phonological encoding in the lexical decision task

This article was downloaded by:[Landsbokasafn Islands - Haskolabokasafn]On: 18 June 2008Access Details: [subscription number 788868639]Publisher: Psychology PressInforma Ltd Registered in England and Wales Registered Number: 1072954Registered office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK

The Quarterly Journal ofExperimental PsychologyPublication details, including instructions for authors and subscription information:http://www.informaworld.com/smpp/title~content=t716100704

Phonological encoding in the lexical decision taskMax Coltheart a; Derek Besner b; Jon Torfi Jonasson b; Eileen Davelaar ba Birkbeck College, University of London,b University of Reading,

Online Publication Date: 01 August 1979

To cite this Article: Coltheart, Max, Besner, Derek, Jonasson, Jon Torfi andDavelaar, Eileen (1979) 'Phonological encoding in the lexical decision task', TheQuarterly Journal of Experimental Psychology, 31:3, 489 — 507

To link to this article: DOI: 10.1080/14640747908400741URL: http://dx.doi.org/10.1080/14640747908400741

PLEASE SCROLL DOWN FOR ARTICLE

Full terms and conditions of use: http://www.informaworld.com/terms-and-conditions-of-access.pdf

This article maybe used for research, teaching and private study purposes. Any substantial or systematic reproduction,re-distribution, re-selling, loan or sub-licensing, systematic supply or distribution in any form to anyone is expresslyforbidden.

The publisher does not give any warranty express or implied or make any representation that the contents will becomplete or accurate or up to date. The accuracy of any instructions, formulae and drug doses should beindependently verified with primary sources. The publisher shall not be liable for any loss, actions, claims, proceedings,demand or costs or damages whatsoever or howsoever caused arising directly or indirectly in connection with orarising out of the use of this material.

http://www.informaworld.com/smpp/title~content=t716100704

http://dx.doi.org/10.1080/14640747908400741

http://www.informaworld.com/terms-and-conditions-of-access.pdf

Dow

nloa

ded

By:

[Lan

dsbo

kasa

fn Is

land

s - H

asko

labo

kasa

fn] A

t: 00

:35

18 J

une

2008

Quarterly Journal of Experimental Psychology (1979) 31, 489-507

PHONOLOGICAL ENCODING I N T H E LEXICAL DECISION TASK*

MAX COLTHEART Birkbeck College,

University of London

DEREK BESNER, JON TORFI JONASSON AND EILEEN DAVELAAR University of Reading

In lexical decision experiments, subjects have difficulty in responding NO to non-words which are pronounced exactly like English words (e.g. BRANE). This does not necessarily imply that access to a lexical entry ever occurs via a phonological recoding of a visually-presented word. The phonological recoding procedure might be so slow that when the letter string presented is a word, access to its lexical entry via a visual representation is always achieved before phonological recoding is completed. If prelexical phonological recodings are produced by using grapheme-phoneme correspondence rules, such recodings can only occur for words which conform to these rules (regular words), since applications of the rules to words which do not conform to the rules (exception words) produce incorrect phonological representations. In two experiments, it was found that time to achieve lexical access (as measured by YES latency in a lexical decision task) was equivalent for regular words and exception words. I t was concluded that access to lexical entries in lexical decision experiments of this sort does not proceed by sometimes or always phonologically recoding visually-presented words.

Introduction

Access to a store of word-representation, to an internal lexicon, is a major component of the act of reading. Many of the information-processing tasks which have been used to investigate reading do not require a subject to access his or her internal lexicon, since the tasks can be executed successfully without such lexical access. Examples of such tasks are tachistoscopic report, the Reicher- Wheeler forced choice task, visual search, same-different judgement, the Sternberg memory- search task, and reading aloud. Each of these tasks can be performed with letter- strings which are not words and therefore have no lexical entries. Thus when letter-strings which are words are used, it is possible that subjects sometimes or always neglect the existence of lexical entries for these stimuli, and perform the task as if the words were non-words. Whether lexical access occurs during the execution of these tasks is not a requirement of the task, but an empirical question

*Requests for reprints to Max Coltheart, Department of Psychology, Birkbeck College, Malet Street, London WCIE 7HX, England.

0033-5~5~/79/030489 + I9 $OZ.OO/O 0 1979 The Experimental Psychology Society

Dow

nloa

ded

By:

[Lan

dsbo

kasa

fn Is

land

s - H

asko

labo

kasa

fn] A

t: 00

:35

18 J

une

2008

M. COLTHEART ET AL. 490

(and of course the existence of such phenomena as word-superiority effects may indicate that words sometimes do receive some form of special treatment).

Considerations of this sort suggest that, if we are specifically interested in the process of lexical access, we ought to study this process by using tasks which compel the subject to access his internal lexicon, rather than leaving this optional. One such task is lexical decision-deciding whether a letter-string is an English word or not. Provided that all the non-words used for such a task are pronounceable and conform to English spelling regularities, this decision can only be made by consulting one’s internal lexicon and determining whether the letter-string one is viewing is present in or absent from this lexicon. How else could a reader decide that LEAT and SHIVE are non-words whilst LEAP and SHINE are words?

This task was used by Rubenstein, Lewis and Rubenstein (1971) to investigate the possible role of phonological recoding in lexical access during reading. Two effects were observed which suggested that subjects were converting visually- presented letter-strings into phonological representations prior to lexical access. Firstly, non-words which were pronounced identically to English words (BURD, BLUD, GROE-we refer to such non-words as pseudohomophones) yielded slower NO latencies than non-words which were not pronounced identically to any English words (ROLT, HOSK, WESP). Secondly, post hoc analysis of YES latencies suggested that homophonic words produced slower YES latencies than non-homophonic words provided that the homophones were the less frequent member of the homophone pair; thus WEAK, MAID or THREW, which are less frequent than WEEK, MADE and THROUGH respectively, produce slow YES latencies, whilst PRAY, RAIN or SAIL do not, since although they are homophones, they have a higher frequency of occurrence than PREY, REIGN and SALE respectively.

