Abstract. Three patients with semantic dementia, involving progres- sive deterioration of semantic memory, performed immediate serial recall of short sequences ...
The Impact of Semantic Memory Loss on Phonological Representations Karalyn Patterson MRC Applied Psychology Unit, Cambridge
Naida Graham and John R. Hodges University of Cambridge Clinical School, Addenbrooke’s Hospital
Abstract Three patients with semantic dementia, involving progressive deterioration of semantic memory, performed immediate serial recall of short sequences of familiar words. On the basis of their performance in other tasks of word comprehension and production, the stimuli were selected individually for each patient as either known or unknown words. All patients showed a marked advantage in recall of known as compared to familiar but now unknown words. Errors consisted primarily
of incorrect combinations of correct phoneme sequences in the stimulus string, with a large number of errors preserving onsethime syllable structure (e.g., mint, rug reproduced as “rint, mug”). Discussion focuses on the implication of these errors for the structure of phonological representations, and in particular on a hypothesis that meaning plays a crucial role in binding the elements of phonological word forms.
INTRODUCTION
domains such as recognition of objects and people; but some of the most striking and early impairments are seen in the domain of language. The typical pattern of language deterioration in semantic dementia is a progressive fluent aphasia (as described by Poeck & Luzzatti, 1988), with severe anomia and reduced meaningful content of speech, but-at least superficially-relatively intact syntactic and phonological structure of speech output. The purpose of the present study was to assess the integrity of the patients’ phonological representations. This goal would be impossible to achieve by analyzing their language production in spontaneous speech or structured output tasks such as naming or picture description: all of the patients are severely anomic, and one of the three has such an extensive loss of word meaning that she says little. It would also be difficult to assess the patients’ phonological representations with a task of single-word repetition: they are virtually perfect at repeating individual words and can even repeat single nonwords. As a method of encouraging the subjects to produce a sufficient number and variety of words, we asked them to listen to, and then immediately reproduce, sequences of three or four spoken words. Following a technique originated by Warrington (1975), the words to be recalled were selected individually for each patient and assigned to two classes: known words, i.e., those that the subject’s spontaneous speech and test performance suggested that he or she still knew, and unknown words,
Errors that people make in word production provide important evidence concerning the nature of phonological representations. Much of this evidence comes from the study of errors in spontaneous speech by normal speakers (see, for example, various chapters in Fromkin, 1980); but because normal speech errors are a relatively rare phenomenon, investigators sometimes resort to more controlled tasks, or to abnormal populations, to study the nature of speech errors. This report concerns word-production errors by three patients with semantic dementia, a form of progressive fluent aphasia (Hodges, Patterson, Oxbury, & Funnell, 1992; Snowden, Goulding, & Neary, 1989), in a task of immediate serial recall of short sequences of words. The three patients have already been described in two publications on semantic dementia, one covering a broad range of the patients’ cognitive performance (Hodges et al., 1992) and the other specifically concerned with their reading performance (Patterson & Hodges, 1992). Semantic dementia is a progressive condition, typically associated with atrophy of temporal neocortex (sometimes predominantly in the left hemisphere), in which disintegration of semantic memory or knowledge is hypothesized to be the primary deficit (Hodges et al., 1992; Saffran & Schwartz, 1993; Snowden et al., 1989;Warrington, 1975).As would be expected from a general semantic deficit, these patients are impaired in many nonverbal 0 1994 M ~ u s e t t Institute s of Technology
Journal of Cognitive Neumckme 61,pp. 57-63
a highly reliable advantage for known over unknown words (all chi-squared tests, with 1 df, yield values significant atp < 0.001). Table 2 presents her performance, analyzed in the same manner but now for three-word sequences only, over the four testing sessions between May 1991 and June 1992. As explained in the Methods section, it is only sessions 1 and 4 for which we have some confidence in the assignment of words to the known condition. As the significance values (for chisquared tests) in Table 2 reveal, the knownlunknown comparison is highly significant, both for whole sequences and words correctly repeated, in sessions 1 and 4. In contrast, the same comparison is either marginally significant (for the words measure in session 2 ) or nonsignificant (for words in session 3, for whole sequences in both sessions 2 and 3 ) when the assignment of words to the known condition was out of date. The reduced difference between known and unknown words at sessions 2 and 3 is almost entirely attributable to decreased recall of known words, relative to session 1, presumably due to the inexorable decline in PP’s semantic memory and vocabulary. When asked at the end of session 4 to repeat each word presented individually, PP correctly reproduced 35 of 36 = 0.97 h u m and 34 of 36 = 0.94 unknown words, confirming that her repetition of single items is nearly flawless. The results of an analysis of errors from all four sessions of the experiment are displayed in Table 3. PP failed to reproduce the correct word in the correct position on a total of 355 stimulus items; 16 of these (1.6% of words presented, 4.5% of errors) were omissions, and are not considered further. The remaining 339 errors of commission comprise the analysis in Table 3. For purposes of this analysis, errors on hzoun-z and unknown words were combined, for the following reasons. First, since PP made almost exactly twice as many errors overall on unknown as on known words, the majority of errors were, in fact, on words from the u h o w n condition. Second,where the number of errors on so-called known words was relatively high, that is in sessions 2 and 3, there is probably no legitimate basis for considering that these words still w e in PP’s vocabulary at the time of testing. Third, even in sessions 1 and 4, there was undoubtedly a degree of uncertainty in our assignment of words to conditions. Finally, there was no notable differ-
i.e.,those to which the subject responded incorrectly on comprehension tests. The results will be used to discuss two different, though linked, theoretical issues. The major focus, addressed by the patients’ relative success in reproducing words from their known and unknown sets, is the relationship between semantic and phonological representations, and in particular how the integrity of phonological word-forms depends on an intact semantic system. Meaning and phonology constitute separate modules in all theories of speech production, with the corollary that each should be independently vulnerable to brain injury or disease; but structural separability is not equivalent to functional independence. The question of whether phonological word-forms can be maintained intact when associated meaning deteriorates is an unanswered, indeed a largely unasked, question. Patients with relatively pure impairments of semantic memory may provide crucial evidence on this question. The second issue, addressed by the nature of the patients’ errors in reproducing word sequences, is the structure of phonological representations themselves. Although three subjects participated in this study, the main focus of the paper will be longitudinal data from one of the three, PP, a patient with profound loss of semantic knowledge (Hodges et al., 1992). Results from single testing sessions with the two other patients, FM and JL, will be presented to enable assessment of the generality of findings with PP. The basic procedure (see Methods for details) in all cases was to present sequences of three or four words, at a rate of one word/sec, for immediate reproduction by the patient. Each sequence was composed of either known or unknown words, and sequences from the two sets were presented in alternation within an experimental session.
