Loquens 2(2) July 2015, e023 eISSN 2386-2637 doi: http://dx.doi.org/10.3989/loquens.2015.023 Predicting segmental substitution errors in aphasic patients with phonological and phonetic encoding impairments Anna Marczyk and Lorraine Baqué Universitat Autònoma de Barcelona anna.marczyk@uab.es, lorraine.baque@uab.es Submitted: 10/04/2015. Accepted: 17/07/2015. Available on line: 13/06/2016 Citation / Cómo citar este artículo: Anna Marczyk and Lorraine Baqué. Predicting segmental substitution errors in aphasic patients with phonological and phonetic encoding impairments. Loquens, 2(2), e023. doi: http://dx.doi. org/10.3989/loquens.2015.023 ABSTRACT: This paper analyses the factors that predict substitution errors produced by four Broca’s and four con- duction aphasic subjects, all native speakers of Spanish, in reading and repetition tasks. Errors were elicited using a list of words where type of consonant, lexical stress and phonetic context were controlled for and where variables related to frequency of occurrence (word and syllable) and phonological neighbourhood characteristics were as- signed using available online corpora. 675 substitution errors were obtained and preferential tendencies to devoice, occlusivise or spirantise were identi- fied. Logistic regression mixed-effect models were performed on these three types of substitution errors to identify the predictors depending on the aphasic profile. While our results lent support to the hypothesis of a concomitant phonetic deficit in fluent aphasia, contrary to the classical claim, it also revealed differential patterns in the phonic behaviour of patients regarding the access to men- tal syllabary or syllabic position effects. Our results are discussed in relation to the phonetic vs. phonological impairments dimension in aphasia and the seri- ality/interactivity axis in speech architectures. Keywords: phonetic and phonological impairments; aphasia; substitution errors. RESUMEN: Estudio de predictores de los errores segmentales de sustitución en pacientes afásicos con un déficit fonético-fonológico. Este estudio se propone identificar los factores que permiten predecir la aparición de errores segmentales de sustitución producidos por pacientes con afasia de Broca y de conducción en tareas de lectura y de repetición. El corpus utilizado para obtener los errores consistía en 240 palabras en las que las consonantes objeto de estudio —oclusivas, fricativas y africadas del español— se encontraban en distintos contextos fonéticos y acentua- les. Posteriormente, se atribuyó a las palabras del corpus los valores relativos a la frecuencia léxica y silábica, así como las características relacionadas con la vecindad fonológica, empleando para ello corpus disponibles en línea. En total, se obtuvieron 675 sustituciones, en las que prevalecen tres tendencias: ensordecimiento de las oclusivas sonoras, refuerzo (oclusivización) de las fricativas y espirantización de las oclusivas. Para identificar los predictores de cualquiera de estos tres tipos de substitución en relación con el perfil clínico de afasia, se efectuó un análisis me- diante el modelo mixto de regresión logística. Los resultados son globalmente congruentes con la hipótesis de un déficit fonético concomitante al déficit fonológico en la afasia de conducción, contrariamente a las predicciones de la hipótesis clásica. Sin embargo, el examen de los errores revela también patrones de tratamiento fónico distintos según el cuadro clínico, en relación con el silabario mental y la influencia de la posición silábica. Se propone una interpretación de estos resultados en el marco del espectro de déficits fonético-fonológicos en la afasia, así como en relación con la dimensión serialidad vs. interactividad que los modelos de codificación del habla. Palabras clave: trastornos fonéticos y fonológicos; afasia; errores de sustitución. Copyright: © 2015 CSIC This is an open-access article distributed under the terms of the Creative Commons Attribution-Non Commercial (by-nc) Spain 3.0 License. 2 • Anna Marczyk and Lorraine Baqué 1. Introduction perceived as segmental substitutions are among the main symptoms of several acquired neurological disor- Speech production models guiding research in the field ders including Broca’s aphasia, apraxia of speech and of acquired speech impairments are anchored in contrasting conduction aphasia, and the underlying mechanisms that viewpoints about the nature of language and mechanisms give rise to them are generally situated at one of the pro- involved in speech production and its impairments. One cessing levels between lexical access and articulation such viewpoint holds that language is composed of relative- (e.g. Code, 1998; Ziegler, 2002, 2008). However, what ly autonomous processing subsystems where each of them mechanism is exactly responsible for these errors re- is assigned a specific functional role such as lexical