The Origins of Formal Paraphasias in Aphasics' Picture Naming

The Origins of Formal Paraphasias in Aphasics' Picture Naming

59, 450–472 (1997) BL971792 BRAIN AND LANGUAGE ARTICLE NO. The Origins of Formal Paraphasias in Aphasics’ Picture Naming Deborah A. Gagnon,* Myrna F...

230KB Sizes 0 Downloads 128 Views

59, 450–472 (1997) BL971792

BRAIN AND LANGUAGE ARTICLE NO.

The Origins of Formal Paraphasias in Aphasics’ Picture Naming Deborah A. Gagnon,* Myrna F. Schwartz,*,† Nadine Martin,† Gary S. Dell,‡ and Eleanor M. Saffran† *Moss Rehabilitation Research Institute; †Temple University; and ‡University of Illinois Accounts of spoken word production differ on whether aphasics’ formal paraphasias derive solely from segmental distortion or whether some derive instead from whole word substitution. Form-related paraphasias produced by nine aphasics during picture naming were examined for evidence of lexical effects (word, frequency, and grammatical class biases) and for the manner in which target phonemes and word shape were preserved. Preservation patterns were consistent with previous descriptions of aphasic and nonaphasic form-related speech errors. Evidence for word and frequency biases was found, as well as a grammatical class bias sensitive to the degree of target–response segmental overlap. In conjunction, the results indicate that formal paraphasias arise, at least in part, via word substitution. The findings are supportive of interactive models with phonological-to-lemma feedback and/or modular models with a grammatically organized lexeme level.  1997 Academic Press

Speech errors produced by nonaphasic speakers have traditionally served as the data upon which theories of speech production have been built (e.g., Dell, 1986; Fay & Cutler, 1977; Fromkin, 1971; Garrett, 1975; Levelt, 1989; Shattuck-Hufnagel, 1979; Stemberger, 1985), but increasingly, aphasic speech errors are being relied upon to inform production theories as well (e.g., Buckingham, 1992; Butterworth, 1992; Dell, Schwartz, Martin, Saffran & Gagnon, in press; Ellis, 1985; Ellis, Miller & Sin, 1983; Garrett, 1992; Kohn & Smith, 1994; Schwartz, 1987). The present investigation continues the latter tradition by focusing on aphasic speech errors that are phonologically related to their targets and using the characteristics of this kind of error to evaluate contrasting theories of spoken word production. The authors gratefully acknowledge Adelyn Brecher, Rita Grewal, and April Roach for their roles in developing the PNT and in data collection; Michael Montgomery for statistical consultation; Michel Salas for research assistance; and anonymous reviewers for providing constructive commentary. This research was supported by NIDCD Grant DC00191. Address correspondence and reprint requests to Deborah A. Gagnon, Division of Social Sciences, One University Place, Widener University, Chester, PA 19013. E-mail: Deborah. [email protected]. 450 0093-934X/97 $25.00

Copyright  1997 by Academic Press All rights of reproduction in any form reserved.

FORMAL PARAPHASIAS

451

Aphasic speech is sometimes marred by the production of formal paraphasias: word utterances that are phonologically similar to their intended targets (e.g., squirrel → school). Such errors are usually not prevalent, however, and are by no means the only type that accompany these deficits: words related to the target in meaning (semantic paraphasias), nonwords related to the target in sound (target-related neologisms), and other types of speech errors may occur as well. Each of these can also be observed, on occasion, in nonaphasic ‘‘slips of the tongue.’’ In nonaphasics, the occurrence of malapropisms (analogs to aphasics’ formal paraphasias) together with semantic errors have motivated a widely accepted two-stage view of word production (Kempen & Huijbers, 1983; Levelt, 1989). The first stage involves retrieval of a word’s semantic–syntactic specification, or lemma; semantic errors are thought to arise from a failure to select the lemma corresponding to the target word. At the second stage, the word’s phonological specification, or lexeme, is retrieved; malapropisms are thought to arise at this stage when the incorrect lexeme is selected. Contemporary production theories maintain the fundamental aspects of the two-stage view, but vary in the details involved. For instance, in an interactive-activation variant of the two-stage model proposed by Dell (1986; 1988), semantic errors and malapropisms may arise at lemma selection. Dell’s model represents lexical information in a hierarchical network of semantic, lemma, and phoneme nodes, with bidirectional connections between levels. Semantic nodes activate lemma nodes, which in turn activate phoneme nodes. At lemma retrieval, the most highly activated lemma node is selected; at phoneme retrieval, the most highly activated phoneme nodes are selected. The critical feature of the model, however, is that nodes at different levels of the network are connected via bidirectional links. Thus, competing for selection at the lemma level are nodes that share semantic features with the target as well as nodes that share its phonemic constituents. The activation of semantic competitors results from the top-down spread of activation from semantic features to all the nontarget lemma nodes that instantiate those features. The activation of phonological competitors results from the bottomup feedback of target phonemes to all the nontarget lemma nodes that are connected to them. Semantic errors arise when a semantically related lemma competitor is selected; malapropisms arise when a phonologically related lemma competitor is selected. In aphasics, the status of formal paraphasias (hereafter, formals) as malapropisms has been called into question because of their rarity and because of their tendency to co-occur with target-related neologisms (hereafter, neologisms). Neologisms are thought to arise subsequent to lemma and lexeme retrieval, at the stage of phonological encoding (Buckingham, 1981; Kohn & Smith, 1994; Shattuck-Hufnagel, 1987). Distortions at this stage may result in an incorrect utterance that is, nonetheless, phonologically related to the intended target word (e.g., octopus → saktapuf). Because formals occur in

452

GAGNON ET AL.