These effects were further investigated by Coltheart, Davelaar, Jonasson and Besner (1977), in view of possible difficulties in interpreting the results of Ruben- stein et al. Firstly, the NO effect observed by Rubenstein et al. could have occurred because of differences between their two classes of non-words in visual similarity to English words. Such effects of visual similarity do occur: for example, Experiment 2 of Coltheart et al. (1977) showed that NO latency in a lexical decision task was slower for non-words that can be made into many English words by single letter changes than for non-words where single letter changes produce few English words. Rubenstein et al. (1971) did not attempt to match their two classes of non-words with respect to similarity to English words, and inspection of their non-words in fact gives the strong impression that their homophonic non-words were of a higher degree of visual similarity to English words than were their non-homophonic non-words. Coltheart et al. (1977) attempted to overcome this difficulty by generating from each of their pseudohomophones a matching non-homophonic non-word which differed by only one letter from the pseudohomophone: for example, WAID/DAID, FLOO/FROO, and so on. With such matched sets of pseudohomophones and non-pseudohomophones, the Rubenstein effect was still obtained; every subject was slower to respond NO to non-words which were homophonic with English words than to non-words which were not. It seems correct, then, to conclude that this effect is genuine evidence

Dow

nloa

ded

By:

[Lan

dsbo

kasa

fn Is

land

s - H

asko

labo

kasa

fn] A

t: 00

:35

18 J

une

2008

PHONOLOGICAL ENCODING AND READING 491

for phonological recoding. Such a conclusion gains further support from the work of Patterson and Marcel (1977). They studied two patients who had suffered injuries to the left hemisphere which produce the syndrome of “deep dyslexia” (Marshall and Newcombe, 1973), also known as “phonemic dyslexia” ( Shallice and Warrington, 1975). One symptom of this syndrome is an inability to carry out phonological recoding of visually-presented letter-strings without use of the lexicon. Evidence for such an inability includes the finding that such patients, when asked to read aloud simple pronounceable non-words such as RUD or GLEM, cannot do so, although the patients are reasonably successful in saying such items when presentation is auditory. If the ability to convert visually- presented letter strings to phonological representations prior to lexical access is in fact lost in these patients, one can investigate whether the slow NO latencies produced to pseudohomophones in lexical decision experiments is a genuine phonological effect, or whether it is due to visual differences between the two classes of non-words, by repeating the Rubenstein et al. (1971) experiment with these patients, The two patients were able to perform the lexical decision task with a high degree of accuracy, making few false positives; and their NO latencies were no greater for pseudohomophones than for non-homophonic words, whereas a group of control subjects judging the same set of words and non-words were significantly slower with pseudohomophonic non-words. Thus the absence of an ability to carry out phonological recoding of visually-presented letter-strings does abolish the NO effect obtained by Rubenstein et al. (1971) and Coltheart et al. (1977).

In connection with the YES effect obtained by Rubenstein et al. (1971), it may be noted that their post hoc analysis of this effect did not control for such variables as word frequency or part of speech, with influence YES latency in lexical decision tasks (e.g. Frederiksen and Kroll, 1976; Scarborough and Springer, 1973). For this reason Coltheart et al. (1977) carried out an experiment specifically designed to investigate this YES effect by comparing YES times to 39 homophonic words (each being the rarer member of a homophonic pair) with YES times to 39 non- homophonic words matched with the homophones on word frequency, number of letters, number of syllables, and part of speech including inflections. There was no difference in YES latency to the two classes of word. This disconfirms the model of lexical access originally proposed by Rubenstein et al., since the model requires that rarer homophones will produce slower YES responses than matched non-homophonic words. Nevertheless, clear evidence for phonological encoding prior to lexical access in the lexical decision task exists, in view of the robustness of the original NO effect and its absence in deep dyslexia.

However, the relevance of this finding to lexical access during normal reading is unclear. It is possible, for example, that, although phonological recoding is occurring during normal reading, and in lexical decision experiments, it is such a slow process that lexical access using a visual word-representation always finishes before phonological recoding is completed. In the lexical decision task, on those occasions when a non-word is presented, the unsuccessful lexical search using a visual representation may take long enough to allow phonological recoding to be completed, and hence to allow effects of phonological recoding to become evident.

This is what Patterson and Marcel (1977) did.

Dow

nloa

ded

By:

[Lan

dsbo

kasa

fn Is

land

s - H

asko

labo

kasa

fn] A

t: 00

:35

18 J

une

2008

492 M. COLTHEART ET AL.

This argument demonstrates that, if our interest is in the possible use of phonological recoding for getting to a printed word's lexical entry, effects on NO responses in the lexical decision task are uninformative; what we want are effects on the YES response. Unless such effects are demonstrated, it will always be possible to argue that phonological recoding has a merely epiphenomena1 role in lexical access.

This problem can be approached by considering what procedure might be employed to obtain a phonological recoding of a visually-presented word without using lexical knowledge. If access to a word's lexical entry is sometimes achieved by first converting the word to a phonological representation and then using this representation for lexical access, there must exist some non-lexical procedure for converting print to phonology.

Three possibilities suggest themselves. The most frequently mentioned possibility is that readers possess and can use a system of grapheme-phoneme correspondences (GPCs), and that printed words are analyzed into their graphemic constituents, to each of which the appropriate phoneme is assigned, thus converting a string of letters into a string of phonemes. A second possibility is that the unit which is used is the syllable, not the phoneme; printed words are analyzed into letter-groups corresponding to syllables, and an internal syllabary exists by means of which a syllable can be assigned to each syllabic letter-group. The third and final possibility is that no analysis of whole words into subword units is carried out; the pronunciation of a word as a whole is obtained from a dictionary of word- pronunciations (a phonological lexicon). Given that we are interested in whether access to a printed word's meaning ever occurs via a prior recoding of the word into phonological form, we must propose that there are at least two separate and independent stores of information about words-a phonological lexicon and a semantic lexicon-if we are to adopt this third possibility. Otherwise, if we regard a word's lexical entry as a single entity containing both semantic and phonological information, we cannot make sense of the idea that a phonological recoding is used to gain access to a word's lexical entry.

These three theoretical approaches to the question of how a reader might go about the task of deriving a phonological representation of a printed letter string have been discussed in detail elsewhere (Coltheart, 1978) and it has been argued there that only the approach based upon GPCs appears to be workable in principle; also, such evidence as there is appears to be consistent only with this approach. We will assume here, therefore, that, to the extent to which prelexical phonologic a1 encoding occurs during reading, it is achieved by the use of GPCs."

The GPC procedure, even though preferable on theoretical and empirical grounds to the other possible procedures, is not without its own difficulties; these too are discussed by Coltheart (1978) and will be mentioned only briefly here.

The relationship of letters to phonemes in English is sometimes many-to-one.

What might this procedure be?

*We assume that readers do not have available more than one of the three possible mechanisms for non-lexical phonological encoding. It is, of course, conceivable that readers can use a GPC procedure and an internal syllabary. We make our assumption through considerations of parsimony, and also because of the absence of any positive evidence for any non-lexical phonological encoding procedure which does not use GPCs.