RESULTS PP.Table 1 presents PP’s performance, as proportion of whole sequences correctly reproduced and proportion of words correctly reproduced, from h w n and unknown sequences of length three and four, when the test was first administered in May 1991. The table shows a dramatic contrast, at both sequence lengths and for both measures, between the two classes of words, with
Table 1. PP’s Performance (May 1991) in Immediate Serial Recall of Three- and Four-Word Sequences of Known and Unknown Words, Given as Both Proportion Correct of Entire Sequences and as Proportion Correct of Words“ Whole Sequences
Words
Known
unknown
Known
Unknown
Length 3
.78
.11
.93
.65
Length 4
.61
0
.86
.47
“N = 18 known and 18 unkrwwn sequences at each length (hence N = 54 for each type of word at length 3 and 72 at length 4).
58
Journal of Cognitive Neuroscience
Volume 6, Number 1
Table 2. PP’s Performance (on Four Different Test Sessions) in Immediate Serial Recall of Three-Word Sequences of Known and Unk.nown Words, given as Both Proportion Correct of Entire Sequences and as Proportion Correct of Words.“ Whole Sequences
WW&
Known
unknown
p-value
Known
1 (May 1991)
.78
.11
.001
.93
.65
.001
2 (Nov 1991)
.44
.22
ns
.78
.59
.05
3 (Mar 1992)
.33
.ll
ns
.67
.59
ns
.53
.08
,001
.76
.48
.001
4 (June 1992) ~
p-value
U&OWn
~~~~~~
“For sessions 1-3, N = 18 known and 18 unknown sequences (hence N = 54 for each type of word); for session 4, N = 36 for each sequence type and 108 for words. For sessions 1 and 4, assignment of words to known and unknown conditions was based on a recent assessment of PP’s language abilities; for sessions 2 and 3,the word lists from session 1 were used. p-values are from chi-squared tests with 1 df.
Table 3. Proportion of Different Error Types in PP’s Reproduction of Three-and Four-Word Sequences“ Type
Examples
Proportion
Monosyllabic (N = 156 errors) Wrong onset, correct rime
comb + /houm/
.53
blade -9 /lei& Wrong terminal consonant, correct onset + vowel
rake -9 /reit/ wine -9 /ward/
.25
Wrong initial and terminal consonants, correct vowel
thorn -9 ( f a d sop -9 /tJOV
.17
Other
bell + /pael/ bird + /bad
.04
Wrong onset, correct remainder
cbickm -9 /‘sikan/ writing --* /‘wartig/
.49
Wrong onset, correct rime in subsequent syllable (or > 1)
parachute -9 /‘pcrasutl elephant + /‘tclamant/
.22
Other
ostrich -9 hstrrd$ button -9 /‘b^ta/
.29
Multisyllabic (N = 183 errors)
Errors preserving correct vowel(s) Monosyllabic
.96
Multisyllabic
.91
Errors preserving correct stress pattern in multisyllabic words
.99
“PP speaks standard southern British English, represented with the following phonetic symbols: uowels:/a/ as in tart,/il as in cheese, /I/ as in fish, /E/ as in chef, /o/ as in lobster, /J/ as in prawn, N as in truffle, /u/ as in food, /u/ as in cook, /3/ as in shirred, /ae/ as in crab, /el/ as in quail, /ru/ as in wine, /au/ as in trout, /ou/as in sole, 1311 as in oil, and /a/ as in the second syllable of cater; co12sonants: all as their alphabetic representations would suggest, except /g/ as in icing. /d3/ as in ginger, /i/ as in yam, /I/ as in sherry, /ti/ as in chocolate, /8/ as in thin, and 16 as in then. For multisyllablic words, ’ marks the beginning of a stressed syllable.
ence between error types on supposedly known and unk?nouw words. The words in the error analysis are divided into monosyllabic and multisyllabic words, not because PP’s performance on these differed (as it happens, she correctly reproduced exactly 65% of both monosyllabic and mul-
tisyllabic words) but because the types of errors afforded by these two classes of words are slightly different. The categories to which errors have been assigned are as follows (see the note to Table 3 for a key to the phonetic transcription used here). PP’s most frequent error to a monosyllabic word was Pattmetal.
59
to produce the correct rime of the word with the wrong onset. ’I)@xllythe whole onset was replaced, even if it consisted of a consonant cluster (e.g., @it + hut/), though there were also a number of instances of simplification of an onset cluster (e.g.,fiit + /ht/ and stool + ItuV). Approximately half of the errors in this category were perfect “Spoonerisms,” where the rimes of two words in the string maintained their correct positions and simply exchanged their onsets (e.g., mint rug + /rInt/ /mAg/).In many cases where the replaced onset of an error did not come from the onset of another word in the sequence, it was a phoneme present somewhere else in the stimulus string (e.g., in syr’nge blade wine + syringe blade /rain/, the onset of /ram/ could be an import from the second, stressed syllable of syringe). Occasionally the substituted phoneme had not occurred in the stimulus string (e.g., sunny rememberfeel + sunny remember /piV). Next in order of probability of errors to monosyllabic words, with less than half as many exemplars as the first category, was a pattern of correct onset and vowel with an incorrect terminal consonant. Only a few of these were the parallel of Spoonerism errors, where two items in the string exchanged terminal consonants; more typically, one word simply duplicated the final phoneme of another (e.g., in hammock wall dart + hammock wall /dak/, the terminal consonant of hammock turns up again in /dak/;and in m e oyster bird + /roud/ oyster /s3d/, the error on rose incorporates the final consonant of bird + /s3d/). PP’s third type of error on monosyllabic words was to reproduce only the vowel of the target word, initiated and terminated by incorrect consonants (e.g., key + him/, m e + kt/. As in the categories above, many of the incorrect phonemes could be interpreted as imports from elsewhere in that stimulus list, though not necessarily from the corresponding position in the word (e.g., in rakeparachute m e + /reIp/ parachute k t / , the initial flc/ of kt/may come from rake and the terminal /t/ from parachute). Finally, the small proportion of “other” errors includes one morphological error (say + saying), three in which the only phoneme retained from target to response was the terminal consonant (e.g., bell --* /pzV>, two that retained only the initial consonant (bird + Am), and two with no phoneme in common between stimulus and response (e.g., hatp + /w3s/). PP’s most frequent error to a multisyllabic word was to reproduce everything correctly except the onset (e.g., mother + /‘rA6a/,parachute + /‘kcraJutf). A few of these were exact Spoonerisms (e.g., stirrup) daisy + /Idtrap/ /‘sterzi/, but most were duplications of onsets (worried fair s&ter + worried fair /'fists/) or of other phonemes from elsewhere in the list (candle w ord cotton + /‘tzndaV sword /‘man/. A second category of error on multisyllabic words, not the next highest in frequency but the second most coGO
Journal of Cognitive Neumcime
-
