BRAIN AND LANGUAGE 47, 279-299 (1994) Graphemic Jargon: A Case Report K. ScHONAUER Psychiatrische Universitdtsklinik Mitinster, Germany AND G. Denes Clinica Neurologica, Universita di Padova, Italy We report on a patient with left hemispheric thromboembolic stroke whose writing performance on single word dictation following recovery from an aphasic syndrome remained severely impaired but fluent. Having only very fragmentary command of the target’s written spelling she produced neologistic nonwords which were approximately the same length and contained, in addition to persev- erative intrusions and unidentifiable errors multiple insertions, deletions, transpo- sitions, and substitutions but only few unpronounceable combinations. The pa- tient was tested 4 and 10 months postonset, before and after having attended a rehabilitation program for dysgraphia. Comparing pre- and post-rehabilitation er- ror corpora we found, besides a slight improvement of her severe dysgraphia, a highly significant predilection of in-class substitutions regarding the consonant/ vowel status of misspelled graphemes. The second error corpus revealed some influence of different consonant/vowel patterns among the targets on the emer- gence of spelling errors. We discuss the hypothesis of an influence of nonlexical “praphotactic’’ features on the spelling process as has been revealed by other cases of acquired dysgraphia published in recent years. © 1994 Academic Press, Inc. INTRODUCTION Recent research on the neuropsychology of the spelling process (as reviewed by Patterson, 1988; Shallice, 1988; Ellis & Young, 1988) has accumulated some empirical evidence that the various error patterns, which occur in written spelling tasks performed by patients with acquired dysgraphias, can be described through a set of empirically consistent hypothesis which have been framed by several attempts to model the ‘functional architecture’’ of the spelling process in a comprehensive and integrated manner. Beyond all conceptual differences most of these mod- Address reprint requests to Gianfranco Denes, Clin. Neurologica, Univ. di Padova, Via Giustiniani 5, Padova 35128, Italy. 279 0093/934X/94 $6.00 Copyright © 1994 by Academic Press, Inc. All rights of reproduction in any form reserved. 280 SCHONAUER AND DENES els have at least one basic assumption, as well as a theoretical framework in common. The latter has been recently described by Caramazza and Miceli (1990) as ‘‘the Information Processing Paradigm’’ which intro- duces the perception of a cognitive process ‘‘as a series of transforma- tions of mental representations.’’ The transformational aspect of any cog- nitive activity that deals with written language is very evident because teading, spelling, and writing imply—or even consist of—exceptionally pure and clear-cut examples for the transformation of one representation “‘into’’ another. These transformations are usually described as acts of transcodification. The basic assumption, which seems to be shared by most contempo- rary attempts to model the functional architecture of the spelling process, is a further subdivision of the transformational activities emerging from the Information Processing Paradigm into rather central and ‘‘genera- tive’’ functions, on the one hand, and more peripheral and ‘‘conserva- tive’ functions, on the other. Although the first have to raise, activate, retrieve, or ‘“‘compute’’ information, usually by confronting the input information with long-term resident algorithms, the latter have to tempo- rarily store or ‘‘hold’’ those transient representations for further pro- cessing. In terms of functional architecture, the first kind of functions are modeled as ‘‘conversion mechanisms”’ or ‘‘buffers.’” Caramazza and co-workers (1987, 1989, 1990) have reported on a dys- graphic patient (L.B.) whose spelling errors revealed a specific impair- ment of the ‘‘Graphemic Buffer,’’ which they defined as a working mem- ory system that temporarily stores graphemic representations and “‘mediates between the processes that compute graphemic representa- tions and more peripheral output processes”’ (Caramazza & Miceli, 1989). Their patient very regularly (on an average about two misspelled letters in three words) committed spelling errors that were not attributable to lexical or phonological factors of the target but covaried with the target’s length. Those errors, which consisted of the presence of a nonrequired letter or the absence of a required letter in the written production, re- vealed deficits of letter-identity information, whereas misplacements of required letters in wrong positions seemed to be suggestive of a presum- ably different kind of information loss, particularly concerning the spatial or sequential positions of letters. Caramazza and co-workers (1987) cau- tiously concluded: The intuition is that the Graphemic Buffer stores a spatially coded sequence of graphemes and that damage to this processing component would result in loss of specific item and/or spatial information. Consequently, errors should take the form of substitutions, insertions, deletions or transpositions of letters. The