Dow

nloa

ded

By:

[Lan

dsbo

kasa

fn Is

land

s - H

asko

labo

kasa

fn] A

t: 00

:35

18 J

une

2008


Therefore grapheme-phoneme correspondences could not be achieved simply by assigning a phoneme to each letter in a letter string. There must first be a parsing of the letter string into those letters or letter-groups each of which corresponds to a single phoneme, i.e., a parsing of the string of letters into a string of “functional spelling units” (Venezky, 1970). Examples of such parsings are TH/I/CK, J/UDG/I/NG and B/R/EE/CH. The relationship of functional spelling units to phonemes is one-to-one, and therefore a stage of word-parsing can be followed by a stage of phoneme assignment, in which an internal table of spelling-unit/phoneme correspondences is used to derive from each spelling-unit its appropriate phoneme.

If this procedure is to work without reference to an internal lexicon, the reader will need to be able both to parse letter strings into their functional spelling units, and also to assign phonemes to these spelling units, without using lexical knowledge. This is impossible, because English is irregular both at the parsing level and at the phoneme assignment level. Thus no procedures exist which will correctly parse all English words; and even when a word has been parsed correctly, no procedures exist which will assign phonemes to spelling units correctly for all English words.

These two problems can conveniently be illustrated with reference to vowel digraphs. The digraphs AI, EA, OA, OE and UI usually correspond to single phonemes in English, and hence each of these letter-pairs is normally a single functional spelling unit; but a parsing procedure which treated them thus would fail for such words as DAIS, REACT, BOA, POET and RUIN. In each of these words, the vowel digraph is two functional spelling units, not one. There are no systematic differences (e.g. in consonant environment) between those words in which the vowel digraphs are single units and those in which they are two units; so no parsing procedure can be devised which will work for all of the words of English containing these vowel digraphs.

But let us suppose this was not true, and that all the words for which the vowel digraph is a single unit were successfully parsed in this way. The assignment of phonemes to these single units run into the problems that, for each of these five digraphs, there are at least two phonemes, and for some digraphs as many as four phonemes, which could legitimately be assigned (RAID/AISLE/PLAID/AGAIN, VEAL/BREAD/STEAK, ROAD/BROAD, TOE/SHOE/DOES, CIRCUIT/ BRUISE/BUILD). Once again, there are no systematic differences between these various words which determine which of the possible phoneme assignments is correct ; so no phoneme-assignment procedui e can be devised which will work for all the words of English containing these vowel digraphs.

These difficulties are not confined to vowel digraphs; there are many other kinds of English words for which no consistent parsing procedure and no consistent phoneme-assignment procedure exists.

However, Wijk (1966) and Venezky (1970) have provided a data-base from which one can derive procedures which work for most English words. The digraphs AI, EA, OA, OE and UI are single functional spelling units in most of the words in which they occur; and when they are single units, their pronunciations are almost always, respectively, as in RAID, VEAL, ROAD, TOE and BRUISE. Any procedure which assumed that this parsing and these assignments held for all words would fail on some words, but the percentage of words on which it would

Dow

nloa

ded

By:

[Lan

dsbo

kasa

fn Is

land

s - H

asko

labo

kasa

fn] A

t: 00

:35

18 J

une

2008


fail is small. The work of Wijk and Venezky has shown that it is in general true that one can devise a set of parsing procedures and a set of assignments of phonemes to spelling units which will work for a large percentage of English words (let us call these “regular words”) and will yield incorrect phonological representations for the remainder of English words (let us call these “exceptions”).

It follows that, if one takes the view that the means by which a reader derives a phonological representation of a printed letter-string without using his interna1 lexicon is to apply a set of GPCs to the string, there exists a set of words (exception words) for which phonological representations simply cannot be obtained without reference to the internal lexicon. Thus the demonstrations by Baron and Strawson (1976) (and by Edgmon: see Gough and Cosky, 1977) of a difference in naming latencies between regular and exception words provides strong support for the psychological reality of the GPC procedure. As they point out, the possible use of GPC procedures for naming visually presented words is confined exclusively to regular words, whereas the alternative method for obtaining the pronunciation of a word (namely, looking it up in the internal lexicon) is available equally for regular and for exception words. The finding that naming latency is shorter for regular words (which have two potentially available methods for their pronunciation, the GPC strategy and lexical lookup) than for exception words (which can only use the lexical lookup method) suggests that GPC procedures do exist and that readers can use them (for regular words).

With respect to the syndrome known as “surface dyslexia”, a reading disorder arising from damage to the temporoparietal region of the left hemisphere (Marshall and Newcombe, 1973; Newcombe and Marshall, 1973 ; Holmes, 1973), detailed analyses of the kinds of incorrect responses made by the patients when reading single printed words aloud strongly suggest that they are failing to apply specific GPCs such as “G and C are soft before E and I, otherwise hard” and “In the sequence vowel+consonant+final E, the vowel is long”. Difficulties the patients show in dealing with vowel digraphs suggest furthermore that the patients sometimes fail at the parsing stage. In addition, both Shallice and Warrington (personal communication) and Saffran (personal communication) have observed patients with reading disorders produced by left hemisphere damage for whom reading aloud is more accurate for regular words than for exception words.

Thus findings from both normal and disordered reading support the view that it is by means of the GPC procedure that prelexical phonological recoding of print is achieved, in spite of the fact that for some words (exception words) this procedure yields an incorrect phonological code.

If we assume that this view concerning the nature of phonological recoding of print is correct, then we can reduce the question “Is lexical access during the lexical decision task ever based on a prior phonological recoding of a visually presented letter string?” to the question “Are GPC procedures employed by subjects when they are carrying out the IexicaI decision task?” The use of GPCs is confined to regular words; when applied to exception words, the GPC procedure

Clinical observations support this view.

Dow

nloa

ded

By:

[Lan

dsbo

kasa

fn Is

land

s - H

asko

labo

kasa

fn] A

t: 00

:35

18 J

une

2008


yields incorrect phonological representations." One can discover whether performance of a task involves the use of GPCs by determining whether performance is better with regular words than with exception words. We were thus led to investigate whether YES responses in a lexical decision task are made more rapidly with regular words than with exceptions,

Pilot study In each of the two experiments reported below, two classes of words are used:

words which conform to GPC rules (regular words) and words which do not (exception words). The two types of word were selected by referring to the rules proposed by Wijk (1966) and Venezky (1970). However, in order to ensure the validity of this selection, a pilot study was carried out to determine whether, with the sets of words we used, the regular words would be named more rapidly than exception words, as was found, for a different collection of words, by Baron and Strawson (1976). Only if this difference is demonstrable can one be sufficiently confident that one has succeeded in selecting one collection of words which conform to GPC rules and another which does not.