or mor- mains a matter of debate (e.g. recent studies or reviews phological processing. In this account, speech production is by Buchwald & Miozzo, 2012; Kurowski & Blumstein, conceptualised as a linear top-down process combining re- 2016; Laganaro, 2015; Pouplier & Hardcastle, 2005). trieval of information stored in the memory and computa- The present paper is highly concerned with this question tion processes involving units of encoding specific to a giv- and seeks to contribute to the existing evidence by ad- en domain. In architectures based on these assumptions dressing some of the understudied issues related to the (e.g. Garrett, 1980; Levelt, 1989, 1999), phonological and mechanisms responsible for substitution errors in Bro- phonetic encodings are thus regarded as two functionally ca’s and conduction aphasia. In the following para- separate and largely independent components organised hi- graphs we will situate our research questions within the erarchically with phonological encoding giving output for broad context of studies on phonemic paraphasias and further phonetic programming. Importantly, the word-form the factors that may constrain them. encoding proceeds according to the principle of seriality, In aphasiology, and especially in the field of phonemic dominant in this account, which postulates that only one paraphasias, one of the main research goals over the past item selected during the phonological encoding can consti- decades has been to establish a threefold correspondence be- tute the entry of the phonetic module, where it is further tween symptoms, underlying causes and aphasic syndromes translated into motor commands. A shortcoming of such ar- (Blumstein, 1973; Blumstein, Cooper, Goodglass, Statlender chitectures, with respect to phonetic and phonological pro- & Gottlieb, 1980; Buchwald & Miozzo, 2012; Buchwald, cessing, consists in the fact that it does not account for the Rapp & Stone, 2007; Buckingham, 1986; Buckingham & variability which is systematically revealed during the con- Christman, 2008; Canter, Trost & Burns, 1985; Lecours & version of higher level representations into overt speech. Lhermitte, 1969; Nespoulous, Joanette, Ska, Caplan & Le- An opposed view, embraced by connectionist models cours, 1987; Nespoulous, Lecours & Joanette, 1983; Tesak (e.g. Dell, Chang & Griffin, 1999; Dell, Juliano & Govind- & Code, 2008 among others). This objective was framed jee, 1993; Dell, Schwartz, Martin, Saffran & Gagnon, with ease within the modular framework of language encod- 1997; Dell & O’Seaghdha, 1992; Thomas & McClelland, ing precisely because of the modularity assumption but also 2008), envisages speech production as a network of ele- because of the apparent naturalness with which this frame- ments in interaction and a continuous processing space work accommodated the double dissociation procedure (but where gradient phenomena are easily accommodated but see Plaut, 1995; Shallice, 1988), thus allowing researchers to where specific functional components are not so easily iso- link selective impairments of a processing component with a lated. In contrast to serial architectures, connectionist mod- particular clinical syndrome. A classical hypothesis in els allow information activated at former stages of encod- aphasiology oriented by this viewpoint has opposed a pho- ing to influence processing at later levels. An important netic deficit, attributed to Broca’s aphasia and apraxia of methodological consequence of such a conceptualization speech, to a phonological deficit in conduction aphasia (e.g. of information flow is that the speech output may contain Alajouanine, Ombredane & Durand, 1939; Béland & Val- traces of information partially activated and streamlined dois, 1989; Lecours & Lhermitte, 1969; Valdois & Nespou- from an earlier to later stage, such as phonological charac- lous, 1998). Evidence to support or reject this claim came teristics of words similar in form to the target word that the mainly from two kinds of studies, on the one hand perceptu- speaker intends to produce. Within the connectionist al studies that attempted to identify the factors which pro- framework, phonological and phonetic domains are thus mote or constrain a given error type (Laganaro & Alario, conceived of as partly overlapping and inextricably inter- 2006; Laganaro & Zimmermann, 2010; Romani, Olson, Se- twined such that clear-cut boundaries do not exist. Howev- menza & Granà, 2002 among others), and on the other hand er, while these models have generated specific hypotheses from instrumental—acoustic and articulatory—studies that concerning phonological encoding, less attention has been sought to discern the source of disruptions on the basis of devoted to the subsequent and ultimate phonetic stage and fine-grained descriptions of errors presenting a similar sur- the exact mechanisms whereby symbolic phonological rep- face