herent as a category, is one that retains the rime of a syllable with the wrong onset, but in a position other than the initial syllable (e.g., supper /sAfa/,or sometimes in both the initial syllable and a subsequent one (e.g., elephant + /Itclamant/. Finally, there was a more mixed category of errors on multisyllabic words (e.g., lobster + /’Ir>dsa/ has errors on the terminal consonant of the first syllable and the onset of the second; hammock + /‘=ma/ has the reverse pattern; and insect --* /sIndct/ has one final consonant and two onset errors). Although the individual error types and their frequencies of occurrence are of considerable interest, perhaps the most striking thing about PP’s word production errors overall is displayed in the last few lines of Table 3. In monosyllabic words, she almost always (0.96) reproduced the correct vowel; in multisyllabic words, .99 of her errors faithfully reflected the correct stress pattern of the target word, and the majority (0.91) reproduced the exact vowel pattern. Moreover, there is a hint of a progressive deterioration in this latter measure: over the four test sessions, the proportions of multisyllabic errors with a vowel pattern identical to the presented word were 0.97, 0.92. 0.91, 0.84. FM: FM’s performance, shown in Table 4 , reveals a highly significant advantage for known as compared to unknown words, both for whole sequences correct [~’(l) = 11.7, # C 0.001] and for number of words correct (x’ = 16.1,p < 0.001). FM made a total of 21 errors, 2 on known words and 19 on unknown words. Unfortunately from the point of view of error analysis, 12 of these 21 errors were omissions, leaving a very small number of word production errors with which to draw conclusions about the integrity of FM’s phonological representations. However, of the nine errors of commission (all on multisyllabic words), five were cases where only the onset of the word was in error and the remainder was correctly reproduced (e.g., trumpet --* /’lAmpar/;guitar + /br’W,cherry + /‘t3ri/). None of these was an exact Spoonerism, where two words simply exchanged their onsets; but for three of the five, the substituted onset was the initial phoneme of another word in the sequence (e.g., the sequence producing trumpet --* /‘l^mpat/ included the word leopard). The remaining four errors of commission included some phonemes from the target word and some that may have been transposed from a neighboring word (e.g., leopard, squirrel + /‘IcpaV). Although proportions from such a small data set should obviously be treated with caution, of FM’s nine word production errors, eight (0.89) preserved the correct stress pattern of the target word and seven (0.78) maintained the correct vowel pattern. JL:Similar to results from PP and FM, JL’sperformance (Table 4) reveals a dramatic advantage for known relative to unknown words, statistically reliable both for whole sequences [ ~ ’ ( l=) 10.8,p < 0.001] and for number of words (x’ = 28.3,p < 0.001). Furthermore, it is possible Volume 6, Number 1
Table 4. Performance by FM and JL in Immediate Serial Recall of Four-Word Sequences of Known and Unknown Words, Given as Both Proportion Correct of Entire Sequences and as Proportion Correct of Words“
FM
.89
.33
.97
.74
JL
.53
.13
.83
.50
“For FM,N = 18 known and 18 unknown sequences at each length (hence N = 72 for each type of word); forJL,N = 30 of each type of sequence and hence 120 for each type of word.
to rule out frequency as the major source of the kmownunknown performance discrepancy for JL. In JL’s complete word sets, the mean/median word frequencies (from Francis & Kucera, 1982) were, respectively, 51.91 29.5 for the known words and 12.0ff.5 for the unknown words. If the 22 highest frequency words are removed from the known set, the meadmedian frequencies of the remaining 38 words are 14.6/6.5, a reasonably close match to the unknown set. JL correctly repeated 78%of these lower frequency known words, an insignificant change from the 83% correct repetition of words from the whole known set, and still reliably superior (x’ = 1 4 . 9 ,< ~ 0.001) to his performance on unknown words at 50%. (This technique of post hoc frequency matching could not be used with PP, who had too few known words, or with FM, who had too few unknown words.) The error corpus for JL is nowhere near as substantial as for PP, because he was tested on one occasion to her four, and also because quite a few of his errors, unlike PP’s, were either omissions of target words or intrusions of whole words from preceding sequences. However, JL made 55 within-sequence errors of commission, which (unlike FM’s meagre nine) seems a large enough sample to warrant detailed description. The relevant error categories and the proportions in each of these are shown in Table 5. Some of the features of the pattern found for PP are also present for JL, such as a greater likelihood that monosyllabic words will split between onset and rime than between onset-plus-vowel and terminal consonant; but JL’s errors are much more variable and therefore the pattern is not as straightforward as for PP. In particular, his largest error category (58% of errors on multisyllabic words) seems to reflect a less clear-cut version of the disintegration of word-forms, in which whole syllables or even more unpredictable segments of words separate and recombine. Also, it is sometimes difficult to be confident about the source of the various phonemes inJL’s errors. For example, when he repeats rooster as /Irma/ in the sequence thimble, swing, vase, rooster, it is tempting to claim that the /I/ in /‘rIzta/ comes from thimble and/or swing and the /id from vase; but there is no way to verlfy this claim. The summary values at the bottom of Table 5 show that, when making an error, JL was more likely than not to reproduce the correct vowel of a monosyllabic target
word and the correct stress pattern of a multisyllabic target word; but because he often imported a whole or part syllable from another word in the string, his errors were rather unlikely to reproduce the exactly correct series of vowels in a multisyllabic word.
DISCUSSION All three patients showed a dramatic advantage in immediate serial recall of three- or four-word strings for known relative to unknown words, where kmown items were defined as words that they produced appropriately in object-naming tasks (FM and JL) or spontaneous speech (PP), and unknown words were items that each patient failed in word comprehension tests. The majority of errors involved migrations of phonological segments among words within the sequence. Several lines of evidence suggest that the superior performance on krnown words reflects something about the patients’ own language capabilities, rather than a general difference (such as word frequency) between the items in the two assignment groups. First, for PP, in test sessions 2 and 3 where there had been no updated assignment of words to the known condition, the benefit for known words was either diminished or virtually absent, indicating that they possessed no inherent advantage over the unknown words. Superior performance for known words then re-emerged in session 4 following a new selection of items. Second, for JL, the reliable difference between word classes was undiminished when the most frequent items were removed from the kmown set to produce a close frequency match between &own and unknown items. Before we attempt to interpret this dramatic deficit in production of unknown words, two comments may help to highlight its importance. First, the phenomenon is not restricted to the repetition technique employed in these experiments. When PP arrived for an appointment wearing a coat that we had not seen before, one of us asked “Is that a new coat?” She replied, ‘‘I don’t understand, what did you say it was, a noat?” Recall of word lists was selected as a vehicle for assessing phonological representations because all three patients, especially PP, have severely impoverished speech output. Unless given appropriate stimulus input in the form of a spoken or P a t t m n mal.