These consist of 39 exception words and 39 matched regular words. These 78 words were divided up into six lists. List I was the first 13 exception words, List 2 the corresponding first 13 regular words, List 3 the second 13 exception words, and so on down to List 6 which was the last 13 regular words.

Here there were 20 exception words. List 8 consisted of the first of each of the set of regular control words of Experiment 11; List 9 consisted of the second item from each of these sets.

Thus nine lists were used in the pilot study: four of these contained exclusively exception words, and five contained exclusively regular words. These lists were typed on nine separate sheets of paper. All nine lists were given, in random order, to each of 23 undergraduates at the University of Reading. The subject was told that he would be asked to read aloud the list of words as fast as he could. For each list, the time between exposing the list to the subject and his completion of the pronunciation of the last word on the list was measured by a stopwatch.

The mean time per word for completion of each of the nine lists is shown in

*Here we reject the possibility of lower-priority GPC rules; e.g., when the rule which specifies that OW is pronounced as in COW fails to produce a word when applied to MOW, we neglect the possibility than an alternative (less probable) rule, that OW is pronounced as in BOWL, is also applied. If these two different rules could be applied simultaneously, then exception words, those to which the lower-priority rules apply, would never suffer relative to regular words, which is inconsistent with the results of Baron and Strawson (1976). If the two different rules were applied in order of priority, then exception words would always be dealt with more slowly than regular words. In this case, to infer from a disadvantage incurred by exception words that GPCs are being used would still be correct. In any case, hierarchical use of GPC rules with differing priorities would seem in principle to be unworkable, since when GPC rules are applied to exception words, the incorrect phonological representation produced is often that of another word (e.g. BREAD becomes BREED, COME becomes COMB, MOVE becomes MAUVE, GAUGE becomes GORGE, SHOE becomes SHOW; there are many such examples). How, then, could the system detect the incorrectness of the phonological representation yielded by application of GPCs to exception words?

The words used in Experiment I are listed in the Appendix.

The words used in Experiment I1 are also listed in the Appendix. This constituted List 7 in the pilot study.

Dow

nloa

ded

By:

[Lan

dsbo

kasa

fn Is

land

s - H

asko

labo

kasa

fn] A

t: 00

:35

18 J

une

2008


Table I. In every case, a regular list was completed more rapidly than its matched exception list. For each subject the times taken to complete Lists I, 3 and 5 (which make up the exception words of Experi- ment I) were summed, as were the times taken to complete Lists 2,4 and 6 (which

These data were analysed as follows.

TABLE I Mean time per word in ms for completion of the nine word lists used in the pilot study

Word list I 2 3 4 5 6 7 8 9

Time (ms) 470 463 517 499 so4 450 48s 477 453

make up the regular words of Experiment I). Thus each subject was given a total time for exception words and a total time for regular words. The total times were significantly greater for exceptions words than for regular words ( t 2 2 = 3 * 6 ~ ,

Then the words used in Experiment I1 were analysed by averaging each subject's times on the regular lists, List 8 and 9, and comparing these regular-word times with the times needed for List 7, which contained exception words. Again, the exception words took significantly longer (t,,= 1.87, P <o.os, one-tailed). Although this difference was significant only with a one-tailed test, the fact that such a test is defensible here would seem to provide some justification for proceed- ing with the use of these words in Experiment 11.

This pilot study demonstrates, then, that on the average subjects do take longer to pronounce the exception words to be used in Experiment I than to pronounce the regular words; and that this is also true of the exception words and regular words to be used in Experiment 11. Grounds are thus provided by this pilot study for the view that our selections of exception words and regular words are valid.

Further evidence for the validity of the sets of words used in Experiment I is provided by Shallice and Warrington (personal communication) ; they used these 78 words in testing the single-word reading ability of a patient in whom, other studies had implied, direct lexical access from print was impaired whilst GPC ability was not abolished. One would expect this patient, then, to be more successful at pronouncing the regular words of Experiment I than the exceptions ; and this was the case.

Experiment I Method

P (0'01).

Subjects

Eo.50 per hour for their participation. (see Apparatus, below).

Stimulus materials

provided by Wijk (1966) and Venezky (1970) yielded incorrect pronunciations.

The subjects were 3 1 undergraduates at the University of Reading, paid at the rate of They were run in groups of up to five at a time

(These are listed in the Appendix). A set of thirty-nine words was chosen for which rules A large

Dow

nloa

ded

By:

[Lan

dsbo

kasa

fn Is

land

s - H

asko

labo

kasa

fn] A

t: 00

:35

18 J

une

2008


variety of forms of irregularity was included (see the column “Exceptions” in the Appendix). For each of these 39 exception words, a matched regular word was selected. The two words in any such pair were closely matched for word frequency (Kucera and Francis, 1967), number of letters, number of syllables, number of morphemes, concreteness/imageability and part of speech (including inflections). Seventy-eight pronounceable non-words were also selected.

Apparatus

strings were displayed on these CROs. subject per CRO. outtons.

Five Advance OS250 CROs slaved to a PDP-12 computer were used. Upper-case letter Subjects were run in groups of up to five, one

In front of each CRO was a response panel containing two response

Procedure Subjects were instructed that their task was to decide as quickly as they could whether

each letter string which appeared on their display was an English word or not. They were told that half of the displays would be words, and that the non-words would be pronounceable and hence would resemble English words. They rested their index fingers on their two response buttons; the right button was used for the YES response, the left for NO.

After the fixation point disappeared, there was a pause of 750 ms. Then a letter string appeared on the visual display. I t remained visible until every subject in the group had responded, and so each trial was terminated by the response of the slowest subject on that trial. There was then a pause of 320 ms, after which the next trial began. The 156 words and non-words were presented in a random sequence, preceded by twenty-four practice trials (12 words, 12 non-words).

Results and discussion

A trial began with the presentation of a fixation point for zoo ms.