the same individuals who make neologisms, and because the incidence of the former is usually low relative to the latter, it has been argued that all formals are actually neologisms that happen by chance to create real words (i.e., jargon homophones; cf. Buckingham, 1980, 1981; Butterworth, 1979; Ellis, Miller & Sin, 1983; Lecours, 1982; Lecours, Deloche & Lhermitte, 1973; Lecours & Rouillon, 1976; Nickels & Howard, 1995). The incidence of formals relative to neologisms is expected to be low from such a perspective because the number of phonologically permissible sequences of phonemes that make real words is low relative to those that make nonwords. Schwartz, Saffran, Bloch, and Dell (1994) reported on a jargon aphasic whose conversational speech errors exemplified this low formal-to-neologism speech pattern. Invoking Dell’s interactive-activation model, Schwartz et al. proposed a pathological lowering of connection strength in the lexical network to explain their subject’s speech pattern (see also Ellis, 1985). Decreased connection strength results in a lower rate of activation spread, and hence, less time for activation to feed back from phoneme to lemma nodes prior to lemma selection. This has two related consequences. First, it prevents the formation of top-down, bottom-up feedback loops that tend to strengthen correct phonemes and weaken alternative ones. Second, it reduces the likelihood that phonological competitors to targeted lemmas will become activated. The joint effect is to promote a high incidence of neologisms and a low incidence of formals. Another way to explain the typically low incidence of formals is to invoke a control process or ‘‘editor’’ that monitors for phonemic errors and suppresses them prior to articulation. Such an editing mechanism has been used to explain so-called lexical biases in the spontaneous and elicited speech errors collected from nonaphasic speakers (e.g., Baars, Motley & MacKay, 1975; Butterworth, 1989; Levelt, 1989). For instance, using a phonological priming error elicitation task, Baars et al. (1975) found that normals’ slips were three times more likely to result in words than nonwords. They explained this apparent lexical bias as resulting from a postlexical editor that is more likely to detect phonemic errors when the outcome of the error is a nonword than when it is an existing word. If this editing mechanism were disrupted in aphasia and more nonword errors were allowed to get through, the ratio of neologisms to formals would be higher than normal, just as Schwartz et al. had found. In contrast to the Schwartz et al. finding and to the general impression that formals are a relatively rare occurrence, Martin and Saffran (1992; Martin, Dell, Saffran & Schwartz, 1994), Blanken (1990), and Best (1996) have reported cases of fluent aphasia in which confrontation naming was characterized by a high proportion of formals. Invoking the same interactive-activation model as Schwartz et al., Martin and Saffran explained this phenomenon as resulting from a faster than normal rate of activation decay: Activated lemma nodes pass activation to constituent phonemes but then too quickly

FORMAL PARAPHASIAS

453

decay back to a resting level of activation. The result is to alter the distribution of activation in the network relative to the normal state: Nodes activated early, including the lemma node that corresponds to the target, will be disadvantaged relative to nodes activated later. The latter includes phonologically related lexical items, which are activated by feedback from the phoneme level. This allows an alternative, yet phonologically related, lexical item to become more highly activated and be selected for production instead. Thus, Martin and Saffran offer an explanation for how formals may arise as a result of substitution at the lemma level of an interactive model. This account was further substantiated in a simulation study of Martin and Saffran’s subject (Martin et al., 1994). The cases reported by Best, Blanken, and Martin and colleagues suggest that not all formals produced by aphasics are simply neologisms that happen by chance to create real words. For one thing, the rate of formals relative to neologisms produced in naming tasks was too high for all three of the subjects studied, ranging from 11 to 23% for formals versus 4 to 20% for neologisms (cf. Best, Table 4). Moreover, Blanken and Martin et al. report that the grammatical class of target words tended to be preserved in the formals produced by their subjects, which should not be the case if these errors were created at phonological encoding. Differences in the nature of segmental (Best; Martin et al.) and word shape (Martin et al.) preservation between formals and neologisms also suggest a dual mechanism. Still, the question of the origin of formals in aphasic speech continues to be a matter of debate. Recently, Nickels and Howard (1995) failed to find evidence for a lexical bias in the form-related paraphasias produced by 15 aphasic subjects in a naming task and further, argued that the errors failed to support the predictions of Dell’s interactive model (Dell & O’Seaghdha, 1991). These authors conclude that their data favor instead a chance outcome explanation for the creation of formals, and that the few reported instances of lexical bias arise from the intactness of a production editor in these subjects. We address the controversy concerning the origins of formals by analyzing a corpus of aphasic form-related errors produced in a naming study. Three questions will be asked in these analyses: (1) Does the incidence of formals in the error corpus support a true lexical influence, or is the incidence within the range that could be expected if formals derived from phonemic distortions at phonological encoding alone? (2) Do the formals behave like lexical substitutions? If so, we would expect that they would preserve the grammatical class of the target (in this case, noun) and that response word frequency would be a factor in their generation. (3) How are target phonemes and word shape preserved in formals and neologisms? Characterizing the nature of segmental and word shape preservation in form-related errors has been a focus of related studies (e.g., Best, 1996; Blanken, 1990; Fay & Cutler, 1977; Martin et al., 1994; Shattuck-Hufnagel, 1979; Valdois, Joanette & Nespoulous, 1989); the present set of errors will also be examined along these di-

454

GAGNON ET AL.

mensions to ensure that they exemplify those upon which previous studies have been based. The questions in 1 and 2 will be addressed by comparing the corpus of errors to a pseudo-corpus that simulates distortion at phonological encoding. Such distortion includes errors of phoneme addition, deletion, transposition, and substitution. Of these, substitution is by far the most common form of distortion (Blumstein, 1973; Lecours, Deloche & Lhermitte, 1973; Miller & Ellis, 1987). Thus, the errors in our pseudo-corpus have been generated solely through the mechanism of phoneme substitution. This allows us to pit phoneme substitution against lexical substitution as a means of generating formals. To evaluate these distinct mechanisms, we enumerate a set of predictions associated with each mechanism and then evaluate these predictions against the error corpus. Two hypotheses are entertained: the single mechanism hypothesis—formals are generated only via phoneme substitution— and the dual mechanism hypothesis—in addition to phoneme substitution, at least some formals are generated via lexical substitution. PREDICTIONS

Single Mechanism Hypothesis: Formals Are Generated Solely from Phoneme Substitutions All models allow for phoneme substitution as a mechanism for generating both neologisms and formals. If formals are generated only via this mechanism, the following characteristics should be observed: A. The incidence of formals should be no greater than chance, that is, no greater than what would be expected given the proportion of real words in the set of all phonologically permissible sequences of phonemes of a given length. B. The average word frequency of formals of a given length should not be statistically different from the average frequency of the set of all lexical items of the same word length. Assuming that phoneme substitutions arise subsequent to lexeme retrieval, there is no means for a frequency bias to arise. C. There should be no systematic preservation of targets’ grammatical class (i.e., noun) in formals. If a phoneme is substituted for in a word and the outcome is also a word, there is only a chance probability that the two items will share the same grammatical class. D. There should be no relationship between the incidence of formals and the incidence of formals that are nouns. There is no means via phoneme substitution for the production of noun outcomes to become correlated with the production of word outcomes, both being chance occurrences. E. The phonological and structural relationship between targets and their form-related errors should be the same for formals and neologisms. Specifically, formals and neologisms should preserve target word shape to the same

