not. Seven of her 16 pairs concerned permissible consonant or vowel doubling: beff/ffeb; seven covered the distribution of <w> and <u>, and <y> and <i>: bei/bey, chym/chim, gri/gry, dau/daw; two concerned <ck>: ckun/nuck.
Two of the three areas tested by Treiman were not used here. Letter-doubling is a highly complex area of its own, for which Carney (1994) gives eleven basic rules. Furthermore there are some differences between British and American usage—British appal versus American appall, British traveller versus American traveler. The contrasts <w>/<u> and <y>/<i> on the other hand lead too closely to the phonological analysis of approximants and to the disputes between English and American phoneticians over the transcription of boy as /˜’/ or /˜y/ and how as /aœ/ or /aw/.
Ten main regularities were tested, in which the correct spelling could be based neither on letter/sound correspondences nor on knowledge of individual words but only on regularities of orthographic visual structure.
initial <k> vs final <ck>: with some exceptions, English words have <k> initially and <ck> finally; frack/frak, both corresponding to /k/
initial <wh> vs final <w>: <wh> occurs at the beginning of words, corresponding to /w/ (when) or /h/ (who), but never at the end; whon/nowh
initial <wr> vs final <wr>: <wr> only occurs initially: wreg/gewr
initial <j> vs final <dge>: <j> is initial, <dge> is final, both corresponding to /d½/; flidge/flij
initial <ch> vs final <tch>: <ch> occurs initially, <tch> finally, largely corresponding to /t§/; chig/tchig
final <v> vs <ve>: final <v> hardly occurs in English, its equivalent being <ve> truve/truv
final <z> vs <ze>: a single <z> also does not occur finally, its equivalent being <ze>; huz/huze
<qu> vs <q>: <q> must be followed by <u>; quong/qong
final <gn>: <gn> occurs finally or medially in English, but not initially; teign/ gneit
initial <rh>: <rh> occurs initially in English words rather than finally; blar/blarh
Again it is possible to find exceptions to each of these, mostly from Albrow’s exotic third system, such as gneiss, shiv, catarrh, Iraq and celeriac (Albrow, 1972).
Each participant was tested on four pairs of each spelling type, varying in letter position; the other six were used as practice items. The intention was to test the overall knowledge of orthographic regularities rather than the specific knowledge of individual regularities. The question displayed on the screen was “Which word looks like an English word?” It was explained to the participants that neither word actually existed but that one was a possible English word, the other impossible.
The participants were students and staff at the University of Essex in Colchester, England, numbering 38 in all. 14 native were native speakers of British English. 24 were L2 users with at least TOEFL 550 or IELTS 6; they consisted of 2 Arabic, 5 Chinese, 1 Danish, 2 Dutch, 1 French, 3 German, 2 Greek, 3 Japanese, 1 Persian, 4.
The experiment used the ERTS PC program (Beringer, 1996) to control presentation, timing, etc and to record answers and response times. Word pairs were presented in the same randomised order within each test; half had the correct answer on the right, half on the left. The response keys were Left Alt, Right Alt, labelled L and R. The pairs were shown at mid-screen in a 26 point type Times New Roman font in lower case. The same sequence of tests was kept for all participants, namely instances, sounds, orthographic regularities. Participants were told that it was a test of spelling. No feedback was provided. During the 6-item practice section of each test, the experimenter ensured that they understood the response they had to make; they were told that both accuracy and speed would be measured. The whole test took about 15 minutes.
The predictions were then:
1) L1 and L2 users of English will score significantly better than chance in terms of accuracy at tests of instances, sounds and orthographic regularities
2) L2 users will be significantly worse than L1 users at all tasks in terms of accuracy and response time.
The results are presented in table 1. The first prediction was that both L1 and L2 users would score significantly better than chance at all three tests, i.e. better than 20 out of 40. Both L1 users and L2 users indeed had a high rate of accuracy on all three tests (one-sample t test, p<0.001 for each test for each group, L1 group df. 13, L2 group df.23), supporting this prediction.
Response Times Accuracy
(mills) s.d. (/40) s.d.
Instances test L1 users 800 139 39.7 0.61
L2 users 1043 329 38.9 6.58
Sounds test L1 users 1658 400 35.5 3.08
L2 users 2234 515 31.3 6.11
OR test L1 users 1504 351 35.4 3.08
L2 users 1827 561 33.3 3.30
Table 1 Results
The second prediction was that L2 users would be significantly worse than L1 users at all tasks, both in accuracy and in speed of response. Starting with accuracy, the L2 users were worse than L1 users when all three tests are taken together (ANOVA between groups, df.1, F=4.9, main effect, p<0.033), as can be seen in Figure 2.
The L1 users showed a ceiling effect on the instances test with 39.7 words out of 40 correct; the L2 users had a greater spread of scores on this test. There were significant differences between the three tests overall (ANOVA between measures, main effect, df. 35, F=56.093, p<.001), with the instances test being the easiest (pairwise comparisons, p<.001) but there were no significant differences between the orthographic regularities test and the sounds test (pairwise comparisons, p=0.197). The deficit for non-native speakers varied between 0.8 (instances) to 4.2 (sounds) but was not significantly different between the three tests (interaction, df.35, F= 1.432, p=0.252).