Each subject’s data were treated in the following way. Firstly, for the purposes of the calculations below, incorrect responses were discarded. Then the standard deviations of the RTs within each of the three stimulus categories (regular words, exceptions, and non-words) were calculated. Any response which lay more than 3 S.D.S above or below the mean for its category was discarded. This procedure resulted in the discarding of 1.16% of responses to regular words, 1.08% of responses to exception words, and 1.24% of responses to non-words. Most of these were unusually long responses, but some were extremely rapid responses, pre- sumably anticipations. Finally, in the case of words, each item which was paired with a discarded item in the matching of regular to exception words was itself discarded. For example, if subject 8’s response to the exception word GAUGE had been discarded (because it had been incorrect, or more than 3 S.D.S above the mean of his responses to exception words, or more than 3 S.D.S below the mean) his response to this word’s regular mate GRILL was also discarded, regardless of whether it was correct or not, and regardless of whether or not it was unusually long, or unusually short. This appears to be the logical way of dealing with two such tightly matched sets of words; it guarantees that for each subject the total set of exception words producing usable YES responses is still exactly matched to the total set of regular words, and it guarantees that for each word-pair the set of subjects contributing to the YES responses with the exception words is identical to the set of subjects contributing to the YES responses with the regular words.

After carrying out this discarding procedure, the means of the remaining

Dow

nloa

ded

By:

[Lan

dsbo

kasa

fn Is

land

s - H

asko

labo

kasa

fn] A

t: 00

:35

18 J

une

2008


responses were calculated across subjects and across items. These means, and the error rates, are shown for each stimulus category in Table 11, and a detailed breakdown of the means for each word is given in the Appendix.

TABLE I1

Mean RTs for correct YES responses, and error rates, in Experiment I ~

Regular words Exceptions

Mean R T over subjects 539 538 Mean R T over words 542 541 Error percentage 7'9 8.3

There was a difference of I ms between the mean YES R T to regular words and the mean YES R T to exception words, whether these means are calculated across subjects or across words. This difference was not significant; nor did the difference of 0.4% in error rates between the two word types approach significance (t3,,=0-65, P >0*05). In this experiment, then, regular words enjoyed no advantage over exception words in the time taken to decide that they are words.

Experiment I1

Method Stimulus materials

The exception words selected for Experiment I were a deliberately heterogeneous group of words embodying many different forms of letter-sound irregularity. The exception words used in the second experiment reported here are irregular in a single way, namely, they all contain vowel digraphs which are pronounced in a way which departs from the most common pronunciation. Twenty such exception words were selected, and nine different vowel digraphs were represented in these 20 exception words; the words are listed in the Appendix.

Lexical decision times for these exception words were to be compared with lexical decision times for matched regular words. If a single regular word was matched to each of the 20 exception words, this would produce a set of 40 words, and if there were also forty non-words, this would result in an 80-trial experiment, which would take about 5 min to run, including the time needed for instructions. Since it seemed wasteful to run such a brief experiment, we decided to use several control regular words for each exception word; this would improve the precision of the comparison between regular and exception words without resulting in an unduly lengthy experiment. An attempt was made to select four matched regular words for each exception word, but this was not always possible; for five of the exception words, only three closely matched regular words could be found, and for one of the exceptions words only two matched regular words could be found. Thus there were 73 control regular words in all.

For each exception word, the two, three or four words in its control set were closely matched with the exception word on word frequency (Kucera and Francis, I977), number of letters, number of syllables, number of morphemes, part of speech including inflections, and concreteness imageability. A set of 93 pronounceable non-words was also produced.

Another set of non-words (shown in Table I11 below) was also selected for the purpose of investigating whether those pronunciations of the vowel digraphs which are claimed to be the most common by Wijk (1966) and Venezky (1970) are in fact the most common when

The entire list of 93 words is given in the Appendix.

Dow

nloa

ded

By:

[Lan

dsbo

kasa

fn Is

land

s - H

asko

labo

kasa

fn] A

t: 00

:35

18 J

une

2008

499 PHONOLOGICAL ENCODING AND READING

subjects are asked to pronounce non-words containing the vowel digraphs. For example, of the four possible pronunciations of the digraph A1 (RAID, AISLE, PLAID, AGAIN) will subjects usually choose the RAID pronunciation, said to be the most common, when asked to say aloud the non-word ZAIB? The items for this test each consisted of a vowel digraph preceded and followed by a consonant, i.e. were of the form C W C .

When subjects pronounce such non-words, they may do so by relying on analogies with real words, rather than on GPCs (Baron, 1976). In an effort to minimise this strategy, the non-words for this test were chosen so that no English word began or ended with the initial CVV component of the C W C non-word, and no English word began or ended with the final W C component of the non-word. The Concise Oxford Dictionary was consulted for deciding whether any English words began with the C W or W C of a non-word, and Walker's Rhyming Dictionary of the English Language (Dawson, 1973) which is not a rhyming dictionary but a reverse dictionary, i.e. a dictionary in which the location of a word depends on the right-to-left order of its letters, was consulted for deciding whether any English ended with the C W or W C or one of our non-words.

On the assumption that the words BANZAI, KEA, DOAB, AIBLINS, DOAT, ZOUNDS and ZOUAVE were absent from the vocabularies of all our subjects, the only analogy available to them was from OUGHT to ZOUG.

Subjects

rate of Eo.50 per hour for their participation. time.

The subjects were twenty-nine undergraduates at the University of Reading, paid at a They were run in groups of up to five at a

Apparatus and procedure These were as in Experiment I, except that after completion of the experiment the

subjects were individually tested with the digraph-pronunciation non-word items. Each item was presented to the subject printed on a card, and he was asked to pronounce it in the way that it would be pronounced if it were an English word. His pronunciation was recorded.

Results and discussion The nine

non-words, their regular pronunciation, and the percentage of subjects producing each of these pronunciations, are shown in Table 111. It is evident that the

We consider first the results of the non-word pronunciation task.

TABLE I11

Percentage of subjects producing the regular pronunciation of each vowel digraph in Experiment 11

Example of regular Non-word item vowel pronunciation

ZAIB maid YAUF haul KEAJ sea ZEWK new DOAB boat JOOV food ZOUG couch or soup VOE toe ZOWF now

Percentage of subjects giving regular pronunciation

72'4 24-1 86.2 93'1 93.1 93'1 93'1 96.6 89.7

Dow

nloa

ded

By:

[Lan

dsbo

kasa

fn Is

land

s - H

asko

labo

kasa

fn] A

t: 00

:35

18 J

une

2008


pronunciations generated by the subject are in fact usually the regular pronunciations, except for the digraph AU, which was frequently pronounced to rhyme with COW. For the remaining eight digraphs, then, the irregular pronunciations possessed by the exception words used in the lexical decision experiment do seem to be the less common forms of pronunciation.