FORMAL PARAPHASIAS

455

degree and show the same pattern of segment preservation if they both arise via the same mechanism. Dual Mechanism Hypothesis: Formals May Derive from either Phoneme Substitution or Lexical Substitution Mechanisms All two-step models of normal word retrieval allow for form-based word substitutions (malapropisms). In models without a feedback mechanism, these may arise at lexeme retrieval (e.g., Fay & Cutler, 1977; Fromkin, 1971; Levelt, 1989); in models with a feedback mechanism, they arise at lemma selection, as a consequence of phoneme-to-lemma connections (e.g., Dell, 1986; Dell & O’Seaghdha, 1992; Harley, 1984; Stemberger, 1985). At issue is whether this mechanism is also responsible for at least some of the formals produced by aphasics. If so, these predictions follow: A. Formals should occur at a greater-than-chance frequency. Note that the appropriate comparison is between actual incidence and chance incidence of formals, not between incidence of formals and incidence of neologisms. Lexical bias does not necessarily imply that more formals than neologisms are expected. It is (at present) impossible to predict how many formals will be generated via a lexical substitution mechanism relative to a phoneme substitution mechanism. A greater incidence of formals produced than expected by chance implies that at least some of these were generated via lexical substitution but says nothing about how many were generated that way. B. Word frequency is known to be an important factor in lexical retrieval. For instance, high frequency words are less susceptible to errors in production than low frequency words (Dell, 1988; Stemberger, 1984; Stemberger & MacWhinney, 1986). On the assumption that high frequency words are more likely to be selected as target substitutes than low frequency words, it is predicted that the average word frequency of formals should be higher than what is expected by chance. Note again that the appropriate comparison is between actual response frequency and chance response frequency. Some prior studies (e.g., Nickels & Howard, 1995; Martin et al., 1994) have tested a different prediction: The average frequency of error responses will be higher than that of their targets. Although this comparison of response frequency and target frequency is an obvious one to make, it is actually quite difficult to interpret. Such a comparison is a complex reflection of any main effect of target frequency (Are low frequency targets more likely to have an error?), any main effect of response frequency (Are high frequency responses more likely to be emitted as errors?), and any target/response frequency interactions. For example, a main effect of target frequency (most errors occurring on low frequency target items) coupled with the absence of a response frequency effect (no effect of frequency within the error words produced) could lead to error responses being, on average, of higher frequency than targets. Thus, one could find that responses are more common

456

GAGNON ET AL.

than targets even though response frequency is not causally involved in the error. To avoid this mistake, a better comparison is between response frequency and chance response frequency. C. Malapropisms overwhelmingly preserve the grammatical class of their targets (Fay & Cutler, 1977). Insofar as the corpus of formals contains ‘‘true’’ malapropisms—lemma or lexeme substitutions—the likelihood is greater than chance that these errors will be nouns—the grammatical class of picture naming responses. As was the case for lexical bias (and for basically the same reason), the appropriate comparison to make in assessing a grammatical class bias is between actual incidence of noun production and chance incidence of noun production, rather than incidence of noun production and incidence of nonnoun production. D. A high correlation between the incidence of formals and the incidence of formals that are nouns would implicate a mechanism, such as lexical substitution, that simultaneously biases for word and noun outcomes. E. Malapropisms tend to deviate more from the target’s phonology and word shape than form-related errors derived from phoneme substitutions (Martin et al., 1994). Thus, if malapropisms contribute to the corpus of formals, we would expect, on average, less word shape and target segment overlap with the target than we find in the neologisms of the naming error corpus. A final note: The assumption underlying the dual mechanism hypothesis—that lexical substitution is responsible for the presence of lexical effects—is not unassailable. Earlier we introduced the concept of a postlexical, prearticulatory editor that monitors for phonemic errors and that is sensitive to the lexical status of errors (i.e., word vs nonword). It is possible to expand the sensitivity of such an editor to make it responsive to other lexical dimensions as well, in which case some of the effects we take to reflect lexical processing could be ascribed to the editor instead. We will take up consideration of this alternative at a later point in the paper. METHODS To assess the above predictions, form-related paraphasias were collected from a group of fluent aphasics who produced them in the context of a confrontation picture naming task.

The Philadelphia Naming Test The Philadelphia Naming Test (PNT) is a computerized confrontation picture naming test consisting of 175 line-drawn objects. The pictures used in the PNT have been digitized and are presented using MacLaboratory experiment-running software (Chute, 1990). The target names for these items consist of one, two, three, and four syllable words, with frequency counts ranging from 1 to 2110 occurrences per million tokens of printed text (see Table 1).1 1 The frequency counts correspond to the singular or mass noun (NN) entry for each test item in the Francis and Kucˇera (1982) frequency counts of American English words.

457

FORMAL PARAPHASIAS

TABLE 1 Distribution of Philadelphia Naming Test Target Labels Word length (No. of syllables)

Word frequency (per million tokens)

1

2

.2

Low (1–25) Medium (26–79) High (80–2110)

44 32 24

37 8 7

20 3 0

The items in the test were normed using 30 nonaphasic control subjects. Each of the items was responded to with the expected target label by at least 85% of these control subjects; 136 of the 175 items yielded the same target label in all 30 control subjects. Across control subjects, each providing 175 responses, only nine responses were form-related errors (seven words, two nonwords). Additional details about the PNT may be found in Roach, Schwartz, Martin, Brecher, and Grewal (1996).

Subjects Twenty-three aphasic subjects have participated in the PNT study. These individuals were selected for inclusion in the study based primarily on their language profile. All were diagnosed as having a fluent-type of aphasia, based on Boston Diagnostic Aphasia Examination criteria (Goodglass & Kaplan, 1983). We excluded individuals diagnosed with a nonfluent form of aphasia or a motor speech impairment (e.g., dysarthria, apraxia) to avoid attributing phonetic and/or articulatory distortions to a phonological origin. Most of the subjects were below 75 years of age and all retained sufficient receptive and expressive skills to participate in the testing. Of the 23 subjects, 9 produced form-related errors (as defined below) on at least 10% of the trials (18 targets). The form-related target–error pairs from these 9 subjects form the data upon which the analyses described below were conducted. Demographic and language profiles for these subjects are presented in Tables 2 and 3, respectively.

Administration All subjects except NC2 were administered the PNT in the same manner. These subjects were seated in front of a computer monitor on which the PNT drawings were displayed one at a time and were instructed to respond to each item with a single word label, as quickly as possible. After it appeared to the experimenter that the subject was finished with his or her attempt to name the object, or after about 30 sec from stimulus onset (whichever came first), the experimenter provided the correct target name and pushed a button to go on to the next trial. A 30-sec trial limit and feedback were included in the procedure in an attempt to prevent aphasic subjects from perseverating. The data from NC were drawn from an earlier version of the test that was not computerized; instead, NC was shown the line drawings of the objects on paper, one at a time. Also, the 175 PNT items were part of a larger set of items named

2 This is the same NC who was the focus of studies reported in Martin and Saffran (1992) and Martin, Dell, Saffran, and Schwartz (1994).