form (e.g. Baqué, Marczyk, Rosas & Estrada, 2015; resentations are transformed into articulatory gestures. Baum, Blumstein, Naeser & Palumbo, 1990; Baum & Slat- It is within these contrasting frameworks that re- kovsky, 1993; Blumstein et al., 1980; Buckingham & Yule, searchers have tried to elucidate the underlying nature 1987; Nespoulous, Baqué, Rosas, Marczyk & Estrada, 2013; of errors affecting the sound shape of words—substitu- Pouplier & Hardcastle, 2005; Tuller & Seider Story, 1988). tions, deletions, insertions and metatheses—commonly While a number of studies reported evidence support- referred to in the aphasic bibliography as phonemic par- ing the classical claim, other studies, in particular instru- aphasias (see Buckingham, 1992 for a review). Errors mental investigations, reported findings that forced re- Loquens, 2(2), July 2015, e023. eISSN 2386-2637 doi: http://dx.doi.org/10.3989/loquens.2015.023 Predicting segmental substitution errors in aphasic patients with phonological and phonetic encoding impairments • 3 searchers to reconsider the phonetic vs. phonological porting a modular and hierarchical structure of specific as- impairments dichotomy. pects of motor control (see Kent, Kent & Weismer, 2000 One such set of findings concerns the phenomenon re- for a review). An examination of speech production archi- ported first by Blumstein et al. (1980) and thereafter termed tectures relative to word-form encoding processes across a subtle phonetic deficit. Subtle phonetic deficit (see Vijayan the seriality/interactivity dimension may thus give further & Gandour, 1995 for a review) refers to the presence of pho- insights into how speech is produced. netic irregularities revealed by acoustic analyses in the out- Secondly, the notion that phonetic irregularities are put of aphasic patients with, presumably, phonological but interpreted as traces of lexical competitors—usually not phonetic deficit (i.e., Wernicke and conduction aphasia). lexemes differing from the target word by one interpho- The irregularities observed concern abnormal distributions nemic distance—is a matter of hypothesis, which in turn of values for the acoustic parameters of both consonants (see depends on the model that guides the research question. Vijayan & Gandour, 1995 for references) and vowels (e.g. Thus, the traces hypothesis is based on two underlying Baqué, 2015; Ryalls, 1986), such as longer segmental dura- assumptions, first, that words compete for selection and, tions, instability of production, narrower vocalic space and a second, that competitors are further processed at later tendency towards more opening in vowels, or overlapping stages of encoding in parallel with the target item, and distributions for voiced and voiceless stops VOT values, thus influence their phonetic characteristics. which are not registered by the listener’s perception. These Regarding the first assumption, previous research car- findings challenge the classical explanation in the sense that ried out within the connectionist framework has shown they seem to suggest either that fluent aphasic subjects (Wer- that the structure of the lexicon, including the number of nicke, conduction aphasia) exhibit a concomitant phonetic phonological competitors and their respective potential to disorder or that the phonetic irregularities observed result be selected, that is, their frequency of occurrence, facili- from a disruption at a higher processing level and are carried tates lexical access and production accuracy in both over to speech output thanks to interaction between the do- healthy subjects and aphasic speakers (e.g. Dell & Gor- mains, against the predictions of the seriality assumption. don, 2003; Goldrick, Folk & Rapp, 2010; Gordon, 2002; Indeed, when reargued from the connectionist per- Gordon & Dell, 2001; Vitevitch, 2002). Since the lexicon spective, phonetic deviations encountered in the speech is language-specific, and morphological structures in of fluent aphasic patients have been hypothesised to re- English and Spanish differ on a number of points, more flect traces of the activated but unselected neighbours of research is needed to test this hypothesis cross-linguisti- the target word (Kurowski & Blumstein, 2016). When the cally. Indeed, Vitevitch and Stamer (2006) found no pho- phoneme belonging to a competitor is activated, the in- nological neighbourhood size effect on response times in formation specifying its subphonemic features is sent a picture-naming task in Spanish speakers and attributed down the system and processed further. This parallel pro- this difference to the differing morphological structure of cessing results in the irregularities captured by acoustic these two languages. To our best knowledge, the effects analyses. Similar patterns have been also observed in nor- of neighbourhood characteristics on aphasic paraphasias mal speakers using the