first indications of distinctive representations for spatial informa- tion and letter-identity information along the spelling process had previ- GRAPHEMIC JARGON 281 ously been published by Kinsbourne and Warrington (1964), who had reported on the prevalence of spatial information deficits in spelling errors committed by patients with finger agnosia. More recent case reports on specific impairments of the Graphemic Buffer (Posteraro, Zinelli & Maz- zucchi 1988; Katz 1991) confirmed the hypothesis that both kinds of infor- mation processing deficits could be attributed to dysfunctions of this buffer. In subsequent analysis of L.B.’s spelling errors, Caramazza and Miceli (1989, 1990) demonstrated, among others, two graphotactic fea- tures which had affected the emergence of his spelling errors: first, they showed that words which were composed of alternating consonant— vowel patterns (e.g., ‘“domenica’’ [Sunday]) caused significantly fewer spelling errors that were predominantly letter substitutions, whereas words (and nonwords) containing also bivocals, trivocals, biconsonants, and triconsonants provoked more spelling errors, in particular deletions. Second, they reported that more than 99% of the letter substitutions in L.B.’s error corpus were in-class substitutions regarding the consonant/ vowel distinction. This means that vowels had been substituted almost exclusively by vowels and consonants by consonants, respectively. This observation raised the idea that, besides the letter-identity information, some information concerning the consonant/vowel status of the required graphemes could be stored in this buffer and that this information con- cerning the consonant vowel status would have to be linked to the spatial information. Otherwise, it would not have been explicable how the pa- tient, having no sufficient access to letter-identity information, replaced the required letter by a vicarious element that belonged to the same cate- gory. This was the first empirical evidence for the hypothesis that likely to characters of non-alphabetic systems (Sasanuma & Fujimura 1972) also letters of alphabetic systems would not only be represented as ab- stract entities but also that their representation could also be surrounded by the processing of additional information concerning particular qualities of these—hence no longer ‘‘abstract’’—entities. Cubelli (1991) confirmed the hypothesis of such a ‘‘consonant vowel tier’’ when he reported on a patient, C.W., who also made on average about two spelling errors among three dictated words which were not affected by lexical features but by the length of the target. C.W. committed predominantly spelling errors of the substitutive type, 99% of which were in-class substitutions regard- ing the consonant/vowel status of the interchanged elements, and he had significantly more difficulties processing vowels than consonants. Fur- thermore Cubelli reported on another dysgraphic patient, C.F., with a transient selective deficit for writing vowels, who wrote down only the consonants of a word and left blank spaces in the sequential positions of the word’s vowels. Although emerging from a fairly pure case of spelling disorder, the clinical and experimental findings, which are presented here, are hardly 282 SCHONAUER AND DENES localizable in terms of functional architecture. Nevertheless, we do re- gard our patient’s errors to be worthy of discussion in the light of the Information Processing Paradigm, because the deficits of representational information, which these errors seem to reveal, are so large that it turns out to be rather remarkable that she remained capable of performing her writing tasks so fluently. The configuration of a fluent, even slightly exceeding, but predominantly fallacious productivity lead us to call our patient’s writing behavior ‘‘jargon-like’’ because it seems to be, in anal- ogy to spoken jargon (Ryalls, Valdois, & Roch Lecours 1988), a fluent production that contains enough deviations for the reader to be unable to make conventional sense of it. As it will be clarified in the course of the work, and although for many errors the source could be recon- structed, the number of graphemic distortions for each word was so high that we preferred to define the patient’s production as jargon rather than multiple phonemic paragraphias. However, the patient's writing difficulty posed a question which may often arise from any kind of jargon-like expressive behavior: if the patients are not able to express or ‘“‘put out’’ what they might have intended to, why are they able to produce so much “‘output’’ that is too heterogenous to be explained by mere ‘‘impover- ishment’’ or ‘‘stereotypy’’? This question cannot be answered yet, al- though it has undergone broad and highly elaborated treatment (Brown, 1981). The focus of our considerations is the attempt to specify those graphotactic and orthographic features that seem to be preserved in our patient's written spelling, notwithstanding the inaccessability of the pre- sumably major part of the information which would be required to accom- plish the tasks. CASE REPORT C.K. is a 64-year-old