Sex

F F F

M

M M

F

M

M

Subject

GB HB LB

NC

JG LH

GL

VP

WR

R

R

R

R R

R

R R R

Handedness

HS1

HS1

HS

Ph.D. HS

HS1

HS HS BA

Education Left temporal occipital bleed Left parietal CVA Left MCA infarct small vessal ischemia lacunar infarcts Left CVA section cerebral aneurysm, mid to posterior superior temporal gyrus Left MCA occlusion Left MCA infarct, posterior, superior, and temporal lobes Left thalamic hemmorrhage s/p section meningioma Left CVA–anterior superior temporal gyrus/parietal Left MCA infarct

Etiology

TABLE 2 Subject Characteristics

61

64

45

76 47

28

83 65 77

Age at testing

8

28

28

55 68

75

40 61 82

% Correct

29

13

46

23 23

16

14 27 10

% Formrelated

Philadelphia Naming Test

458 GAGNON ET AL.

Language profile

Wernicke Conduction Anomic Conduction Anomic Conduction Wernicke Anomic Wernicke

Subject

GB HB LB NC JG LH GL VP WR

2 3 4 2 3 5 3 1 1

Severity rating 48 81 87 35 83 89 70 65 29

Aud. comp. (mean on four subtests)

35

81 90 69

52 88 92

Visual confront. naming 50 70 70 Not available 70 70 50 Not available 90

Word repetition

BDAE, percentiles of subtest summary profile

75

75 60 55

60 0 60

Phrase repetition

TABLE 3 Subjects’ Language Profiles

3

0

3

7

Neologisms

Not available 1 Not available 2 Not available 5

2 Not available

Literal

13

2

1

9

Verbal

0

0

5

Other

78 43 35 55 23 25 5

15

BNT % correct

No. paraphasias in the vis. conf. naming subtest FORMAL PARAPHASIAS

459

460

GAGNON ET AL.

by NC. However, only NC’s responses to the 175 PNT items were used in the analyses described below. The PNT was split into two blocks of test items for presentation. Ordering of the blocks was randomized across subjects; the order of trials within a block was constant. Testing usually took place within a single session but sometimes two sessions were needed. Sessions began with a set of 10 practice items, none of which appeared on the test itself. Each session was tape recorded for future transcription purposes; the experimenter also performed a rough online transcription while administering the task.

Scoring Following testing, subjects’ responses were phonetically transcribed by two speech pathologists, using loose International Phonetics Association transcription rules. Discrepancies in transcription were resolved by the two transcribers; another opinion was obtained if a particular discrepancy could not be resolved in this way. Next, the responses within a trial were labeled as first attempts, first complete attempts, and final attempts at naming. For the purpose of this study, a subject’s first complete response to each item was extracted. The first complete response was taken to be the subject’s first response to an item which minimally contained a consonant and non-schwa vowel, was followed by a noticeable pause and/or had clear downward or rising intonation. A response was deemed incomplete if it failed to meet the above acoustic criteria for completion, or if it was a monosyllabic response to a multisyllabic target (i.e., a three or four syllable target), unless the response was a one morpheme response to a compound target (e.g., cheer for cheerleaders). So, for example, the self-interrupted response /glæ-/ for glass would be deemed incomplete, but /glæ/ followed by a pause or change in intonation would be deemed complete. Interrater agreement for determining the first complete attempt at naming was good (average across six aphasic subjects was 93% for two raters). Once the first complete attempt for each trial was determined, it was coded with regard to the relationship it had to the correct target name (e.g., semantic, phonological, picture part, etc.). Interrater agreement for this component of the scoring procedure was also good (average across six aphasic subjects was 93% for two raters). Only responses coded as purely formrelated paraphasias (the response shared only a phonological relationship to the target) were used in the present analyses. That is, a response that named an object that was semantically or visually related to the target (e.g., horse → donkey; bicycle → spokes) was excluded, regardless of the degree of phonological relatedness it held to the target label. Phonological relatedness was determined by comparing each response to a transcription of the target item. Target transcriptions were guided by those provided in Kenyon and Knott (1953), taking into account the dialectical variation of the Philadelphia region. The subjects who participated in this study were either all life-long residents of the Philadelphia area or lived the vast majority of their lives there, so dialectical issues were minimal. One of the following criteria had to be met in order for a response and target pair to be considered phonologically related (all criteria exclude consideration of unstressed vowels): (1) overlap of first or last phonemes, (2) overlap of at least two phonemes in any syllable or word position, or (3) overlap of at least one phoneme in the same syllable and word positions, aligning words left to right. The majority of the errors we have coded as phonologically related actually exceed these criteria and bear a close phonological resemblance to their targets, e.g., close to half (42%) of these errors contain 50% or more of their target’s phonemes, in the same word position they occupied in the target (Gagnon & Schwartz, 1996). Finally, the formrelated errors were coded as to their lexical status: word or nonword. In order for an item to be coded as a word (a formal), it had to appear in a dictionary of the English language (Merriam-Webster’s Collegiate Dictionary, 1993, was used) and could not be a proper name. A response not meeting these criteria was coded as a nonword (a neologism).

FORMAL PARAPHASIAS

461

ANALYSES AND RESULTS

The approach in the analyses is to compare the characteristics of the formrelated errors to the characteristics predicted under the hypotheses above. Each of the predictions is assessed in turn: A. Incidence of Formals Relative to Chance To determine whether the overall proportion of formals differs from what would be expected given the distributional properties of the lexicon, the lexical density of a sub-section of phonological space was estimated. This estimation was carried out by generating all the possible consonant–vowel– consonant (CVC) sequences using a subset of consonant (/b/, /d/, /g/, /p/, /t/, /k/, /f/, /s/, /l/, /m/, & /n/) and vowel (/i/, //, /e/, /ε/, /{/, /ɑ/, /o/, / U /, /u/, & /ö/) phonemes of English. We chose to estimate density based on CVC sequences because these correspond to a domain of lexical space that is perhaps one of the most dense and would thus provide a relatively high estimate of lexical density. The particular phoneme segments chosen represent some of the most frequent phonemes in English (Blumstein, 1973; Carterette & Jones, 1974), also biasing toward a high estimate of lexical density. If the proportion of formals in the subjects’ corpus of formrelated errors is at least as high as that computed from this dense area of the lexicon, then it would be unlikely that the formals were all merely the outcome of a phoneme substitution mechanism. Generating all possible combinations of the consonants and vowels from the phoneme sets given above resulted in 1210 CVC sequences. Of these, 490 sequences (40%) were words and 730 (60%) did not correspond to real words (using the same criteria described above for determining the lexical status of subjects’ responses). Thus, lexical space, defined along phonological dimensions, is relatively sparse compared to the domain of all phonologically permissible possibilities: If a phoneme is randomly substituted for in a word (obeying phonotactic constraints), there is, at best, a 40% chance that the resulting outcome will also be a word. In short, if phoneme substitution is solely responsible for generating word outcomes, then word outcomes should occur no more than 40% of the time. The number (and proportion) of formals produced by each subject is provided in Table 4.3 Overall, formals accounted for 46% of subjects’ formrelated responses. To determine whether more formals were produced than 3 Variations among subjects will appear throughout the analyses discussed in this paper. The data, however, are not powerful enough to prove individual differences: Most subjects produced too few form-related paraphasias to make any strong claims about any one individual’s tendencies. Moreover, our goal is not to demonstrate individual differences among subjects but rather, to demonstrate a group effect. We simply aim to determine whether, among a group of aphasics who make form-related errors, there is or is not a tendency for these to reflect lexical influences.