61
Table 5. Proportions of Different Error Types in JL's Reproduction of Four-Word Sequences." Examples
Tvpe
Proportion
Monosyllabic ( N = 32 errors) Wrong onset, correct rime
swing + /rig/ stool + /puu
.34
Wrong terminal consonant, correct onset + vowel
goat + /gouts/ desk /dest/
.19
Wrong initial and terminal consonants, correct vowel
pipe + /rait/ bluse /pautJ/
.13
Migration of syllable or part-syllable
bowl, pumpkin + pumpJpullanl leaf, bicycle + /lais/, bicycle
Wrong vowel, 1 or 2 consonants correct
frog + /pragl
.25
Migration of syllable or part-syllable
kettle, barrel + /'kcral, 'baetaY pigeon, carpet + /'piQat/ celery, trumpet + celery, /'tr^mpi/ shirt, potato, giraffe + shirt, potato, /Jaft/
.58
Plural error
potato + potatoes
.06
Other
beetle + /'sedaV
.36
-
Multisyllabic ( N = 33 errors)
Errors preserving correct vowel(s) Monosyllabic
.66
Multisyllabic
.15
Errors preserving correct stress pattern Multisyllabic
.70
aOmissions and intrusions of words from preceding sequences have been excluded, leaving a small corpus of errors. See Table 3 for key to IPk
written word, they produce a sparse variety of different lexical types (though sometimes, in the case of a highly fluent patient like FM, many tokens of those types), and of course they never spontaneously produce previously familiar but now unknown words. Repetition of word lists was therefore useful for evoking production of the appropriate range of words; but it seems likely that the results of this technique merely reflect what would be observed in extensive conversations with the patients about new coats, etc. In other words, we claim that the patients' performance informs us about their everyday language comprehension and production, not just about their recall of word lists. The second point is that the quantitative and qualitative deficit obtained for unknown words was by no means obvious prior to this study (plus a very few others in the literature suggesting similar effects, e.g., Breedin & Saffran, 1993; Martin & Saflran, 1990; McCarthy & Warrington, 1987). All theories of speech production represent meaning and phonological form as separate modules, and although spontaneous speech clearly relies on efficient communication from one to the other, repetition is a different matter. For a person with normal auditory 62
Journal of Cognitive Neumcience
processing and normal ability to produce the sound patterns of hidher own language, the task of repeating familiar words is so easy and automatic that it has been described as a "privileged loop" (McLeod & Posner, 1984) that can operate without reference to meaning. For words similar to those employed here, normal adult subjects have a span of about five items (Baddeley, 1966; Brener, 1940); the patients, who have entirely normal auditory-verbal short-term memory capacity as measured by digit span, and who must (premorbidly) have had well-established phonological representations for words such as chicken and sbirt, are virtually incapable of reproducing a single sequence of three or four such (now unknown) words. This is a dramatic testimony to the impact of deteriorated meaning on the integrity of phonological word forms. Our interpretation of the pattern of observed effects, already expressed in connection with several other lines of inquiry (Patterson & Hodges, 1992;Patterson & Marcel, 1992), is as follows. First, and least controversially, we propose that words for speech production are represented as sets of phonological elements rather than as preassembled packages. As Levelt (1992) expresses it, "A Volume G,Number I
word’s phonetic form is not a ready-made template that can be retrieved as a whole. Speech error research has made it abundantly clear that a word’s ultimate shape is to be constructed time and again” (p. 9). The second and more controversial aspect of our proposal is that, in the normal speaker, the phonological elements of a word are helped to emerge in the right combination by two different factors that we have termed sources of coherence or “glue.”The more obvious kind of coherence is to be found within the phonological representation itself: because the elements of a word are activated together every time that the speaker utters the word, the elements should establish strong links in an autoassociative net. There is, however, another interaction that occurs whenever a person hears or produces a word, and that is the link between the word’s pronunciation and its meaning. We speculate that this link also serves a binding function and that, when there is deterioration of components of semantic memory that define a word, one major source of the coherence of the word’s phonological representation is lost. In a more general form, the basic idea underlying this hypothesis is that cognitive computations are inherently variable and noisy, and that the process whereby a representation at one level “settles down” or gets “cleaned up” sufficiently to support a response is partly accomplished by communication from other levels. This concept, probably derived originally from the interactive activation model of McClelland and Rumelhart (198l), is now widespread in cognitive theories; see, for example, Chertkow, Bub, and Seidenberg (1989), Plaut and Shallice (1993), Taft (1991), and Van Orden, Pennington, and Stone (1991). The absence of semantic binding should not, in principle, corrupt or distort the individual elements of a phonological representation. That is why, on our hypothesis, the patients are successful when asked to repeat single words (or even nonwords) or to read single words with regular spelling-sound correspondences (see Patterson & Hodges, 1992). We would argue that these latter tasks can be adequately performed when only subword phonological elements are available. Provided that a person can access the correct set of elements in the correct order, these can be concatenated to produce the correct whole-stringpronunciation, even if it has no meaning as an extra source of binding. This is how normal speakers repeat and read single nonwords, and how the patients repeat single real words whose meaning they no longer know (though of course the latter task may benefit from the purely associative phonological connections for familiar words discussed above). What happens, however, when the phonological system is stressed by a sequence of three or more items? If there is no extra glue between phonological elements from links to the semantic system, then the job of keeping all of the elements of each item together will be more difficult, and there will be a risk of element N from item 1 combining with element N + 1 of item 2. This is reflected in the errors of our patients
in serial recall of unknown words, and in those of normal subjects in serial recall of nonsense words (Treiman & Danis, 1988). The technique of comparing immediate recall for known and unknown words was introduced by Warrington (1975), in a study of two patiene (EM and m), and was also employed by McCarthy and Warrington (1987) with one patient (NHB). These three previously studied patients had characteristics very similar to those in our investigation (though none of the three was as severely impaired as PP) but produced rather different patterns of performance in this paradigm. In Warrington’s (1975) experiment, items were assigned to known and unkrnown sets on the basis of the patients’ ability to define words (no details are given of criteria for scoring these definitions), the sequence length was 5 words, and only 15 words in each category were used (with each word occurring a number of times across different sequences in the experiment). Although there was a slight numerical advantage in recall of known relative to unknown words for both patients, the difference was not significant. Six months later, a different version of the experiment was performed, in which Warrington compared each patient’s recall of the same unknown words with recall of nonwords (nonsense syllables). Here, both patients achieved significantly better performance on the unknzown words than the nonwords. In McCarthy and Warrington’s (1987) study, words were assigned to known and unknown sets on the basis of NHB’s ability to define pictures and words, the sequence length was 3 words, and there were 12 words in each set. NHB achieved virtually identical repetition performance on the two sets. In contrast, when asked to reproduce six- or seven-word sentences each containing a known or unknown word, he showed a marked effect, correctly repeating 61% of the sentences with a known word but only 17% of those with an unkmwn item. Furthermore, the authors comment on NHB’s errors in this sentence-repetition task, which were very similar to those of the three patients in our study, with many migrations of phonemes between words, e.g., theflag was colored bngbt red “the blag was fullered with a right breg” (McCarthy & Warrington, 1987, p. 1555). It is never possible to identify, with certainty, the reasons for different patterns of performance by similar (but of course not identical) patients on similar (but not identical) tests; nonetheless, several points of contrast seem worth making. First, the method used to select knownunknown words was rather different in these studies. Second, all three of the other patients were near ceiling levels of performance on the unknown words, which may have masked a potential effect of the known-unknown manipulation. Third, in the Warrington study, it would have been of substantial interest to have data for the known-unknown contrast on the second testing occasion, when EM’S performance on unknown words had declined markedly. If EM still had some limited Pattmon etal.