right-handed female patient with 8 years of schooling who became aphasic after left hemispheric stroke. CT scan evidenced a cortical left fronto-temporo-parietal perisylvian lesion. Three months postonset, she showed good improvement with the exception of a persistent writing disorder. Her spontaneous speech had recovered very well, auditory comprehension was nearly flawless and her oral naming and capacity to read and understand single words or short phrases were good. In her daily life she still felt taxed when reading books or longer texts, but could extract the essential information from a newspaper article with some effort and identify signposts or brand names on packages with- out difficulty. There was no spontaneous writing activity in her daily life. In the clinical testing situation, her handwriting was composed of well-formed and recognizable letters, but on single word dictation or in a written naming task, she produced only neologisms, few of them being recognizable at first sight as target related. Neurological examination was GRAPHEMIC JARGON 283 unremarkable apart from the aphasic disturbances. She was not apraxic, but mildly acalculic. Her digit span was 5. In a standardized language examination battery (Italian version of the Aachener Aphasie Test Luz- Zatti et al., 1991), she scored in a pattern consistent with mild amnestic aphasia: Spontaneous speech scoring: Articulation/prosody 5 out of 5 points, automatic speech 4 out of 5 points, semantic structure 4 out of 5 points, phonemic structure 5 out of 5 points, syntactic structure 5 out of 5 points, communicative skills 4 out of 5 points (= spontaneous speech total: 27 out of 30 points). Task testing: Token test 23 out of 50 errors, repetition 139 out of 150 points, reading 23 out of 30 points, written spelling 15 out of 60 points, oral naming 73 out of 120 points, comprehen- sion 105 out of 120 points. Experimental Tasks Given the characteristics of Italian ortography, a ‘'shallow’’ language without irregular or ambiguous words, subjects with 8 years of schooling are expected to perform flawlessly in writing tasks. To study the patient’s written transcodification abilities in more detail, C.K. was submitted to the following tests: Written Output Tasks Single letter writing. The patient was asked (1) to write down by heart the whole Italian alphabet. In a subsequent testing session she had (2) to write all letters of the Italian alphabet (in randomized order) to dictation. Allographic conversion from model. One hundred and eighteen letter strings (words and nonwords) printed in uppercase block letters were presented on individual index cards and (3) had to be converted into cursive handwriting. Retarded allographic conversion. Two hundred and thirty-one items (14 syllables, 31 nonwords, and 186 words) printed in uppercase block letters on index cards were individually shown to the patient for as long as she requested (usually a few seconds) and then removed. Stimulus features varied in the list included word class (concrete and abstract nouns, proper names, inflected verbs, adjectives, words with uncommon tonic stress, and functors), lexical frequency (according to the sample of Bortolini et al., 1972), and word length. The task was (4) to write the item down immediately after removal of the model. Although it was not explicitly required in the instruction, C.K. wrote all of her responses in cursive handwriting. This task was very similar to the ‘“‘delayed copy” task, which has been proposed by other authors and which introduces an additional fixed delay of 3 (Caramazza et al., 1987) or 10 (Posteraro et al., 1988) sec between the interruption of visual access and the begin- ning of writing down the answer. 284 SCHONAUER AND DENES Writing to dictation. The same 231 items described in the previous paragraph were administered (5) as a writing to dictation task. Typing to dictation. We asked the patient (6) to type single words and nonwords to dictation using the keyboard of a desktop computer. The task was performed very slowly and turned out to be tiresome, probably because she had never been accustomed to typing. Therefore only a part of the battery was administered for a total of 110 items. Written naming. The patient was asked (7) to write down the names of 20 visually presented items (line drawings and colors) which were taken from the Aachener Aphasie Test and had been (orally) named cor- rectly by her in a previous testing session. Oral Output Tasks Reading and repetition. The entire battery (231 items) described above was presented visually for (8) reading aloud and orally for (9) repetition. Oral spelling/phonemic decomposition. C.K. was unable (10) to spell orally any word or nonword presented to her auditorily. Instead she (11) fluently pronounced the individual phones (i.e. realizations of phonemes) of any item in a row, claiming to have no access to the required letter names. This strategy implied that she usually “‘fused”’ those letter groups which are represented by a single phone (most geminates, but in Italian also (gn), (gl), (ch), and others) to one single output element. All of these tasks were