462

GAGNON ET AL.

TABLE 4 Distribution of Subjects’ Form-Related Errors and of the CVC Corpus Formals Subject GB HB LB NC JG LH GL VP WR CVC Corpus

All 15 (.65) 22 (.46) 3 (.17) 12 (.48) 14 (.35) 12 (.33) 37 (.46) 19 (.83) 25 (.51) 490 (.40)

CVC 4 11 2 3 5 2 12 4 3

(.50) (1.00) (1.00) (.50) (.50) (.50) (.71) (1.00) (.60)

490 (.40)

Nouns CVC log frequency 2.81 2.17 1.81 3.44 2.99 3.08 2.65 2.87 3.67 1.87

All 13 17 0 9 7 9 26 16 21

(.87) (.77) (.00) (.75) (.50) (.75) (.70) (.84) (.84)

314 (.64)

1 phoneme deviates 4 8 0 5 1 3 3 1 2

.1 phoneme deviates

(1.00) (.67) (.00) (.83) (.20) (.50) (.50) (1.00) (1.00)

314 (.64)

9 9 0 4 6 6 23 15 19

(.82) (.90) (.00) (.67) (.67) (1.00) (.74) (.83) (.83)

314 (.64)

expected by chance, the number of CVC form-related errors produced by each subject was multiplied by .40 (the proportion of CVC responses expected to be words by chance). This chance-predicted number of formals was then subtracted from the subject’s actual number of CVC formals (cf. Table 4). Using a one-sample t test, the null hypothesis was rejected in favor of the one-tailed hypothesis that the average deviation (of actual rate of formal production from chance rate of formal production) was greater than zero (t(8) 5 2.86, p , .02).4 This lexical bias is not unique to CVC responses since the non-CVC subset of responses consisted of 41% formals. It is reasonable to assume that this also is a greater-than-chance proportion of formals, considering that this set of responses is composed of phoneme strings that correspond to much less densely populated areas of the lexicon (e.g., two, three, and four syllable sequences) than the CVC neighborhood upon which the 40% estimate was based. To summarize: Even with a generous estimate of lexical density, the proportion of formals in our corpus of form-related paraphasias exceeds it. This implicates a mechanism of some sort that has the ability to generate lexical bias. B. Relationship between Actual Response Frequency and Chance Response Frequency For this analysis, log frequencies were used that were based on collapsed homophone frequencies (sum of occurrences across all of the grammatical

4

An α level of .05 is used as the criterion for significance in all parametric tests reported.

FORMAL PARAPHASIAS

463

categories for which the spoken word form occurs in English).5 The average log frequency of the CVC formals for each of the nine subjects is provided in Table 4. The 490 lexical items in the CVC corpus described under Analysis A were found to have an average log frequency of 1.87. A comparison between this expected log frequency and subjects’ actual response log frequencies showed that the frequencies of subjects’ CVC formals were significantly greater than what would be expected from the chance estimate (t(8) 5 4.99, p , .0006), implicating a mechanism sensitive to the frequency of potential word outcomes. C. Incidence of Nouns Relative to Chance The counts and proportions of noun formals produced by each subject are given in Table 4. Since many words have a dual (or even multiple) grammatical interpretation (e.g., fly), we categorized a formal as a noun if the highest frequency entry for that item in the Francis and Kucˇera word counts corresponded to a noun category. Across subjects, 74% of the formals were nouns, using this criterion. This degree of noun production can be compared to the occurrence of noun entries in the CVC corpus. In that corpus, 64% of the 490 CVC words were nouns, using the same criterion used to establish grammatical class in subjects’ responses. This is also very close to a 60% noun estimate from a pseudo-corpus analysis reported by Martin et al. (1994). To test the hypothesis of no difference between actual number of nouns produced and number predicted by chance, the total number of formals produced by each subject was multiplied by .64 (the percentage of nouns in the CVC corpus). This predicted number of nouns was then subtracted from the subjects’ actual number of nouns produced. Subjects’ rate of noun production was found to be significantly greater than chance (t(8) 5 2.24, p , .03). Thus, there is evidence of a tendency for subjects to produce nouns at a greater-than-chance rate, implicating an influence on formals that cannot be explained via recourse to a phoneme substitution mechanism alone. A comparison of formals that involve one phoneme deviations from their targets to those that involve more-than-one phoneme deviations lends further support to this argument. On the assumption that formals involving a change in more than one phoneme are more likely to arise from lexical substitution and that one phoneme deviates are more likely to arise from phoneme substitution, it would be expected that a grammatical class effect would be weaker in the set of one phoneme deviate responses than in the set of greater-thanone phoneme deviate responses (Martin et al., 1994). This appears to be the case (cf. Table 4): Only 61% of subjects’ one phoneme deviate responses 5 Recent empirical evidence (Jescheniak & Levelt, 1994) suggests that collapsed homophone frequency (e.g., combining the frequency-of-occurrence for made and maid) is the appropriate measure on which to base production frequency analyses.

464

GAGNON ET AL.

TABLE 5 Preservation of Target Word Length (Number of Responses) Response length (No. of syllables) Formals

Neologisms

Target length (No. of syllables)

1

2

.2

1

2

.2

1 2 .2

91 21 0

11 23 6

1 2 4

46 7 1

21 49 17

1 4 37

were nouns (27 of 44 responses) compared to 79% of their greater-than-one phoneme deviate responses (91 of 115 responses). The grammatical class effect does not hold up for responses that deviate by only one phoneme from their targets (t(8) 5 .32, p . .6), while the effect is significant in the set of responses that deviate by more than one phoneme from their targets (t(8) 5 3.46, p , .005). Furthermore, the difference in the effect between one phoneme deviate and greater-than-one phoneme deviate responses was significant (t(8) 5 3.73, p , .006). An adequate account must explain why the grammatical class effect in formals should be sensitive to the degree of target phoneme deviation. D. Relationship between Production of Formals and Nouns If a phoneme substitution mechanism is solely responsible for the generation of formals, there should be no systematic relationship between the tendency to produce formals and the tendency to produce formals that are nouns. We assessed the existence of such a relationship in our corpus using the data presented in the ‘‘All’’ columns of Table 4. These data show that the proportion of each subject’s form-related errors that are formals, and the proportion of formals that are nouns, are positively correlated (ρ(8) 5 .85, p , .02). This is consistent with a mechanism that creates a bias for nouns when words are produced. E. Preservation Patterns Word shape preservation. Table 5 shows a breakdown of the number of trials in which target and response were both one, both two, or both greaterthan-two syllables in length, or when target and response were combinations of these lengths. The numbers along the diagonals in the matrices of Table 5 reveal the degree of word shape preservation. Preservation of target word shape is prevalent in both formals and neologisms: Target syllable number was preserved in 74% of formals (117 of 159 responses) and 70% of neolo-