63
semantic knowledge about the unknown words at the time of initial testing, then perhaps the absence of a significant advantage for known words is not so surprising, and the effect might, if tested, have emerged on the second occasion. As a final point, although our hypothesis would of course have predicted a significant advantage for known over unknown words for these three other patients, we do not think that it is weakened by the significant advantage that all three showed for unknown words over nonwords. Indeed, our patients also show a substantial lexicality effect in repeating sequences (these data have not been included here, partly because nonword recall was not assessed on all of the relevant testing occasions, but mainly because our interest is in performance on real words that are comprehended or not). An advantage for unknown words over nonwords is consistent with the view that phonological representations derive some of their coherence from links to meaning, both because the other hypothesized source of this coherence (purely associative links within the phonological lexicon) will always favor real words, and also because our claim is that the patients have lost a signgcunt degree of semantic knowledge, not necessarily all of it, for unknown words. We turn now to the other major issue addressed by the data from this study, namely, implications of the patients’ word reproduction errors for the structure of phonological representations. When incorrectly reproducing a target word, both PP and FM almost unfailingly produced the correct number of syllables and (for multisyllabic words) the correct stress pattern; and although JL’s performance was less consistent in this regard, 0.70 of his errors to multisyllabic words also preserved the correct stress pattern. The suprasegmental accuracy of the patients’ errors supports the view that the “frame”of a word’s pronunciation is largely independent of the content that fills the frame. This notion has been described as “the most fundamental insight from modern speech error research’ ( b e l t , 1992, p. 10). Our data suggest that, at least in the task of word-list reproduction, the frame of an unknown word is substantially more durable than its specific phonemic content (perhaps with the exception of vowel phonemes, see below). The word frame apparently specifies not only such suprasegmental aspects of phonology as the number of syllables and their stress pattern, but also some characteristics of within-syllable structure. To use Levelt’s example and analysis, in a Spoonerism speech error such asf i l like playing + “peel like flaying,”having omitted the segmental information It7 in constructing the pronunciation of feel, the speaker nevertheless said “peel” rather than “eel”; this suggests that a frame requiring a consonant onset was already active, and was filled by the available segment /PI. Likewise, the majority of the patients’ errors respected the onset-rime frame of the target word, not only in perfect Spoonerisms but in other errors as well. For example, in PP’s error wowied fair 64
Journal of Cognitive Neuroscience
dter + worried fair /‘fist/ la/, although she appears to have lost the initial phoneme of sister altogether, she reused the intial phoneme of fair rather than violating the frame and producing IS^/ /a/. Performance by both PP and FM was also dramatic in its preservation of the correct vowel phoneme(s); once again this was true to a lesser, though non-negligible, extent for JL. Vowels may come through unscathed, even when the initial and/or terminal consonants of syllables do not, simply because vowels carry most of the energy in speech. They will therefore be prominent in the initial representation of each stimulus word, and so most likely to survive the stages of working memory and speech production required between hearing a word sequence and reproducing it (Brady, Shankweiler, & Mann, 1983). This prominence of vowels does not appear to characterize the serial recall of lists of CVC nonsense words by normal subjects: in the data from Treiman and Danis (1988), when only one phoneme from a CVC nonword was correctly reproduced, it was no more likely to be the V than one of the Cs, and indeed the phoneme most likely to survive was the initial C. In other regards, however, there is striking similarity in the errors of normal adults for sequences of six nonsense words (Treiman & Danis, 1988), of third-grade children for sequences of five words (Brady et al., 1983), and of our patients for sequences of three to four unknown words. In particular, in all three sets of data, errors of onset, leaving the rime of a syllable intact,occurred substantiallymore often than errors maintaining the initial C plus V. This pattern supports theories of phonological structure (e.g., Fudge, 1987; MacKay, 1972) that propose a three-level hierarchy of the syllable, in which the syllable consists of an onset plus a rime and the rime then splits into (vowel) peak and coda, rather than the alternative two-level view (Clements & Keyser, 1983) in which the syllable has three equal constituents. Returning briefly to our main hypothesis, that loss of semantic knowledge has direct consequences for the integrity of phonological form, it should be noted that the hypothesis fits other aspects of these patients’ cognitive performance. For example, we argue that their acquired surface alexia (a significant deficit in assigning the correct pronunciation, when reading aloud, to words with atypical spelling-sound patterns) is a reflection of deteriorated semantic representations (Patterson & Hodges, 1992). In computing a phonological representation for a written word, several different alternative pronunciations may be activated,especially for exception words like PINT whose spelling-sound relationship deviates from that of other words with a similar orthographic structure (MINT, LINT, PRINT, etc). The existence of a semantic representation corresponding to the correct pronunciation of PINT may reinforce the correct response for the normal reader, and impaired knowledge of meaning in semantic dementia may thus be responsible for the common regularization errors (PINT read Volume 6,Number 1
aloud as if it rhymed with “mint”) observed when such patients read irregular words. If the pronunciation of a regular word like MINT must be based on unlinked elements, it will still have a very good chance of being correct, because subword elements (like IN and INT) have pronunciations congruent with those of the whole regular word MINT. For an exception word like PINT, by contrast, translation from orthography to phonology on the basis of these subword elements will tend to result in a regularization error. Furthermore, data from a surface alexic patient studied by Hillis and Caramazza (1991) plus some of our own most recent data (Graham, Hodges, & Patterson, 1993) demonstrate a significant mrd-qec$c concordance between exception words that the patients fail to comprehend and those on which they err in reading aloud, providing additional evidence that the intactness of meaning has direct implications for the integrity of phonological word forms. We close this discussion with a few additional comments on the brain regions affected in semantic dementia. All three patients reported here, in common with other cases of progressive loss of semantic memory (see Hodges et al., 1992), show selective and often asymmetric temporal-lobe abnormalities. This is apparent on structural, and particularly on functional (PET or SPECT), brain imaging. When changes are seen on CT or MRI, they invariably involve the left temporal neocortex; in some cases (such as FM), the left has been selectively involved, while in other patients (such as JL) both temporal lobes are affected. The pathological process shows a predilection for the temporal pole and the inferior, middle, and superior gyri. It is interesting to note that in the PET results for patient PP, the posterior portion of the superior temporal gyrus (i.e., Wernicke’s area) appeared to be spared. These findings are in keeping with reports of selective impairment of semantic memory in the context of nonprogresive brain damage, for instance following herpes simplex virus encephalitis or traumatic brain injury (e.g., Pietrini, Nertempi, Vaglia, Revello, Pinna, & Ferro-Milone, 1988; Warrington & Shallice, 1984). Thus, there is converging evidence implicating left-temporal neocortical areas as the region most critically involved in the maintenance of semantic memory. The pathological basis of semantic dementia remains uncertain, but we have grounds for believing that it is unlikely to be Alzheimer’s disease; many of the clinical, neuropsychological, and radiological findings point to a diagnosis of Picks disease (for fuller discussion of this aspect, see Hodges, 1993 and Hodges et a]., 1992).