Firstly, for each of the 20 sets of two, three or four regular words used as controls, the meant R T of the correct responses of that subject within each set was calculated. From this point on, these 20 means were treated as single RTs matched with the RTs to the corresponding 20 exception words.

The same procedures were then applied to these data as had been used for the data of Experiment I. The percentage of correct responses discarded because they were more than 3 S.D.S away from the mean was 1.36%. After the discarding procedure had been completed, the means of the remaining responses were calculated across subjects and across items. These means, and the error rates, are shown for the two categories of words in Table IV. A detailed breakdown of the means for each word is given in the Appendix.

Each subject's lexical-decision data was treated in the following way.

TABLE IV

Mean RTs for correct YES responses, and error rates, in Experiment II

Regular words Exceptions

Mean R T over subjects 592 594 Mean RT over words 587 585 Error percentage 12'2 10.9

There was a difference of 2 ms between the mean YES R T to regular words and the mean YES R T to exception words, whether these means are calculated across subjects or across words. This difference was not significant; nor did the difference of 1.3 yo in error rates approach significance (tI8=0.647, P >o.o5).

This experiment produced the same result as Experiment I. There was no evidence whatsoever that lexical decision times are slower to exception words than to regular words, nor any evidence that the accuracy of lexical decision is influenced by whether a word is regular or an exception.

General discussion

The difficulty experienced by subjects in responding NO to such pseudohomophones as BRANE in lexical decision tasks, and the absence of this effect in deep dyslexia, indicates that pre-lexical phonological recoding occurs in these tasks. If it is accepted that the only workable method by which such phonological recoding can be achieved is by means of grapheme-phoneme correspondences, as we argued earlier, then such recoding can only be correctly performed for regular words. It follows that, if such recoding ever assists the YES response, there

Dow

nloa

ded

By:

[Lan

dsbo

kasa

fn Is

land

s - H

asko

labo

kasa

fn] A

t: 00

:35

18 J

une

2008


should be an advantage for regular words over exceptions. We have found no evidence for such an advantage.”

Thus a somewhat paradoxical situation has arisen : phonological recoding plays a part in NO responses, but not in YES responses. One way of understanding this, which we mentioned briefly in an earlier paper (Coltheart et al., 1977) is to suggest that phonological recoding is a relatively slow process. When a letter string is a word, its lexical entry is always reached by the visual route before phonological recoding can be completed, even though such recoding is being performed, Thus YES responses are always generated via the visual route. When a letter string is not a word, an unsuccessful attempt at lexical access is carried out; this takes longer than access to a lexical entry when the stimulus is a word, and so is still in progress when the phonological recoding has been completed. This allows difficulties caused by pseudohomophones to intrude and to delay the NO response.

Our conclusion that when a word is presented to a subject in a lexical decision task access to that word’s lexical entry never depends on a prelexical phonological recoding agrees with the conclusion reached by Frederiksen and Kroll (1976) but conflicts with that reached by Meyer, Schvaneveldt and Ruddy (1974). On each trial in the experiment of Meyer et al. (1974) the subject saw two letter strings (words or non-words), the second one occurring as soon as a response had been made to the first. Some of the words had ambiguous grapheme-phoneme correspondences, e.g. COUCH or BREAK. When such a word was preceded by another word which contained the same ambiguous letter sequence but with a different phonemic correspondence (e.g. when COUCH was preceded by TOUCH, or BREAK by FREAK) the YES response was slower than when there were no shared letter sequences (e.g. FREAK then COUCH, or TOUCH then BREAK) and slower than when there were shared letter sequences but no ambiguity (e.g. BRIBE-TRIBE or FENCE-HENCE). The explanation of this result proposed by Meyer et al. (1974, p. 317) was as follows: “Suppose that a word like FREAK has just been processed and that the string of letters B-R-E-A-K is encountered next. If it is phonologically encoded to rhyme with FREAK, then the resulting phonological representation will sound like BREEK, and will not be found in lexical memory. To avoid an error, the S would have to recode the second string in the correct phonological representation and check it during another pass through memory. According to this explanation, when a word has more than one possible phonological representation, these possible representations are tried one after the other, and a NO response made only after the last representation in this series has been sought unsuccessfully in lexical memory. Presentation of COUCH before TOUCH increases the priority of one of the incorrect phonological representations of TOUCH (the one which rhymes with COUCH) and hence lengthens the time before the lexical entry for TOUCH is accessed using the correct phonological representation. This explanation implies, of course, that subjects in a lexical decision task at least

*After this paper was submitted, a report by Stanovich and Bauer (1978) appeared: they found no regularity effect on lexical decision times with instructions stressing speed, but an effect with instructions stressing both speed and accuracy.

The repetition would produce relatively long RTs”.

Dow

nloa

ded

By:

[Lan

dsbo

kasa

fn Is

land

s - H

asko

labo

kasa

fn] A

t: 00

:35

18 J

une

2008


sometimes gain access to the internal lexicon via a phonological recoding of a visually presented word, since the COUCH-TOUCH effect occurs with the YES response.

However, the explanation offered by Meyer et al. (1974) appears to be contra- dicted by anothei feature of their results. In Condition 4 of their second experiment, every second word presented had at least two possible phonological representations. In Condition 2, less than one-third of the second words had two possible pronunciations. On the Meyer et al. (1974) hypothesis of serial testing of phonological representations, subjects would obtain the correct phonological representation earlier on the average in Condition 2 than in Condition 4, and therefore would respond YES faster in Condition 2 than in Condition 4. This did not happen; in fact the YES response was (non-significantly) faster in Condition 4. A second problem with these data is that when Becker, Schvaneveldt and Gomez (1973) repeated this experiment using a different set of words, they failed to observe any slowing of the YES response-subjects were no slower at responding YES to COUGH when it was preceded by DOUGH than when it was preceded by MINT. For these two reasons, the results of Meyer et al. (1974) cannot be taken as strong evidence for the use of phonological encoding when making the YES response in a lexical decision task.

We have accepted the view that GPCs are used in a naming-latency task-hence the superiority of regular words over exceptions observed by Baron and Strawson (1976)-but not for accessing lexical entries of words in a lexical decision task. Why is it that the two experimental tasks differ in this way?