tongue-twisters paradigm (Gol- produced by Spanish speakers have not been yet exam- drick & Blumstein, 2006; Pouplier, Marin & Waltl, 2014). ined. Moreover, neighbourhood factors may exert an in- The connectionist interpretation of observed acoustic fluence on speech encoding beyond frequency effects. patterns raises several interesting questions and suggests Both connectionist and serial models account for frequen- promising research avenues. The first of them concerns in- cy of occurrence effects. In serial frameworks, frequency teraction. Even within connectionist accounts of speech effects are epistemologically related to the processes of production, the exact scope of interaction is still an unset- information retrieval and are not limited to lexeme fre- tled matter. The controversies concern the question wheth- quency but include as well specific hypotheses regarding er interaction influences only adjacent or also distant do- syllabic frequency (e.g. Cholin, Dell & Levelt, 2011; Je- mains—for example, lexical to articulatory or phonological scheniak & Levelt, 1994; Levelt & Wheeldon, 1994). to articulatory processes—or whether it operates exclu- The second assumption concerns the interaction between sively forward or also backward. An assessment through the lexical and phonetic processing. Since the competitors, computational simulation of speech architectures ranging as mentioned above, differ from the target word in one seg- along a serial/interactive dimension for their capacity to ac- ment, usually at one interphonemic distance, the hypothesis count for speech errors (Rapp & Goldrick, 2000) concludes is that the intermediary phonetic realizations of the seg- that these are best accounted for by cascading models with ment—i.e., neither [s] and nor [z] but something in between the interaction limited to semantic-phonological domains. (Kurowski & Blumstein, 2016)—are due to the influence of Thus, on the one hand, the hypothesis that competition at the competitor. However, it is possible to imagine that the the lexical and phonological levels exerts influence on ar- traces may be due to other factors and influences, whether ticulatory processes needs to be substantiated by more re- cognitive variables such as memory, general cognitive effort, search studying involving not only speakers with speech error awareness and errors monitoring, or phonetic factors pathology but also healthy. On the other hand, the exist- such as articulatory tension. While such questions may be ence of co-activation at certain levels does not necessarily difficult to translate into experimental designs, they are theo- rule out the serial organization of certain processes occur- retically possible and could contribute to a better under- ring at the motor level. Indeed, there is some evidence sup- standing of the mechanisms involved in speech production. Loquens, 2(2), July 2015, e023. eISSN 2386-2637 doi: http://dx.doi.org/10.3989/loquens.2015.023 4 • Anna Marczyk and Lorraine Baqué This state-of-the-art-summary shows that at present creation and were attributed to the lexical items of the there does not exist a model that can encompass the multi- stimuli lists at a later stage. This includes the factors de- ple dimensions of phonemic paraphasias analyses. The pre- scribing the neighbourhood characteristics of the target sent study does not claim to fill this gap but rather is merely item such as neighbourhood density (henceforth PTHN) intended to contribute to the existing body of evidence, and neighbourhood frequency mean (henceforth PTHF), taking as a starting point some of the questions that have as well as its frequency aspects, including word/lexeme been raised by previous studies. Specifically, it seeks to frequency, syllabic frequency and phonemic frequency. identify the predictors of segmental substitution errors elic- PTHN is defined as the number of lexical entries in a ited in reading and repetition tasks, which examine speech given lexicon that result from a deletion, addition or substi- production accuracy at a post-lexical level. The substitu- tution of a segment with regards to the target item (Vite- tions we are concerned with involve a change in the per- vitch, 1997). Words can belong to either sparse or dense ception of voicing or manner of articulation category. Thus, neighbourhoods. For example, a lexical entry cava can have our primary research goal is to explore the relationship be- various phonological neighbours including casa, cara, tween segmental substitution errors and the following: cama, cabo, daba, caza, calla, acaba, cabra, cavar, clava, etc. and is said to belong to a dense neighbourhood, in con- • Factors related to frequency including word and trast to a word like flama which has only a few phonologi- syllable frequency and neighbourhood size and cal neighbours like llama, flaca, lama or fama and thus is neighbourhood frequency