comprehension. A simple summary of extensive assessment is that PP was severely impaired on any task requiring declarative conceptual knowledge of objects, people or words. She never, in any test, correctly produced the name of an object, either when asked to name real objects or pictures, or in category fluency tests. She could not categorize, at a level significantly above chance, line drawings of real objects versus chimeric compositions of two different real objects (e.g., she accepted as real objects the body of a deer with the head of a lion, and the blades of scissors on the handle of a screwdriver); she could not match a picture of an object to its characteristic sound; she was consistently at or near chance in matching pictures to their spoken names; she could not recognize photographs of famous people or even of her own family. None of these failures could be attributed to low-level perceptual problems, since she attained entirely normal performance for her age on the Block Design subtest of the WAIS and in matching objects photographed from different angles; nor were her impairments attributable to generalized intellectual deficit since her score on a nonverbal test of intelligence (Raven’s Colored Matrices) was above average for her age. Although already severely impaired at initial assessment in July 1990, PP’s performance on tests of semantic memory has shown a dramatic decline over 30 months of subsequent investigation. For example, in 1990, while unable to sort pictures of animals and objects into specific categories referring to their attributes (e.g., native/ foreign for animals, electricaYnonelectrica1 household items for manmade objects), she achieved perfect performance in sorting the same pictures into living versus manmade things. She maintained a score of 100% on this test over three test sessions between August 1990 and July 1991; in the subsequent year, however, her performance gradually declined to 47% (chance). Of particular relevance to the list recall paradigm used in this study, one of PP’s best preserved cognitive skills was short-term auditory-verbal memory, at least as measured by digit span. When first assessed in 1990, she had a digit span of 9 forward and 6 backward (giving a WAIS age-scaled score of 17), and even in June 1992 her forward digit span was measured at 8. A positron emission tomography (PET) study performed in 1991 at the Hammersmith Hospital, London, indicated marked reduction of cerebral metabolism confined to left temporal and adjacent parietal regions. The stereotactically obtained quantitative data for oxygen consumption (as a measure of cerebral metabolic rate) in selected brain regions are shown in Table 6.
METHODS PP: Case Report
PP: Experimental Method
PP, a right-handed woman who had worked as a clerical
PP’s performance in object naming (zero), action naming (zero), and word-picture matching (at or near chance)
officer and secretary, presented in 1990 (aged 68) with a 3-year history of progressive loss of vocabulary and
suggests that she no longer knows most specific referential words (nouns, verbs, and adjectives). At the beginPattenon eta].
G5
Table 6. Results of PET Study Showing the Absolute Values for Regional Oxygen Consumption (mu100 mumin) in PP and (in parentheses) These values as a Percentage of the Mean in Six Normal Control Subjects (Mean Age = 63, SD = 6) Right
L??
Inferior frontal gyrus Brodmann area 45
2.48
(83%)
2.90
(95%)
Brodmann area 46
2.66
(89%)
2.83
(94%)
Inferior, Brodmann area 20
1.67
(56%)”
1.89
(64%1
Middle, Brodmann area 21
1.85
(57%)“
2.53
(82%)
Superior, Brodmann area 22
2.39
(74%)
3.07
(92%)
Posterior, Brodmann area 37
1.89
(56%1“
2.69
(77%)
Thalamus
3.12
(96%1
3.94
(121%)
Angular gyrus
2.63
(79%1
2.78
(84%1
Inferior parietal lobe
2.13
(70%)“
2.76
(89%)
Temporal lobe
“Greater than 3 SD below control mean.