In terms of response time (Figure 3), the three tests produced significant differences (between measures effect, df.2, F=85.5, p<0.001); for both groups the instances test was fastest, the sounds test slowest. All 3 tests differ from each other (pairwise comparisons, p<0.002). The difference between the L2 users and the L1 users varies from 243 milliseconds slower (instances) to 576 milliseconds slower (sounds) but does not differ significantly from test to test (interaction, df.2, F=2.474, p=0.091). The L1 users were faster than the L2 users overall (between groups, main effect, df=1, F=12.41, p<0.001). The instances test responses for L1 users again show a ceiling/floor effect with a smaller spread of scores than the others. Quite clearly the second prediction is met: L2 users are less accurate and slower than L1 users.
Both L1 and L2 users do therefore have a knowledge of instances, sounds and orthographic regularities. This establishes that language users know, not only the two main components of sounds and instances that have been studied in the literature, but also orthographic regularities, in both L1 and L2. It is not necessary to be a monolingual speaker of English to know all three types of information. The dual process model cannot readily accommodate these results since it provides no obvious way in which these regularities can follow either the sounds route or the visual route. The argument applies equally to L1 and L2 users: they both know orthographic regularities that are neither visual instances nor sound rules to an equivalent level of success and with very similar response times.
L2 users do score slightly less than L1 users, despite their high level of accuracy, ranging from 0.8 to 4.2 out of 40 in favour of the L1 user; they have response times that are 21% to 34% slower. The L2 user is slightly less effective than a monolingual on many other tasks (Cook, 1997). There is then an overall deficit for L2 users across tasks and across measures. This experiment has shown that advanced L2 users have efficient access to all three types of knowledge, though with slightly less speed than native speakers. Because of the mixed languages of the participants, this slight deficit cannot be ascribed to interference from any particular L1 writing system.
The main thrust of this paper is to argue that, in addition to the knowledge of letter/sound correspondence rules and visual instances, users also know regularities of orthographic form, as the results of both experiments clearly show. As Seidenberg (1992b, p.248) said of the two-route model, though for different reasons, ‘This dichotomy is not rich enough to capture facts about human performance’. It is doubtless a moot question whether these regularities play more than a minor role in real-life reading. What the argument reveals, however, is a gap in the standard two-component model given in Figure 1. Seidenberg (1992a) talks of an ‘equitable division of labour’ between the two sides; the results here suggest some labour is distributed elsewhere.
Can orthographic regularities fit into the dual process model? Such knowledge cannot be readily accommodated within the ‘rules’ of sound/letter correspondence, since these regularities are visual rather than phonological. Nor does it fit into a list of ‘instances’, since it does not consist of individual idiosyncratic words but of systematic knowledge. Accounting for this information requires something else: people know not only rules for relating letters to sounds and a list of idiosyncratic words but also how letters pattern with each other.
Seidenberg (1992b, p.109) insists ‘both routes are capable of handling words but only the orthographic to phonological computation is relevant to non-words’. If only conversion from one medium to another is involved, this may be correct but, so far as the visual route is concerned, the regularities test here has demonstrated that people can handle non-words orthographically just as well as they handle them phonologically, whether L1 users or L2 users from a variety of backgrounds. Non-words can have orthographic regularities unrelated to letter/sound correspondences.
One way of incorporating orthographic regularities might be to adopt a connectionist model such as Plaut et al (to appear). If all differences amount to the weighting of connections within a single system, orthographic regularities can presumably be covered within the same network as the other types of information. The Plaut et al (t.a.) model includes two routes, one going directly from orthography to semantics, one from orthography to phonology to semantics. Yet, while in principle orthographic regularity seems to fit this formulation, two additions are needed, neither of which seem to be explicitly stated as yet. One is the elaboration of the semantic route; the link between orthography and meaning needs to have a structure of its own, as yet undescribed in Seidenberg (1992a) or Plaut et al (t.a.). The second is the recognition that the facts of orthographic regularities need to be accommodated; so far the connectionist model has tried to map the information found in the sounds and instances components but not orthographic regularities.
The other main point is that these three aspects of orthographic competence are readily available to all users of a writing system. It does not make much difference to using English whether one is an L1 user, a Japanese, or a German: all three components are familiar to the users even if they have never been taught them directly and are not consciously aware that they know them. The human mind apparently has no problems in acquiring these three types of knowledge equally even when one of them has been comparatively unimportant in the acquisition of their first writing system. All three therefore have to be accommodated in one way or another in models of spelling or reading, of the individual as well as of the writing system as a whole, and of L2 users as well as L1 users.