465

FORMAL PARAPHASIAS

TABLE 6 Response Length as a Function of (a) Target Length and (b) Response Lexicality

Target length (No. of syllables) 1 2 .2

(a) Response length (No. of syllables) 1

2

.2

137 (.80) 28 (.26) 1 (.02)

32 (.19) 72 (.68) 23 (.35)

2 (.01) 6 (.06) 41 (.63)

(b) Response Lexicality Formals Neologisms

112 (.70) 54 (.30)

38 (.24) 87 (.47)

9 (.06) 42 (.23)

gisms (128 of 183 responses). This difference was not significant across subjects (p . .57). To determine whether there were any differences in the preservation patterns of formals and neologisms, a 2 3 3 3 3 (response lexicality 3 target length 3 response length) log linear analysis was conducted (Fienberg, 1991). The best-fit model contained two two-way interactions (target length 3 response length and response lexicality 3 response length; G 2 (6) 5 7.09, p . .31). The two components are shown in Table 6. The target length by response length component suggests an overall tendency for syllable number preservation regardless of lexical type: One syllable targets tend to result in one syllable responses, two syllable targets tend to result in two syllable responses, and greater-than-two syllable targets tend to result in greater-than-two syllable responses. This confirms the lack of a difference between formals and neologisms in overall syllable number preservation. The response lexicality by response length component suggests, however, that despite overall similar degrees of preservation, there is a difference in the particular nature of the preservation patterns of formals and neologisms: There is a tendency for formals to be one syllable in length and a tendency for neologisms to be two syllables in length, regardless of target length. This result appears to represent a true lexical bias, specifically, a bias toward the selection of short words for which a full phonological specification is more likely to be available. But there is an alternative explanation that does not invoke lexical bias: Short responses will tend to be words, and long responses nonwords, simply because of the distributional properties of the lexicon. In other words, phonological deviates that are one syllable in length have a greater likelihood of corresponding to a real word by chance alone; likewise, deviates that are more than one syllable in length have a greater likelihood

466

GAGNON ET AL.

TABLE 7 Proportion of Target Phonemes Preserved in Formals and Neologisms Subject GB HB LB NC JG LH GL VP WR

Formals

Neologisms

.51 .55 .35 .60 .43 .56 .37 .31 .26

.60 .53 .50 .60 .53 .62 .53 .50 .23

of not corresponding to a real word. Thus, the word–nonword difference is inconclusive with respect to the mechanism underlying form-related errors. The response lexicality by length effect underscores the importance of taking response length into account, however, whenever a particular lexical effect is investigated, since estimates of chance probability will vary widely depending on the subset of the lexicon being sampled (cf. Nickels & Howard, 1995). Segment preservation. In this analysis, segment preservation refers not only to preservation of segment identity (i.e., what phoneme it is), but preservation of segment position within the target as well. In other words, for a target segment to be considered preserved, the response had to preserve its phonemic identity, target syllable position (initial consonant, vowel, or final consonant), and target word position (first or second syllable). Across subjects, there was a 51% overall preservation of segments for neologisms compared to a 42% overall preservation of segments for formals (see Table 7). This difference was consistent across subjects (t(8) 5 2.87, p , .03). To further explore the nature of segment preservation, the percentage of target phonemes preserved in formals and neologisms for each of six target segment types was computed. Segment type was defined in terms of target syllable position (initial consonant, IC; vowel, V; final consonant, FC) and target syllable in which the segment occurred (first, 1; second, 2). The percentages were computed for first and second syllables only because there were relatively fewer three or four syllable targets and responses, and thus very few phonemes on which to base an analysis for these word positions. The pattern of results in Fig. 1 shows that response–initial phonemes (IC1) were best preserved for both formals and neologisms, and were preserved to virtually the same degree in both types of responses (63 vs 64%). Second, for each segment type except IC-1, the degree of preservation in formals was less than that in neologisms. To determine whether the observed word–nonword difference was statisti-

FORMAL PARAPHASIAS

467

FIG. 1. Proportion of target phonemes preserved in formals and neologisms for six segment types: initial consonant-first syllable (IC-1), vowel-first syllable (V-1), final consonant-first syllable (FC-1), initial consonant-second syllable (IC-2), vowel-second syllable (V-2), and final consonant-second syllable (FC-2).

cally reliable, an ANOVA comparing response–initial consonants to the set of noninitial segments (i.e., IC-1 vs the collapsed set of V-1, FC-1, IC-2, V-2, and FC-2 segments) was conducted. The difference between IC-1 and the rest of the segment types was significant (F(1, 32) 5 9.98, p , .004). However, the effect of lexicality was not significant (p . .24), nor was the interaction of lexicality with segment type (p . .5). This analysis demonstrates that the initial consonant position has special status in both the formals and the neologisms in our corpus. It fails to confirm the word–nonword overall segment preservation difference that was found in the initial analysis, but this latter analysis is limited to first and second syllables and thus, does not make use of all the data that the first does. The suggestion of a word– nonword difference in the degree of post-IC-1 segment preservation remains just that, pending further evidence. DISCUSSION

The form-related paraphasias upon which our analyses were based appear to be typical of those reported in similar investigations, supporting the generality of the conclusions drawn from them. For instance, Best (1996) reports

468

GAGNON ET AL.