mean. said to belong to a sparse neighbourhood. According to cas- • Phonetic and phonological factors including lexical cading models of speech production, the neighbours of the stress, phonetic context, phonological category of “correct” lexical item are hypothetically co-activated during the substituted sound and position in the word. encoding. Moreover, the coactivation strength of competi- • How the above factors and potential differences in tors may vary depending on the frequency of occurrence of how they influence errors depending on the aphasic each item. It has been hypothesised that not only the target profile. word frequency but also the frequency of its competitors can influence speech production. In particular, it has been argued that the higher the frequency of a given neighbour, 2. METHOD the stronger its chances of being selected against the target less-frequent lexical entry, which offers a plausible explana- 2.1. Stimuli tion for why a less frequent word like cava is likely to be replaced by a more frequent lexical item like casa. This ef- 2.1.1.  Criteria for stimuli preparation fect is measured by PTHF, defined as the average number of occurrences of all the words in the neighbourhood of the The errors we study here were elicited using a corpus target lexical item (Vitevitch, 1997). compiled by Baqué et al. (2008). The principal objective For the majority of items in our stimulus list, the values of this corpus was to examine errors that affect the word for both neighbourhood variables were obtained from an at the level of the segment and to test them for phonetic on-line cross-linguistic database for phonological and or- variables. Thus, the stimuli list consisted of a set of 240 thographic neighbourhood information (CLEARPOND: words which contained target consonants /p, t, k, b, d, ɡ, Marian, Bartolotti, Chabal & Shook, 2012). This resource f, s, θ, x, ùʃ/ in different phonetic contexts, stress condi- used a film and TV series subtitle corpora consisting of tions and vowel environments. 39,935,628 words to extract neighbourhood characteristics The lexical items belonged to different lexical catego- of the contemporary Spanish lexicon (Cuetos, Glez-Nosti, ries, comprising nouns, verbs, adjectives and adverbs and Barbón & Brysbaert, 2011). including both inflected and uninflected forms (e.g., For words that were not found in the database, the gastar, suspira, gatas, etc.) as well as common and proper corresponding PTHN and PTHF were calculated by the nouns (e.g., Zamora, trabajo, etc.). The set contained two software automatically. monosyllabic items (boj, faz), four tetrasyllabic words Apart from neighbourhood characteristics, this study (practicante, brazalete, dramático, administrar) and one is also concerned with determining the effects of lexical pentasyllabic word (desfavorable). The remaining 233 item frequency and syllabic frequency. The word fre- stimuli were either bisyllabic or trisyllabic. quency values for all but 44 items on our list (82% of the total) were obtained from a database of 81,323 written to- kens by Alameda and Cuetos (1995), available online. 2.1.2. Frequency and neighbourhood aspects of the The obtained values for the stimuli set ranged from 1 to stimuli 702, with a mean value of 47.06 (SD1 = 101.40), the 44 items were coded for this variable as NA. Some of the variables which we sought to assess in The syllabic frequency values were obtained from the the study were not taken into account during the stimuli online resource SYLLABARIUM by Duñabeitia, Cholin, 1   High values of SD indicate that word and syllable frequencies are highly skewed, as is habitual in natural languages (Baayen, 2001). Loquens, 2(2), July 2015, e023. eISSN 2386-2637 doi: http://dx.doi.org/10.3989/loquens.2015.023 Predicting segmental substitution errors in aphasic patients with phonological and phonetic encoding impairments • 5 Corral, Perea and Carreiras (2010), based on B-Pal lexi- fied inclusion criteria, such as presence of phonemic con by Davis and Perea (Davis & Perea, 2005). The val- paraphasias compatible with a Broca’s or conduction ues for this variable in our dataset ranged from 1 to 2701, aphasia profile, intact comprehension and absence of with a mean value of 547.84 (SD = 775.25). executive function alterations or visual and hearing im- Figure 1 represents schematically all the factors taken into pairments. All subjects were native speakers of Spanish account in the error analyses carried out in the present study. or bilingual Spanish-Catalan speakers. Out of 10 pa- As depicted in the chart, a given phoneme occupies a tients examined, 8 met the above criteria and were fur- specific position in a word, it may pertain to either frequent ther assessed using the MTBA-BCN aphasia battery or infrequent syllables, either stressed or unstressed, and (Baqué et al., 2006), which confirmed the initial diag- may