performed 4 months postonset. Table | presents a survey of C.K.’s performance measured by the relative frequencies of correctly processed items in the various tasks. TABLE | Relative Frequencies of Correctly Processed Items in Various Transcodification Tasks Written output Letters (1) Alphabet generation 81% (17/21) (2) Letter dictation (rand. order) 95% (20/21) Words (3) Allographic conversion from model 74% (88/118) (4) Retarded allographic conversion 35% (81/231) (5) Writing to dictation 3% (6/231) (6) Typing to dictation (keyboard) 10% (11/110) (7) Written naming 0% (0/20) Oral output (words) (8) Reading 79% (181/231) (9) Repetition 94% (218/231) (10) Oral spelling (No output achievable) (11) Phonemic decomposition 59% (37/63) Note. For tasks (1) and (2), items are single letters; in all other cases, single words. GRAPHEMIC JARGON 285 These data are only meant to give a rough orientation. They do not allow statistical distribution testing because they have been raised in too different contexts. Nevertheless, they reveal C.K.’s severe and fairly pure impairment of spelling which was about equally evident in all rele- vant tasks (5, 6, 7, 10) in the absence of quantitatively comparable deficits in single-letter-processing (tasks 1, 2) or reading (task 8). The dissociation between the two allographic conversion tasks (3, 4) raises the hypothesis that dysfunctional working memory might be involved in the spelling disorder. Further Observations on Single Word Dictation Qualitative Analysis Although single word dictation was-—apart from written naming—the most difficult writing task for her, it was nevertheless performed most fluently. Having listened to the stimulus word, she usually made up her mind for a few seconds and, once starting to write, she concluded her response rapidly with at most one break. She reported that the task was much easier for her in connected handwriting than in single letters or block letters. While writing to dictation she was aware of not exactly achieving the targets but was very undecided when asked to specify which part or aspect of the items she had reproduced correctly and which were incorrect. (This indecision was also present in those six cases where she achieved the target). She also reported having access to the word’s semantic content when writing and could identify the nonwords as such. Although the instructions did not contain any particular restriction in this respect, the patient did not repeat the words aloud or silently when writ- ing. She described her performance on the dictation task as ‘‘following a spontaneous but very vague track.”’ There was a good correlation (r = .763) between the length of any target and her respective response, the latter being on average 1.2 letters longer. But comparing the graphemic elements, there were only weak correspondences. On average only a fraction of .58 of the letters of her responses was part of the respective target and only a fraction of .35 appeared in corresponding sequential positions. Nevertheless it should be stressed that the ‘‘correctness’’ of her responses was markedly above any level that could be achieved by chance. Apart from the six items she spelled correctly, all other responses produced to dictation were grapho- tactically acceptable written neologisms. Twenty-one of her responses were equally or more related to preceeding stimuli or responses than to the actual item hence giving evidence for occasional perseverative “‘intrusions.’’ However, it must be stressed that perseverations occurred exclusively in this particular task. 286 SCHONAUER AND DENES Comparison with Retarded Allographic Conversion Working on the retarded allographic conversion task, C.K. still pre- ferred unconnected but cursive handwriting. She committed fewer but qualitatively the same kind of errors (described in more detail under Error Analysis below) as in dictation. Here, an average fraction of .84 of the response letters was part of the target and a fraction of .79 appeared in corresponding positions. (The smaller difference between these two fractions compared to the dictation task could be interpreted as a rela- tively better preservation of ‘‘spatial’’ information. This observation could be discussed in relation to the presumably different ‘processing routes”’ of the two tasks: in the dictation task, both the letter-identity information and the spatial information have to be completely ‘‘gener- ated,’’ whereas in the retarded allographic conversion task they are—as an alternative or as a sort of ‘‘repair device’’—also accessible via visual input. In a further step it could be hypothesized that the processing of spatial information might benefit more from this visual access because it could have a more intrinsically visual character than letter identity information. In the absence of any further validating data, we decided to refrain from a discussion of this hypothesis.) Effects of Word Classes, Lexical Frequency, and Length on Single Word Dictation Outcome The correctness of responses in dictation tasks is usually measured by the relative frequency of correctly processed words (and nonwords), which allows