a similar degree of target word shape preservation in her subject’s formals and neologisms, as we report here. She also found that formals not preserving target word shape tended to contain fewer syllables than their targets, similar to what we observed of our subjects’ formals. With regard to segment preservation, she reports a greater degree of segment preservation in her subject’s neologisms than in his formals, a difference also found in our subjects’ neologisms and formals. This last finding is suggestive of dual origins for the two error types, but is not enough to warrant any strong conclusions in this regard. In addition to being characteristic of other aphasics’ form-related paraphasias, the formals from our corpus appear to be similar in nature to nonaphasics’ malapropisms. Fay and Cutler (1977) report a large degree of word shape preservation, good initial segment preservation, and strong grammatical class preservation in the malapropisms they collected, all characteristic of our subjects’ formals as well. If, as is widely held, malapropisms derive primarily from lexical substitution, the similarities between malapropisms and formals hint at a lexical origin for formals as well. In the remainder of the paper, we argue that the cumulative results of our analyses strongly supports a lexical influence on the generation of formals. It is clear that lexical effects are present in the set of form-related paraphasias produced by our subjects. First, the incidence of formals in our subjects’ corpus of form-related errors is higher than the rate expected based on the density of words in the phonological lexicon. Second, the average word frequency of the CVC formals was found to be higher than that expected based on the frequency of words in a CVC lexicon. Third, grammatical class effects were found, such that more nouns were produced than expected by chance and the proportion of formals produced and nouns produced were significantly correlated across subjects. These findings conform to the predictions from the dual mechanism hypothesis, which proposed that at least some formals are malapropisms— form-based lexical substitutions. Can these findings adjudicate between the discrete two-stage retrieval model, which assigns such substitutions to the lexeme stage, and the interactive model, which assigns them to the lemma stage? The fact that formals are constrained by grammatical class and word frequency suggests that the level at which these substitutions arise is responsive to these variables. If there were independent evidence that frequency and class effects operate exclusively at the lexeme level, the discrete model would be supported; if the evidence suggested that these effects operate exclusively at the lemma level, the interactive model would be supported. Unfortunately, no such clarity exists. While there is strong motivation for assigning grammatical class effects to the lemma level (lemmas are the units that are manipulated by syntactic structures; see Levelt, 1989) there is also evidence from neuropsychological dissociations that these effects may operate at a lexeme level that is organized

FORMAL PARAPHASIAS

469

by grammatical class (Caramazza & Hillis, 1991). Frequency effects, too, have an ambiguous locus. For example, Dell (1990) argues for a lemmalevel explanation, while Jescheniak and Levelt (1994) argue for a lexemelevel locus. As they are currently formulated, versions of both discrete and interactive models can account for form-based word substitutions sensitive to frequency and grammatical class. Our primary motivation for establishing the existence of lexical effects in the error corpus was to strengthen the case for the dual mechanism account of formals, which allows for form-based lexical substitution, against the single mechanism account, which does not. The single mechanism account is problematic on its face: Given that the normal production system is vulnerable to form-based word substitutions, why not the aphasic system? The existence of individual aphasics who make formals at rates that exceed chance adds to the implausibility of the single mechanism account. And the data reported here undermine this account still further by showing that this lexical bias is evident as well in a large corpus of errors from unselected subjects. However, lexical bias does not unambiguously implicate lexical substitution. As was noted in the Introduction, an editor that monitors for phonemic errors and corrects them prior to articulation could produce lexical bias if it were responsive to the wordness of the monitored string. Indeed, if this responsiveness to wordness were the consequence of a more general sensitivity to the familiarity of the monitored string, then the word frequency effect would be predicted as well: A phonemic error that creates a familiar word would be more likely to go undetected than an error that creates an unfamiliar word or a nonword. By multiplying the properties of the editor, other observed effects can be accounted for as well: An editor that is sensitive to the grammatical class of the target word would be more likely to detect a phonemic error that creates a word of a different grammatical class (in this case, a nonnoun). And the chances of letting through a nonnoun could be affected by the phonemic similarity to the target, such that nonnouns would be more likely to pass when they closely resemble the target than when they do not. This would account for our finding that the tendency for formals to be nouns was significant only for that subset of formals that deviated from the target by more than one phoneme. In our view, the output editor account suffers from being post hoc and thus, too powerful. There is no a priori basis for attributing to an output editor a particular set of properties, and nothing to constrain its further elaboration as the need arises. The force of this criticism is blunted somewhat if we assume that the editing function is subserved by the language comprehension system (e.g., Levelt, 1983; Monsell, 1987), in which case the proposed sensitivity to lexicality, frequency, and grammatical class does have some independent motivation. On the other hand, such a proposal shifts the burden for explaining the presence or absence of lexical effects in phonological errors

470

GAGNON ET AL.

from the production system to the comprehension system. We know of only one study that has evaluated this proposal directly, and the evidence there was negative (Nickels & Howard, 1995). In that study, no correlation between aphasics’ form-related errors in a naming task and their phonological errors in comprehension tasks (minimal pair judgments, lexical decisions, and synonymy judgments) was found. Also, there was no evidence of a relationship between monitoring behaviors (e.g., word fragments, self-interruptions, self-corrections) and degree of auditory comprehension. To summarize, given the strong evidence for lexical effects in the data, the most straightforward account is that the aphasic system is vulnerable to lexical substitutions at a level of the retrieval system that is responsive to word frequency and grammatical class. This implies that at least a subset of the aphasics’ formals are malapropisms, and are generated in the same way. This further extends the parallels between paraphasias and normal speech errors and, hence, the relevance of aphasia data for models of normal speech production. REFERENCES Baars, B. J., Motley, M. T., & MacKay, D. G. 1975. Output editing for lexical status in artificially elicited slips of the tongue. Journal of Verbal Learning and Verbal Behavior, 14, 382–391. Best, W. 1996. When racquets are baskets but baskets are biscuits, where do the words come from? A single-case study of formal paraphasic errors in aphasia. Cognitive Neuropsychology, 3, 443–480. Blanken, G. 1990. Formal paraphasias: A single case study. Brain and Language, 38, 534– 554. Blumstein, S. E. 1973. A phonological investigation of aphasic speech. The Hague: Mouton. Buckingham, H. W. 1980. On correlating aphasic errors with slips-of-the-tongue. Applied Psycholinguistics, 1, 199–220. Buckingham, H. W. 1981. Where do neologisms come from? In J. W. Brown (Ed.), Jargonaphasia. New York: Academic Press. Buckingham, H. W. 1992. Phonological production deficits in conduction aphasia. In S. Kohn (Ed.), Conduction aphasia. Hillsdale, NJ: Erlbaum. Butterworth, B. 1979. Hesitation and the production of verbal paraphasias and neologisms in jargon aphasia. Brain and Language, 8, 133–161. Butterworth, B. 1989. Lexical access in speech production. In W. Marslen-Wilson (Ed.), Lexical representation and process. Cambridge, MA: MIT Press. Butterworth, B. 1992. Disorders of phonological encoding. Cognition, 42, 261–286. Caramazza, A., & Hillis, A. E. 1991. Lexical organization of nouns and verbs in the brain. Nature, 349, 788–790. Carterette, E. C., & Jones, M. H. 1974. Informal speech: Alphabetic and phonemic texts with statistical analyses and tables. Los Angeles: University of California Press. Chute, D. L. 1990. MacLaboratory for psychology. Devon, PA: MacLaboratory Inc. Dell, G. S. 1986. A spreading activation theory of retrieval in sentence production. Psychological Review, 93, 283–321. Dell, G. S. 1988. The retrieval of phonological forms in production: Tests of predictions from a connectionist model. Journal of Memory and Language, 27, 124–142. Dell, G. S. 1990. Effects of frequency and vocabulary type on phonological speech errors. Language and Cognitive Processes, 5, 313–349.