ning of this study (spring 1991), however, there were a small number of such words that she used appropriately in conversation or picture-description tasks (e.g., nouns such as m o t h , sister, daughtq tonight, water; verbs such as washing, staying, read; adjectives such as little, south, worried).These were designated known words. We also counted as kmown words some items produced by PP in a word-fluency task. Although unable to produce even a single item in semantic-categoryfluency (responding to the instruction: “Tell me the names of some animals” with “I wish I could remember what an animal was”), she did manage to produce a few words in response to letter-fluency instructions (e.g., “tell me some words beginning with the letter S”). Although we have no direct evidence of PP’s comprehension of these words in a specifically semantic task, the fact that she produced only meaningful words and no neologisms in letter fluency, and that the words produced were (like the known words culled from her spontaneous speech) high-familiarity nouns (e.g., m@er) and adjectives (e.g., silly),suggests that these words still have some integrity within her language system. The unknown words for the experiment were selected from items on which PP had either ( 1 ) selected the wrong alternative when asked to point to a picture corresponding to a spoken word, or (2) given the wrong reply in response to a simple semantic question about a spoken word (e.g., “does a sparrow have wings?”). In May 1991, we managed to identify a total of 36 content words (nouns, verbs, and adjectives) that could be classified as kmown. These 36 words were concatenated, in different groups and orders, to form 18 lists of unrelated three-word sequences (e.g. feet, hughter, washing and thinking, read, children) and 18 lists of unrelated four-word sequences (e.g., easy, south, daugh66
Journal of Cognitive Neuroscience
ter, read). We then created lists of three- and four-word sequences of unknown words, matched to the hown word sequences on syllable length but not frequency. Given that PP’s vocabulary was restricted to a small number of very familiar words, it was not possible to match the two sets on word frequency/familiarity. However, all of the unknown words are sufficiently common that they were certainly known to PP premorbidly (e.g., for the known sequence thinking, read, children, the matching unknown sequence was chicken, sword, daisy). Each sequence was read aloud to PP at a rate of one word/sec, and she was asked to repeat it immediately. The entire set of four-item sequences was given first, then the set of three-word sequences; within each sequence length, known and unknown sequences were given in alternation. The identical test was administered to PP on two further occasions, in November 1991 and March 1992, without a reassessment of whether it was appropriate to consider the so-called known words as still within her language competence. In fact, given the dramatic deterioration over this 10-month period in PP’s spontaneous speech and in other abilities reflecting semantic memory, it is almost certain that many of the words originally assigned to the known set no longer belonged in that category. Therefore, when the recall task was administered for a fourth time, in June 1992, new sequences (now strings of three words only) were created on the basis of a record of PP’s spontaneous speech from her most recent test session. By this stage in the progression of her semantic dementia, PP’s speech consisted mainly of repetitive, stereotyped phrases (“I don’t understand at all,” “I’m terrible today”). We therefore had at our disposal an even smaller number of content words used appropriately in nonstereotyped productions, though Volume 6, Number 1
there were a few (e.g., nouns: house, p k e , time; verbs: coming, doing, remember;adjectives, adverbs: long, now, very). To make the number of known words up to 36, we were forced to use some of the function words that still appeared appropriately in PP’s speech (as in her question, asking about her husband, “Will he be coming soon?”). Unknown words for this fourth session were once again failed items from word-picture matching tasks, and matched to the known words for syllable length as before. At the end of this session, PP was also asked to repeat all 72 words (36 known and 36 unknown) presented as single items.
FM: Case Report Fh4, a left-handed woman working as a care-assistant in an old people’s home, presented in 1991 (age 54) with a 1-year history of increasing anomia for things and people. Her performance on tests of visuospatial ability, nonverbal reasoning, and memory was within normal limits, as was her digit span (6 forward, 5 backward). Her ability to match spoken names of objects to corresponding pictures was only mildly impaired. Anomia was noticeable in her otherwise fluent spontaneous speech, and confirmed by dramatically poor performance on tests of confrontation naming, naming to description and category fluency. She was above chance but significantly impaired, relative to normal subjects, in object decision (judging real vs. chimeric objects). Having apparently been a keen reader and an excellent speller premorbidly, she has developed prominent surface alexia and agraphia. This picture has remained reasonably static over 20 months of investigation, with some further decline in tasks such as naming, but no marked deterioration of comprehension (see Hodges et al., 1992, for further details). A SPECT scan demonstrated reduction in perfusion to left temporal regions, and an MRI scan with coronal and temporal-lobe oriented horizontal cuts showed striking focal atrophy of the left temporal neocortex, involving the superior, middle, and inferior temporal gyrus but with preservation of the hippocampal complex (see Fig. 1).
FM: Experimental Method Although FM has a severe anomia, she often succeeds when asked to perform other sorts of tests requiring knowledge of the objects that she fails to name. The difficulty in devising lists of known and unknown words for FM, therefore, is precisely the opposite of that with PP: it is words that can confidently be considered unknown that are scarce for FM. The following sources of u h o w n words were used: (1) items failed on at least two separate occasions in word-picture matching tests; (2) words poorly defined (e.g., FM’s responses when asked to define a lion-“An animal, quite tall, can’t think
Figure 1. T1-weightedMRI scan image in patient FM:a coronal slice through the anterior part of the temporal lobes demonstratingmarked focal atrophy of the left temporal neocortex. Note that left-sided structures are shown, by convention, on the right side of the image.
of anything else” or an eagle-“A bird, it stands up, has big eyes, does things at night”-suggest impoverished/ inaccurate knowledge of lions and eagles); (3) items from the semantic feature questions section of our semantic battery (where the patient is asked eight simple questions about an object, e.g., “Does a lion have a mane?” “Does a lion have stripes?”) for which FM answered more than half of the questions incorrectly; and ( 4 ) words that FM herself indicated, in the course of testing, that she did not comprehend (e.g., one of the semantic-feature questions about a helicopter is “Does a helicopter have a propeller?” to which FA4 replied “What’s a propeller?” Therefore, even if her responses over the eight questions about a helicopter did not place it in the unknown category,pmpeller was considered an unknown word for purposes of word-list recall. In October 1991, we were able to identify 36 words from the sources described above that could be assigned to the unknown category for FM, and then chose 36 known words (matched to the unknown items for syllable length) that she produced correctly when asked to name the Snodgrass and Vanderwart (1980) pictures. All Pattemnetal.
67
72 words were object names. Eighteen four-word se-
quences were created from each set, each specific word occurred twice, once in the first nine sequences and once in the second, embedded in a different sequence on each of the two occasions. As for PP, the sequences were read aloud to FM, with kmwn and unknown sequences alternating, and she was asked to repeat each sequence immediately after hearing it.
JL: case Report JL,a right-handed 60-year-old company director, presented in 1991 with a 1-year history of decline in both expressive and receptive vocabulary for names of places, people, and even common words such as food names. His day-to-day and autobiographical memory were relatively well preserved, and he had a normal digit span (7 forward, 5 backward). He performed well on all tests of visuospatial ability. Speech was fluent with normal articulation and syntax but some word-finding difficulties. Naming was very impaired, as was the ability to match spoken names to their pictorial referents. His performance on a test of object decision, though above chance, was significantly below normal (see Hodges et al., 1992, for further details). A functional (SPECT) brain scan revealed bilateral temporal lobe hypoperfusion, and MR imaging showed strikingly focal atrophy of both temporal lobes involving the superior, middle, and inferior gyrus but with relative sparing of the hippocampal complex. Over the 20 months since initial testing,JL has shown a steady and in some instances marked deterioration in various cognitive abilities. He has also become agitated and easily distractable.