nonparametric distributional testing to compare the sub- ject’s performance in various subsamples. This procedure is not feasible regarding C.K.’s written spelling performance because the relative fre- quency of correctly processed words in the dictation task (6/231) is too low. This is why we specified for each single item the fraction of letters that occur in the target as well as in the response (a) in whatever position and (b) in corresponding positions. These fractions, to which we will in future refer as ‘“‘hitrates,”’ are meant to be parametric estimations of correctness in relation to the stimulus-length because the absolute num- ber of ‘‘shared letters’ of any single target-response pair is always di- vided by the number of letters of the target. Figure | displays the average hitrates per single word (a) regardless of the positions of the letters ‘‘shared’’ between responses and targets (solid black columns) and (b) taking into consideration only those ‘“‘shared”’ letters which appear in corresponding positions (striped columns). These hitrates are computed separately in all three subsamples that have to be drawn in order to compare among each other the two degrees of fre- quency/familiarity, the nine word classes, and the different word lengths as they are represented in the battery. Beside these columns are listed GRAPHEMIC JARGON g 2 g & 3 2 8 2 2 3 a 3 5 104 8 8 8 2 2: 228 @, 8 & 5 6 = = 8 2 3° 0.8 2 @ w 0.6 0.4 ANOVA BB RegardlessPos p=0.25 Regarding POS p=0.6 0.2 Average Hitrates per Word by Grammatical Classes 1.0 0.8 0.6 0.4 ANOVA GE RegardiessPos —p=0.46 0.2 EB RegardingPos — P=0.07 0 . = Average Hitrates per Word by Frequency/Familiarity 1.0 8 letters Alators slotors Gletters Tletters Sletters 0.6 0.4 ANOVA $B RegardiessPos p=0.152 0.2 RegardingPos P=0.0001* Average Hitrates per Word by Wordlength 287 Fic. 1. Hitrates in the first dictation task broken down to word classes, lexical frequency and wordlength. *Hypothesis of equidistribution can be rejected. 288 SCHONAUER AND DENES the probabilities for erroneous rejections of the absence of distributional class effects as an outcome of one-way analysis of variance within each subsample. (Of course we compared only solid with solid and striped with striped columns. The difference between the fraction of shared letters regardless their position and the fraction of those in corresponding posi- tions must be ‘‘significant’’ qua definitionem because the latter are always included in the first.) The first impression when viewing the columns is that class effects seem to be mostly rather weak. This impression is confirmed by the testing procedure which reveals a significant (p < 0.05) effect on the distribution of hitrates only regarding corresponding positions across the various wordlengths, The influence of the category ‘‘frequency/famil- iarity’’ on the positional hitrate just fails to meet the—arbitrarily fixed— significance threshold. What is well worth noting is the fact that the sup- posed influence of wordlength in the dictation task does not evolve in a linear way. The most correctly transcodified items are those composed of four or five letters, respectively, and items shorter or longer than this are transcodified less correctly. In fact five of the six items perfectly written to dictation belong to these two length categories. Error Analysis Any comparison of written responses on single-word output tasks with their corresponding targets contains some speculative or conjectural ele- ment. This amount of speculation can be estimated as very near to zero as long as these comparisons reveal predominantly single errors. A written reproduction of the word ‘“‘DONA”’ [imperative of ‘‘to donate’’] as ‘‘doma’”’ offers as a very obvious hypothesis the presumption that one of its constituting elements has been substituted with another. It is similarly probable that the transcodification of the word ‘‘MAIALE”’ [‘‘pig’’] as ‘‘maniale’’ and the nonword ‘‘PLA”’ as ‘‘pa’’ leads to the hypothesis of a single insertion or deletion, respectively. With a considerably lower degree of probability the transcodification of ‘MANO’ [‘‘hand’’] as ‘‘bana”’ induces hypothetically the emergence of two substitutions. If we consider the transcodification of the Italian surname ‘‘GIO- VANNI” as ‘‘lograrunio”’ as it is sketched in Fig. 2, it seems to be at [male surname] Vi ACN)N I 6190 PKI | |? fo girjalr nite Fic. 2. Paragraphic example from C.K.'s error corpus. GRAPHEMIC JARGON 289 least plausible that the first three elements have undergone a sort of transposition, that three of the subsequent elements ((a), (n) and (i)) have been transcodified correctly, that possibly a (v) has been replaced by an (r), and that the final (o) could be regarded as the insertion of an element. There is another corresponding feature to mention: Usually the patient produced only lowercase letters with the occasional exception that she wrote most of the proper names with an initial uppercase letter, as she did with lograrunio. If the hypothesis of a mutual transposition of the first three elements of Giovanni is valid, in this case this would mean that the