A first possibility is that the differences between regular words and exceptions in the naming experiments of Baron and Strawson (1976) do not demonstrate an effect of the use of GPCs on the latency to begin pronouncing a word. There are at least two conceivable artifacts in these experiments. Firstly, exception words may differ from regular words not in naming latency but in naming duration. This would show up as a difference in the total time needed to utter a list of words, even if the naming latencies of exception words and regular words did not differ. Secondly, one could argue that even if naming of words relied exclusively on visual access to the lexicon, exception words may still suffer, in the following way. If phonological recoding always lags behind visual access but nevertheless does occur, at some point during the naming of a list of words the phonological recoding of an exception word will be completed. If subjects notice that this recoding conflicts with the way in which they actually pronounced the word (a conflict which would not occur with regular words), this may disrupt whatever processing they are carrying out at that moment, even though it is processing of words later in the list; hence a difference, in favour of regular words, in the time taken to name the whole list would emerge, even though no difference in individual naming latencies existed. These two difficulties show that it may be unwise to investigate the effects of GPC regularity on naming latency by measuring the time needed to name a list of exceptions or regular words. If individual naming latencies are measured, both problems are avoided.

In an unpublished experiment by Edgmon, described by Gough and Cosky (1977), precisely this was done: a set of 56 exceptions and 56 closely matched

Dow

nloa

ded

By:

[Lan

dsbo

kasa

fn Is

land

s - H

asko

labo

kasa

fn] A

t: 00

:35

18 J

une

2008

503 PHONOLOGICAL ENCODING AND READING

regular words was presented in random order, one word at a time, and individua naming latencies were measured. Median naming latency was 600 ms for regular words and 627 ms for exception words, a highly significant difference. A problem for individual-latency experiments which does not arise for list-reading experiments is that throat microphones are differentially sensitive to different initial phonemes ; however, Gough and Cosky report that the words used by Edgmon were matched on initial phonemes, so that the effect he observed cannot be an artifact of differences in initial phonemes in the two sets of words.

It should be noted that Mason (in press), measuring individual naming latencies for the forty regular and forty exception words used by Baron and Strawson (1976, Table I), failed to find a naming-latency advantage for regular words. A possible reason for this is that, as Baron and Strawson point out, their exception words had a higher mean word frequency than their regular words; since naming latency is inversely related to word frequency (e.g. Forster and Chambers, 1973 ; Frederiksen and Kroll, 1976), this frequency difference may have concealed an exception/regular difference in Mason’s experiment. We will assume, then, that the finding that regular words have shorter naming latencies than exception words is not artifactual.

However, even if not artifactual, this result is ambiguous with respect to theoretical interpretation. Even if we accept that the pronunciation of a regular word is assisted by the use of GPCs to derive a phonological code of that word, the naming-latency advantage of regular words could arise in either of two ways, one non-lexical and other lexical. The first possibility is that the phonological code is converted to an articulatory code and this code is then executed, with no use made of the word’s lexical entry. The second possibility is that the phonological code is used for lexical access, and from the lexical entry of the word an articulatory representation is retrieved and executed. Either possibility would produce a naming-latency advantage for regular words. Our failure to find a lexical- decision advantage for regular words, however, suggests that the second possibility may be rejected, since it appears that lexical access via a phonological recoding is always slower than visual access; if naming of words were always a lexical process, then regular words would not be named faster than exception words, since the phonological access to the lexicon available for regular words would not influence their naming latencies. Thus, when the naming-latency and lexical-decision data are taken in conjunction, they suggest that the advantage in naming latency enjoyed by regular words is contributed by a non-lexical route using GPCs to proceed from print through phonology to articulation, a route which is irrelevant for tasks requiring lexical access. An unresolved problem here is that, when exception words and regular words are intermingled randomly and individual naming latencies are recorded, as in the experiment described by Gough and Cosky (1977)~ a subject who is employing a non-lexical GPC-based route for pronunciation (in addition to a lexical route using visual access) would seem to have no way in which he could avoid deploying this for exception words as well as regular words. This would produce incorrect pronunciations for exception words on that proportion of trials where the non-lexical route was completed earlier than the lexical route. Sometimes, however, exception words are pronounced wrongly in naming latency

Dow

nloa

ded

By:

[Lan

dsbo

kasa

fn Is

land

s - H

asko

labo

kasa

fn] A

t: 00

:35

18 J

une

2008


experiments, and sometimes these wrong pronunciations are those produced by application of GPCs.

Although our position is not without its problems, then, we suggest that in naming latency experiments the subject is operating two routes in parallel and whichever is the fastest on a particular trial is responsible for his pronunciation on that trial. The first route consists of visual access to a lexical entry followed by retrieval of an articulatory code from that entry and execution of that code. The second route does not use the internal lexicon: it consists of the application of GPCs to a letter string to produce a phonological code, conversion of that code into an articulatory code, and execution of the latter code. The lexical route is available equally to regular words and to exceptions, but not to non-words, since these have no lexical entries. The non-lexical route is available equally to regular words and to “pronounceable” non-words, but not to exception words, since these are incorrectly recoded by GPCs. The naming-latency advantage of regular words over exceptions is due to the use of two parallel routes (with overlapping distributions of finishing times) vs. only one of these routes. On this view, regular words should show no latency advantage over exception words in the lexical decision task, since the route which gives them this advantage is a non- lexical one, and hence irrelevant when the task is lexical decision rather than naming aloud; and our results provide a demonstration that in the lexical decision there is no advantage for regular words over exceptions.

T o the extent to which lexical access, as defined by the requirements of the YES response in a lexical decision task, corresponds to lexical access as the first step in comprehending single printed words during normal reading, our results suggest that pre-lexical phonological recoding is not used in normal reading.

This problem clearly needs further attention.

This work was supported by Grant HR 4071 from the Social Sciences Research Council.

References BARON, J. (1976). Mechanisms for pronouncing printed words: Use and acquisition. In

LABERGE, D. and SAMUELS, S. J. (Eds), Basic Processes in Reading: Perception and Comprehension.

Use of orthographic and word-specific knowledge in reading words aloud. Journal of Experimental Psychology: Human Perception and Performance, 2, 386-93.

Semantic, graphemic and phonetic factors in word recognition. Presented at Psychonomic Society Meeting, St. Louis.

COLTHEART, M. (1978). In UNDERWOOD, G. (Ed.), Strategies of Information Processing.

COLTHEART, M., JONASSON, J. T., DAVELAAR, E. and BESNER, D. (1977). Access to the internal lexicon. In DORNIC, S. (Ed.), Attention and Performance VI. New York: Academic Press.

DAWSON, L. H. (1973). London: Routledge and Kegan Paul.

FORSTER, K. I. and CHAMBERS, S. M. (1973). Lexical access and naming time. Journal of Verbal Learning and Verbal Behavior, 12, 627-35.