JL:Experimentat Method Although JL is the most difficult patient of the three to test, because restlessness and distractability make him a rather uncooperative subject, he was the easiest of the three for whom to select materials for this experiment; his performance on tests from the semantic battery provides a much more adequate balance between h u m and u h w n words. Following similar techniques to those described for FM,we culled 60 known items from JL’s successes in picture naming, and 60 unkrnoum items from his failures in word-picture matching, word definitions, and semantic feature questions. All 120 words were object names. Thirty four-word sequences from each set were constructed in which (1) words from the two sets were matched for syllable length, (2) h w n and unkrnourlz sequences were presented in alternation, and (3) each word occurred twice, once in the first half of the testing order and once in the second half, in the context of a different list of target words on the two occasions. The testing procedure was identical to that employed with PP and FM. 68
Journal of Cognitive Neuroscience
Acknowledgments We greatly appreciate the assistance of Richard Wise and Guy Sawle (MRC Cyclotron Unit, Hammersmith Hospital, London) in performing and analyzing the PET study with case PP. We are also grateful to Sally Andrews, Alan Baddeley, Marlene Behrmann, Vicki Fromkin, and Becky Treiman for thoughtful discussions and comments on the ideas presented here. Reprint requests should be sent to Dr. K. E. Patterson, MRC Applied Psychology Unit, 15 Chaucer Road, Cambridge CB2 2EF, U.K.
REFERENCES Baddeley, A. D. (1966). Short-term memory for word sequences as a function of acoustic, semantic and formal similarity. Quurter!y Journal of ,%phmual Psydwlogy, 18, 362-365.
Brady, S., Shankweiler, D., & Mann, V. (1983). Speech perception and memory coding in relation to reading ability. Journal of , T @ e r i M l Child Psychology, 35, 345-367. Breedin, S. D., & S h , E. M. (1993). S e n t e n c e m n g in the p r e s e n c e of semantic loss:A case study. Paper presented to the Sixth Annual CUNY Sentence Processing conference. Brener, R (1940). An experimental investigation of memory span.Journal of E3qberimental Psychology, 26, 467-482. Chertkow, H., Bub, D., & Seidenberg,M. S. (1989). Priming and semantic memory loss in Alzheimer’s disease. Brain andLmguage, 36,420-446. Clements, G. N., & Keyser, S. J. (1983). C V ~ o l o g yCam. bridge, MA: MIT Press. Francis, W. N., & Kucera, H. (1982). Frequency amlysk of Engl&b w g e . Boston: Houghton Mifflin. Fromkin, V. A. (1980). E m in linguisticpaformarue:Sl@ of the tongue, m, pen and h n d . N e w York:Academic
Press. Fudge, E. (1987). Branching structure within the syllable.
Journal of LinguMcs, 23, 359-377. Graham, K., Hodges,J. R, & Patterson, K. (1993). The relationship between comprehension and oral reading in fluent progressive aphasia. N m m l o e , in press. Hillis, A. E., & Caramazza, A. (1991). Mechanisms for accessing lexical representationsfor output Evidence Erom a category specific semantic deficit. Brain and Languuge, 40, 106-144. Hodges, J. R (1993). Pick’s disease. In A. Bums & R Levy (Eds.), Dementia.London: Chapman and Hall (in press). Hodges, J. R, Patterson, K., Oxbury, S., & Funnell, E. (1992). Semantic dementia: Fluent aphasia with temporal lobe atrophy. Brain, 115, 1783-1806. b e l t , W. J. M. (1992). Accessing words in speech production: Stages, processes and representations. Cognirion, 42, 1-22. MacKay, D. G. (1972). The structure of words and syllables: Evidence from errors in speech. Cognitfve Psychology, 3, 210-227. Martin, N., & Sa&an,E. M. (1990). Repetition and verbal STM in transcortical sensory aphasia: A case study. Brain and Language, 39, 254-288. McCanhy, R A, & Warrington, E. K. (1987). The double dissociation of short-termmemory for lists and sentences. Brain, 110, 1545-1563. McClelland, J. L., & Rumelhart, D. E. (1981). An interactive activation model of context effects in letter perception: Pan 1. An account of basic findings. Psychologicul Review, 88, 375-407. McLeod, P., & Posner, M. I. (1984). Privileged loops from perVolume 6, Number 1
cept to act. In H. Bouma & D. G. Bouwhuis (Eds.), Azfention & p e @ i n c e X.London: Erlbaum. Patterson, K., & Hodges, J. R. (1992). Deterioration of word meaning: Implications for reading. Neuropsychologia, 30, 1025-1040. Patterson, K., & Marcel, k (1992). Phonological ALEXIA or PHONOLOGICAL alexia? In J. Aegria, D. Holender, J. Junca de Morais, & M. Radeau (Eds.), Analytic approaches to hu?nun cognition.Amsterdam: Elsevier. Pietrini, V., Nertempi, P., Vaglia, k , Revello, M. G., Pinna, V., & Ferro-Milone, F. (1988). Recovery from herpes simplex encephalitis: Selective impairment of specific semantic categories with neuroradiological correlation. Journal of Neumlogy, Neurosuqe?y,and Psycbiahy, 51, 1284-1293. Plaut, D.,& Shallice, T. (1993). Deep dyslexia: A case study of connectionist neuropsychology. CognitiwNeuropsychologv, 10,377-500. Poeck, K., & Luzzatti, C. (1988). Slowly progressive aphasia in three patients: The problem of accompanying neuropsychological deficit. Brain, 111, 151-168. SafFran, E. M., & Schwartz, M. F. (1993). Of cabbages and things: Semantic memory from a neuropsychological perspective-a tutorial review. In C. Umilta & M. Moscovitch (Eds.), Attention and Perfonane XV (in press). Schwartz, M. F., & Chawluk, J. B. (1990). Deterioration of h-
guage in progressive aphasia: A case study. In M. F. Schwartz (Ed.), Modular deficrts in Alzbeimer--typedementia. Cambridge, MA: MIT Press. Snodgrass,J. G., & Vanderwart, M. (1980). A standardized set of 260 pictures: Norms for name agreement, image agreement, familiarity, and visual complexity.Journal o f E p n mental Psychology:Human Learning and Memory, 6 174215. Snowden,J. S., Goulding, P. J., & Neary, D. (1989). Semantic dementia: A form of circumscribed cerebral atrophy. Behavioural Neurology, 2, 167-182. Taft, M. (1991). Reading and the mental I d o n . Hove and London: Erlbaum. Treiman, R,& Danis, C. (1988). Short-term memory errors for spoken syllables are affected by the linguistic structure of the syllables.Journal of E3qberimental Psychology: Learning, Memory and Cognition, 14, 145-152. Van Orden, G. C., Pennington, B. F.,& Stone, G. 0. (1990). Word identification in reading and the promise of subsymbolic psycholinguistics. Psychological R&, 97, 488-522. Warrington, E. K. (1975). The selective impairment of semantic memory. QuarterlyJournal of E . % $ e r i ~ tPsychology, al 27, 635-657. Warrington, E. K., & Shallice, T. (1984). Category specific semantic impairments. Brain, 107, 829-854.
Pmemnetd.
69