FORMAL PARAPHASIAS

471

Dell, G. S., & O’Seaghdha, P. G. 1991. Mediated and convergent lexical priming in language production: A comment on Levelt et al. 1991. Psychological Review, 98, 604–614. Dell, G. S., & O’Seaghdha, P. G. 1992. Stages of lexical access in language production. Cognition, 42, 287–314. Dell, G. S., Schwartz, M. F., Martin, N., Saffran, E. M., & Gagnon, D. A. (in press). Lexical access in normal and aphasic speakers. Psychological Review. Ellis, A. W. 1985. The production of spoken words: A cognitive neuropsychological perspective. In A. W. Ellis (Ed.), Progress in the psychology of language, Vol. 2. Hillsdale, NJ: Erlbaum. Ellis, A. W., Miller, D., & Sin, G. 1983. Wernicke’s aphasia and normal language processing: A case study in cognitive neuropsychology. Cognition, 15, 111–144. Fay, D., & Cutler, A. 1977. Malapropisms and the structure of the mental lexicon. Linguistic Inquiry, 8, 505–520. Fienberg, S. E. 1991. The analysis of cross-classified categorical data. Cambridge, MA: MIT Press. Francis, W., & Kucˇera, H. 1982. Frequency analysis of English usage: Lexicon and grammar. Boston: Houghton Mifflin Co. Fromkin, V. A. 1971. The non-anomalous nature of anomalous utterances. Language, 47, 27– 52. Gagnon, D. A., & Schwartz, M. F. 1996. The origins of neologisms in picture naming by fluent aphasics. Brain and Cognition, 32, 118–120. Garrett, M. F. 1975. The analysis of sentence production. In G. Bower (Ed.), Psychology of learning and motivation. New York: Academic Press. Garrett, M. F. 1992. Disorders of lexical selection. Cognition, 42, 143–180. Goodglass, H., & Kaplan, E. 1983. The assessment of aphasia and related disorders. Philadelphia: Lea & Febiger. Harley, T. 1984. A critique of top-down independent levels models of speech production: Evidence from non-plan-internal speech errors. Cognitive Science, 8, 191–219. Jescheniak, J. D., & Levelt, W. J. M. 1994. Word frequency effects in speech production: Retrieval of syntactic information and of phonological form. Journal of Experimental Psychology: Learning, Memory, and Cognition, 20, 824–843. Kempen, G., & Huijbers, P. 1983. The lexicalization process in sentence production and naming: Indirect elicitation of words. Cognition, 14, 185–209. Kenyon, J. S., & Knott, T. A. 1953. A pronouncing dictionary of American English. Springfield, MA: G. & C. Merriam Co. Kohn, S. E., & Smith, K. L. 1994. Distinctions between two phonological output deficits. Applied Psycholinguistics, 15, 75–95. Lecours, A. R. 1982. On neologisms. In J. Mehler, E. C. T. Walker, & M. F. Garrett (Eds.), Perspectives on mental representation. Hillsdale, NJ: Erlbaum. Lecours, A. R., Deloche, G., & Lhermitte, F. 1973. Paraphasies phone´miques: Description et simulation sur ordinateur. In Colloquies IRIA—Informatique Medical. Rocquencourt: Institut de Recherche d’Informatique et d’Automatique. Lecours, A. R., & Rouillon, F. 1976. Neurolinguistic analysis of jargonaphasia and jargonagraphia. In H. Whitaker & H. A. Whitaker (Eds.), Studies in neurolinguistics, Vol. 2. New York: Academic Press. Levelt, W. J. M. 1983. Monitoring and self-repair in speech. Cognition, 14, 41–104. Levelt, W. J. M. 1989. Speaking: From intention to articulation. Cambridge, MA: MIT Press. Martin, N., Dell, G. S., Saffran, E. M., & Schwartz, M. F. 1994. Origins of paraphasias in deep dysphasia: Testing the consequences of a decay impairment to an interactive spreading activation model of lexical retrieval. Brain and Language, 47, 609–660. Martin, N., & Saffran, E. M. 1992. A computational account of deep dysphasia: Evidence from a single case study. Brain and Language, 43, 240–274. Miller, D., & Ellis, A. W. 1987. Speech and writing errors in ‘‘neologistic jargonaphasia’’:

472

GAGNON ET AL.

A lexical activation hypothesis. In M. Coltheart, G. Sartori, & R. Job (Eds.), The cognitive neuropsychology of language. Hillsdale, NJ: Erlbaum. Monsell, S. 1987. On the relation between lexical input and output pathways for speech. In A. Allport, D. MacKay, W. Prinz, & E. Scheerer (Eds.), Language perception and production: Relationships between listening, speaking, reading and writing. New York: Academic Press. Nickels, L., & Howard, D. 1995. Phonological errors in aphasic naming: Comprehension, monitoring and lexicality. Cortex, 31, 209–237. Roach, A., Schwartz, M. F., Martin, N., Grewal, R. S., & Brecher, A. 1996. The Philadelphia Naming Test: Scoring and rationale. Clinical Aphasiology, 24, 121–133. Schwartz, M. F. 1987. Patterns of speech production deficit within and across aphasia syndromes: Application of a psycholinguistic model. In M. Coltheart, G. Sartori, & R. Job (Eds.), The cognitive neuropsychology of language. Hillsdale, NJ: Erlbaum. Schwartz, M. F., Saffran, E. M., Bloch, D. E., & Dell, G. S. 1994. Disordered speech production in aphasic and normal speakers. Brain and Language, 47, 52–88. Shattuck-Hufnagel, S. 1979. Speech errors as evidence for a serial ordering mechanism in sentence production. In W. E. Cooper & E. C. T. Walker (Eds.), Sentence processing. Hillsdale, N.J.: Erlbaum. Shattuck-Hufnagel, S. 1987. The role of word-onset consonants in speech production planning: New evidence from speech error patterns. In E. Keller & M. Gopnick (Eds.), Motor and sensory processing in language. Hillsdale, NJ: Erlbaum. Stemberger, J. P. 1984. Structural errors in normal and agrammatic speech. Cognitive Neuropsychology, 1, 281–313. Stemberger, J. P. 1985. An interactive activation model of language production. In A. W. Ellis (Ed.), Progress in the psychology of language, Vol. 1. Hillsdale, NJ: Erlbaum. Stemberger, J. P., & MacWhinney, B. 1986. Frequency and the lexical storage of regularly inflected words. Memory and Cognition, 14, 17–26. Valdois, S., Joanette, Y., & Nespoulous, J-L. 1989. Intrinsic organization of sequences of phonemic approximations: A preliminary study. Aphasiology, 3, 55–73. 1993. Merriam-Webster’s collegiate dictionary, tenth edition. Springfield, MA: MerriamWebster, Inc.