be embedded in a frequent or infrequent word, or a nosis of the speech therapist. On the basis of the diag- word belonging to a sparse or dense neighbourhood, with nostic information gathered, the final population was competitors characterised by higher individual frequencies grouped into two clinical categories consisting of four or not. To take an example, /p/ is word-initial in the noun Broca’s aphasic subjects and four conduction aphasic piso ‘flat’ and verbs pisó ‘(s/he) stepped’ and parar ‘to subjects. Table 1 provides detailed information about stop’, and can be embedded in either a stressed (/ˈpiso/) or the subjects’ characteristics. The exact scores for sub- unstressed syllable (/piˈso/, /paˈɾaɾ/), one syllable being jects’ performance in aphasia assessment tasks as well more frequent (/pa/) than another (/pi/), all of them embed- as their linguistic description can be found in the ded in words characterised by different individual frequen- Appendix. cies of occurrence and pertaining to varying phonological neighbourhoods (e.g., 19 phonological neighbours in the case of piso and pisó and 11 in the case of parar). 2.3. Procedure 2.3.1.  Error elicitation 2.2. Subjects The lexical items containing target consonants were All the aphasic subjects preselected for this study randomised and placed at the end of short declarative sen- were recruited from the Hospital Universitari de Bell- tences. The list of sentences was printed in 16-point Arial vitge (Barcelona, Spain) according to a set of prespeci- type with 2.0 line spacing, with the target item in bold- face (e.g., Esta película es un drama ‘This film is a dra- ma’), and presented to the subjects in two experimental Figure 1: Variables taken into account for any given phoneme conditions, as either a reading or a repetition task. This in the substitution error corpus. choice was motivated by the fact that the goal of this study was to elicit errors that arise at the post-lexical level of speech encoding. Subjects were recorded either read- ing or repeating the stimulus sentences using a Sony ICD- CX50 visual voice recorder and a Sony high quality mi- crophone in a soundproof room at the Hospital Universitari de Bellvitge. These tasks were carried out in 4–6 sessions for each subject over a period of 3–4 weeks. The sessions were independent of the speech-language therapy sessions that the subjects were receiving during the study period. Table 1: Summary of the information regarding the aphasic subjects examined in this study. Months post Subjet Sex Age at testing Aphasia type Etiology Lesion site onset FNG M 68 Broca  6 CVA frontoparietal, left hemisphere FRG F 61 Broca  8 CVA left middle cerebral artery (MCA) MFB F 45 Broca  6 CVA multiple supratentorial lesions JLLV M 40 Broca 36 CVA left MCA CPB M 48 Conduction 17 CVA subdural parietal hematoma, left hemisphere LFC M 57 Conduction  8 Tumor glioblastoma multiforme, left temporoparietal lenticular region, left MCA, bypassing the JMC M 50 Conduction  7 CVA insular and parietal region JAOF M 57 Conduction  1 CVA left MCA Loquens, 2(2), July 2015, e023. eISSN 2386-2637 doi: http://dx.doi.org/10.3989/loquens.2015.023 6 • Anna Marczyk and Lorraine Baqué 2.3.2.  Error coding Figure 2: Substitution errors related to the voicing feature by pathology. Audio recordings of all productions thus obtained were coded according to the subject, task and stimulus number. Every target consonant was identified either as a correct realization of given phoneme (e.g., /k/ in kilo  à /k/ à CORRECT) or as a perceived substitu- tion where the replacement consonant pertained to one of the preselected phoneme categories (e.g., /ɡ/ in gata à /k/ à ERROR). All the substitutions where the replacing consonant was different from the target set of consonants (e.g., initial /b/ in baba replaced by /m/) as well all the other types of errors were excluded from the present study. Most of the substitutions led to the creation of non- words (lexical substitutions where the error and the target were semantically related were removed from further anal- yses). The transcription was carried out by the first author of this paper. In order to validate this classification, two in- With regard to manner of articulation, two opposite dependent native speakers of Spanish were presented with tendencies were identified, a tendency to occlusivise an identification test consisting of randomly chosen items voiceless fricatives (N = 136) and a tendency to spirantise from the data set, including both errors and correct produc- stops (N = 89). However, Broca’s and conduction aphasic tions, corresponding to 1% (N = 78) of the total number of groups showed different patterns of errors. In Broca’s stimuli, in conditions that closely resembled the original group there was a dominant tendency to occlusivise transcription. Cohen’s Kappa was used to assess the inter- voiceless fricatives, accounting for 25.8% (N = 119) of all rater reliability in SPSS (version 21 for Windows). The in- substitutions, while these errors accounted for 8% ter-judge reliability was 94.2% (k  = 0.033) between the (N  = 17) of all substitutions in the conduction aphasic author and first rater and complete (k = 1.00) between the group. Conversely, the spirantisation errors accounted for author and second rater. 