Hillsdale, New Jersey : Erlbaum. BARON, J. and STRAWSON, C. (1976).

BECKER, C. A., SCHVANEVELDT, R. W. and GOMEZ, L. M. (1973).

Lexical access in simple reading task. London : Academic Press.

The Rhyming Dictionary of the English Language.

Dow

nloa

ded

By:

[Lan

dsbo

kasa

fn Is

land

s - H

asko

labo

kasa

fn] A

t: 00

:35

18 J

une

2008


FREDERIKSEN, J. R. and KROLL, J. F. (1976). Spelling and sound: Approaches to the internal lexicon. Journal of Ex$erimental Psychology : Human Perception and Perform- ance, 2, 361-79.

In CASTELLAN, N. J., PISONI, D. B. and Pons , G. R. (Eds), Cognitive Theory, Volume 2. Hillsdale: Erlbaum Press.

Dyslexia: A neurolinguistic study of traumatic and developmental disorders of reading.

Computational Analysis of Present-day American English.

Patterns of paralexia. Journal of Psycho- linguistic Research, 2, 175-99.

From print to sound in mature readers as a function of reader ability and two forms of orthographic regularity.

Functions of graphemic and phonemic codes in visual word recognition.

Stages in recovery from dyslexia following a left cerebral abscess.

Aphasia, dyslexia, and the phonological coding of written words.

Evidence for phonemic recoding in visual word recognition. Journal of Verbal Learning and Verbal Behavior,

SCARBOROUGH, D. L. and SPRINGER, L. (1973). Noun-verb differences in word recognition.

SHALLICE, T. and WARRINGTON, E. K. (1975). Word recognition in a phonemic dyslexic

STANOVICH, K. E. and BAUER, D. W. (1978). Experiments on the spelling-to-sound

VENEZKY, R. L. (1970). WIJK, 0. (1966).

GOUGH, P. B. and COSKY, M. J. (1977). One second of reading again.

HOLMES, J. M. (1973).

KUCERA, H. and FRANCIS, W. N. (1967).

MARSHALL, J. C. and NEWCOMBE, F. (1973).

MASON, M. (In press).

MEYER, D., SCHVANEVELDT, R. and RUDDY, M. G. (1974).

NEWCOMBE, F. and MARSHALL, J. C. (1973).

PATTERSON, K. and MARCEL, A. J. (1977).

RUBENSTEIN, H., LEWIS, S. S. and RUBENSTEIN, M. A. (1971).

Unpublished Ph.D. thesis, University of Edinburgh.

Providence : Brown University Press.

Memory and Cognition.

Memory and Cognition, 2, 309-21.

Cortex, 9, 329-32.

Quarterly Journal of Experimental Psychology, 29, 307-18.

10, 645-57.

Presented at Psychonomic Society Meeting, St. Louis.

patient.

regularity effect in word recognition.

Quarterly Journal of Experimental Psychology, 27, I 87-99.

Memory and Cognition, 6 , 410-15. The Structure of English Orthography. The Hague: Mouton.

Rules of Pronounciation for the English Language. London: Oxford

Received 8 August 1978 University Press.

Dow

nloa

ded

By:

[Lan

dsbo

kasa

fn Is

land

s - H

asko

labo

kasa

fn] A

t: 00

:35

18 J

une

2008

M. COLTHEART E T A L .

Appendix

Exception words and regular words used in the four experiments, and mean correct YES latencies for each word in each experiment

Experiment I

Exceptions Regular words

GAUGE AUNT LAUGH BREAK STEAK DEBT PINT SIGN MORTGAGE CASTLE COME GLOVE LOVE SHOVE LOSE MOVE PROVE GONE GROSS BURY BOROUGH THOROUGH SCARCE ANSWER SWORD YACHT SURE BLOOD FLOOD COUGH TROUGH BOWL SOUL BUILD BISCUIT CIRCUIT SUBTLE SEW BROAD

655 513 534 507 506 551 560 505 604 5 24 510 573 5 I4

535 497 533

563

528 561 533 602 593 604 5 27 520 553

523 491

579 472 5 27 504 507 553

488

576

582 538 571

GRILL GANG TREAT DANCE SLATE CULT PINE BASE DISTRESS SHERRY TAKE SPADE TURN SHRUG SAVE SORT SPEND KEPT QUICK DUEL CAPSULE SPLENDID STREWN COUNTY SPEAR TROUT FREE HORSE TOOTH BARGE THRONG PLUG MILE CHECK SHAMPOO PROTEIN STUPID RUB FRESH

507

519 509

608 514 507 532

560

583

518 487

506 564

526 538 533 5x6 554 602 585 65 5 538

484

536

643

508 489 556 578

595

544

529 544

524

5 94

523

520 533 473

Dow

nloa

ded

By:

[Lan

dsbo

kasa

fn Is

land

s - H

asko

labo

kasa

fn] A

t: 00

:35

18 J

une

2008


Experiment II

Exception words

~

Regular control words

~

EXP. 2 (Normal)

Mean of Exception regular

word controls

AISLE MAUVE SAUSAGE AUNT GAUGE BREAK STEAK SEW BROAD CANOE SHOE BLOOD FLOOD HOOD WOOL FOOT DOUBLE POUR BOWL DISOWN

FROST SPINE SPOON STAIN 623 BRASH SLEEK 724

SOAP GANG GULF HERD 638 COUCH GRILL SHELF SPARK 664 CHECK DANCE SIGHT SHAPE 575 DITCH FLOCK MOUSE SLATE 607 DIP PRY RIP RUB 566 FRESH ROUND TERSE 635 BASIN CABLE CARGO FILLY 654 BARK FORK HAWK MIST 525 HORSE PLANT PLANE 490 CROWN SLOPE TOAST SPELL BULB LACE LUMP MINT 555 SLUG PORK COIN DRUM 498 SONG RAIN DUST WINE 522 ATTEND REDUCE SAMPLE 530 TOSS YELL HOOT WIPE 567 DECK LOCK PLUG SINK 544 BEFELL DEPART ENRAGE REGAIN 728

SPANGLE THICKET CUTLASS CURRANT 707

5 20

575 661 701 554

571 575 570 583 619

607

560

573

601 555 544 549 5 87 5 87 541 717

This shows the 20 exception words used in Experiment 11, the mean latency of correct YES responses to each word, and the mean latency of correct YES responses to each of the sets of regular words serving as control items for the exception words.

phonological encoding in the lexical decision task

Documents