Albrow, K.H. (1972). The English Writing System: notes towards a description. London; Longman
Cook, V.J. (1997). ‘The consequences of bilingualism for cognitive processing’. In A. de Groot & J.F. Kroll (eds.), Tutorials in Bilingualism: Psycholinguistic Perspectives, Lawrence Erlbaum
Beringer, J. (1996). Experimental Run Time System. Frankfurt; Berisoft, Version 3.18
Bradley, L. & Bryant, P. (1985). Rhyme and Reason in Reading and Spelling. Oxford; Blackwell
Carney, E. (1994). A Survey of English Spelling. London; Routledge
Castles, A. & Coltheart, M. (1993). ‘Varieties of developmental dyslexia’. Cognition, 47, 149-180
Chomsky, N. (1972). ‘Phonology and reading’ In H. Levin (ed.), Basic Processes in Reading. Harper and Row
COBUILD (1995). Collins COBUILD English Dictionary. London, HarperCollins, new edition
Coltheart, M., Curtis, B., Atkins, P. & Haller, M. (1993). ‘Models of reading aloud: dual route and parallel-distributed-processing approaches’. Psychological Review, 100, 4, 589-608
DeFrancis, J. (1989). Visible Speech. University of Hawaii Press
Funnell, E. (1983). ‘Phonological processes in reading: new evidence from acquired dyslexia’. British Journal of Psychology, 74, 159-180
Hanson, V.L., Goodell, E.W. & Perfetti, C.A. (1991). ‘Tongue twister effects in the silent reading of hearing and deaf college students’. Journal of Memory and Language, 30, 319-330
Haynes, M. & Carr, T.H. (1990). ‘Writing system background and second language reading: a component skills analysis of English reading by native-speaking readers of Chinese’. In T.H. Carr & B.A. Levy (eds), Reading and its Development: Component Skills Approaches. San Diego; Academic Press, 375-421
Katz, L. & Frost, R. (1992). ‘Reading in different orthographies: the orthographic depth hypothesis’. In R. Frost & L. Katz (eds.), Orthography, Phonology, Morphology and Meaning. Amsterdam; Elsevier, 67-84
Kreiner, D.S. (1992). ‘Reaction time measures of spelling: testing a two-strategy model of skilled spelling’. Journal of Experimental Psychology: Memory and Cognition, 18, 4, 765-776
Manis, F.R., Seidenberg, M.S., Doi, L.M., McBride-Chang, C. & Petersen, A. (to appear). ‘On the bases of two subtypes of developmental dyslexia’
McCandliss, B.D., Posner, M.I. & Givón, T. (to appear). ‘Brain plasticity in learning visual words’
Nunberg, G. (1990). Linguistics of Punctuation. Stanford; CSLI
Olson, R.K., Kleigl, R., Davidson, B.J. & Foltz, G. (1985). ‘Individual and developmental differences in reading ability’. In G.E.Mackinnon & T.G.Waller (eds), Reading Research: Advances in Theory and Practice Vol. 4 New York; Academic Press, 1-64
Paap, K.R., Noel, R.W. & Johansen, L.S. (1992). ‘Dual-route models of print to sound: red herrings and real horses’. In R. Frost & L. Katz (eds.), Orthography, Phonology, Morphology, and Meaning. Amsterdam; Elsevier, 293-318
Patterson, K.E. & Morton, J. (1985). ‘From orthography to phonology: an attempt at an old interpretation’. In K.E. Patterson, J.C. Marshall & M. Coltheart (eds.), Surface Dyslexia. London; Erlbaum, 335-359
Perfetti, C.A., Zhang, S. & Berent, I. (1992). ‘Reading in English and Chinese: evidence for a ‘universal’ phonological principle’. In R. Frost & L. Katz (eds.), Orthography, Phonology and Meaning. Amsterdam; North Holland
Plaut, McClelland, Seidenberg & Pattison (in press). ‘Visual word recognition: are two routes really necessary?’. Psychological Review
Seidenberg, M.S. (1992a). ‘Beyond orthographic depth in reading: equitable division of labour’. In R. Frost & L. Katz (eds.), Orthography, Phonology, Morphology and Meaning. Amsterdam; Elsevier, 85-118
Seidenberg, M.S. (1992b). ‘Dyslexia in a computational model of word recognition’. In P. Gough, L. Ehri, & R. Treiman (eds.), Reading Acquisition. Erlbaum, 243-274
Seidenberg, M.S. & McClelland, J.L. (1989). ‘A distributed, developmental model of word recognition and naming’. Psychological Review, 96, 523-568
Smith, J.S. & Schmidt, D.L. (1996). ‘Variability in written Japanese: towards a sociolinguistics of script choice’. Visible Language, 30, 46-71
Treiman, R. (1993). Beginning to Spell. Oxford University Press
Yamada, J. (1992), ‘Why are kana words named faster than Kanji words?’. Brain & Language, 43, 682-693
Yang, L. & Givón, T. (1993). ‘Tracking the acquisition of L2 vocabulary: the Keki language experiment’. Tech report 93-11, University of Oregon, Institute of Cognitive and Decision Sciences
Appendix. Word pairs
Instances test (adapted from Olson et al, 1985)
Sounds test (adapted from Olson et al, 1985)