22.1% (N  = 47) of all substitutions in the conduction aphasic group and 9.1% (N = 42) in the Broca’s aphasic group. This pattern is depicted in Figure 3. 3. RESULTS Finally, all the groups of errors were significantly influ- enced by phonetic context. Chi square and Fisher tests 3.1.  Identifying preferential tendencies were performed to examine the relation between phonetic context and different kind of errors. The relation between Out of 4824 consonants obtained in both reading and these variables was significant in all cases and indicated repetition tasks by the aphasic population studied, we ob- that the devoicing, occlusivisation and spirantisation errors tained 675 productions that could be regarded as substitu- were not equally distributed across all phonetic contexts. tions for the target sound. Substitutions represented the For devoicing errors, as can be seen in Figure 4, we most frequent error type and accounted for 14% of all the observe more errors in the word-initial position than in any production and 60% of all the segmental errors. This cat- other phonetic context (χ2(4, N = 1068) = 64.88, p < .001). egory was followed by segmental deletions (158 errors), distortions (72 errors), contextual substitutions (49 er- rors) and segmental additions (14 errors). The remaining Figure 3: Substitution errors related to the feature of manner errors were classified into one of the following catego- of articulation by pathology: O = stop, F = fricative, ries: lexical substitutions, multiple errors, metathesis er- A = affricate, where O‑>F stands for spirantisation; F‑>O rors or “other”. The analysis of errors belonging to one of occlusivisation, etc. the latter groups is beyond the scope of the present paper. An initial analysis of substitution errors was carried out to detect preferential tendencies related to the phono- logical features of voicing and manner of articulation. As illustrated in Figure 2 (all bar graphs were done with sj- Plot, Lüdecke, 2015), we observed a strong tendency to- wards devoicing of voiced stops in both aphasic groups. Devoicing errors accounted for 53% (N = 245) of all sub- stitutions in Broca’s group and 48% (N = 104) in the con- duction aphasic group. Sonorization errors were rare in both groups, accounting for less than 5% of all substitu- tions (19 and 10 errors of this type in the Broca’s and con- duction groups respectively). Loquens, 2(2), July 2015, e023. eISSN 2386-2637 doi: http://dx.doi.org/10.3989/loquens.2015.023 Predicting segmental substitution errors in aphasic patients with phonological and phonetic encoding impairments • 7 As depicted in Figure 5, occlusivisation errors occurred 3.2.  Predicting substitution errors more frequently in phonetic contexts that correspond to the syllabic onset (word-initial, intersonorant and post-conso- The regression analyses reported below take as a nantal positions), where word-initial position as well as in starting point the results of the descriptive analyses post-consonantal position seems to be particularly vulnera- summarised in the previous paragraphs. Substitution ble to this type of error (χ2(4, N = 1130) = 30.87, p < .001). error is a broad, non-homogeneous category as sug- Concerning spirantisation errors, the majority of them gested by the presence of preferential tendencies. were observed in syllabic coda (χ2(4, N = 2518) = 84.84, Hence, the following analyses will focus on the three Fisher p < .001), as illustrated in Figure 6. major types of substitution errors revealed by earlier examination: devoicing, occlusivisation and spiranti- sation errors. A common characteristic of all the error Figure 4: Percentage of devoicing errors in voiced stops types is their sensitivity to phonetic context and syl- according to the phonetic context. labic structure. Clearly, for devoicing errors the word- initial position is more error-prone in comparison to other phonetic contexts. On the other hand, it is the po- sition in the syllabic structure, onset vs. coda, which seems to trigger different rates of occlusivisation and spirantisation errors. Therefore, for the analyses re- ported below, we will examine the factors predicting devoicing errors in two phonetic contexts, word-initial and non-initial position, while occlusivisation and spi- rantisation error categories will be controlled for their position in the syllable, and, when possible, also for the word position. 3.2.1.  Setting the model Figure 5: Percentage of occlusivisation errors in voiceless All statistical analyses reported in this paper were per- fricatives according to the phonetic context. formed with the R Studio interface (RStudio Team, 2015) for R statistical software version 3.1.2 for PC (R Develop- ment Core Team, 2011). Several mixed-effects logistic re- gression models were conducted on three binary outcome variables related to the presence of different types of sub- stitution error in different phonetic environments. Subjects and items were entered as random intercepts which cap- tures our data, where both main factors and their interac- tions are within-subject and within-item. Initial models in- cluded the following predictor variables: pathology, lexical stress, task, PTHN, PTHF, lexeme frequency, syllabic fre- quency, type of consonant and, depending on the outcome variable studied, phonetic context and syllable position. Before entering them into the analysis, all predictors were first checked for collinearity by calculating the variable in- Figure 6: Percentage of spirantisation errors in voiceless flation factor (VIF) for each of them (function available in fricatives according to the phonetic context. the package car, Fox & Weisberg, 2011). All the VIF val- ues tested for each model fell below 2.0, thus confirming the collinearity assumption. 3.2.2.  Rationale for interactions Apart from the main effects, we sought to assess, where possible, the effects of interactions between the pathology and the rest of the predictors. The hypothesis that the different underlying error-generating mecha- nisms are a function of the clinical picture of aphasia would be supported if the interaction between the pa- thology and the other predictor proved to be statistically Loquens, 2(2), July 2015, e023. eISSN 2386-2637 doi: http://dx.doi.org/10.3989/loquens.2015.023 8 • Anna Marczyk and Lorraine Baqué significant. Our second hypothesis on the nature of the labic frequency, lexeme frequency, nor any interaction be- underlying deficit in fluent vs. non-fluent aphasic group tween the pathology and other predictors reached the sig- would be supported if post-hoc pairwise analyses of nificance level. simple effects went in the expected direction, that is, if The results obtained with the lsmeans function show (1) no effect of “higher level encoding” variables on the that the odds ratio of a devoicing error in the word-ini- probability of substitution errors was observed in the tial context is 18.66 times lower for conduction aphasia Broca’s aphasia group (2) no effect of “low-level encod- in comparison to Broca’s group. Moreover, it is 2.45 ing” variables on the outcome variable was observed in times lower (p = .000) for consonants in stressed sylla- the conduction aphasic group. bles vs. unstressed ones (the probability of observing a devoicing errors being 43% and 24% in unstressed and stressed syllables respectively). Finally, the stops /b/ 3.2.3.  Predictors of devoicing errors and /ɡ/ are more sensitive to devoicing than /d/, the re- spective probabilities of error for each of these conso- Two analyses were conducted to identify predic- nants being 34%, 48% and 18%. The results of the final tors of devoicing errors in initial and non-initial con- model are listed in Table 3. texts. A devoicing error may occur only on a voiced Secondly, the logistic regression model for devoicing stop (the phonological contrast of voicing in Spanish errors in non-initial contexts, including the same set of is limited to stops), hence, for the analyses reported predictors as above, showed an effect of PTHN on the below we retained all voiced stop consonants /b, d, ɡ/ probability of observing a devoicing error, independently (N = 520). Table 2 shows the confusion matrix for de- of the pathology group (χ2(1) = 4.84, p = .028). The result voicing errors in both groups and across all phonetic indicates that PTHN has a facilitating effect on accuracy. contexts. In other words, the probability of a devoicing error is 1.6 The first logistic regression model was intended to times lower in words pertaining to denser neighbour- identify factors predicting devoicing errors produced on hoods vs. sparse neighbourhoods in both groups. The word-initial consonants. The following factors were can- likelihood of observing a devoicing error is also influ- didates for predicting devoicing errors: lexical stress, enced by the type of consonant (χ2(2) = 15.13, p = .000). PTHN and PTHF, lexeme frequency, syllabic frequency, Analyses revealed that the probabilities of observing a consonant type, task and pathology, as well as their inter- devoicing error are 11%, 5% and 19% for /b/, /d/ and /ɡ/ actions with pathology. respectively. The difference between the error probabili- We identified the following explanatory variables pre- ties of /d/ and /ɡ/ is statistically significant (p  =  .000), dicting devoicing errors: pathology (χ2(1) = 6.67, p = .010), while other contrasts do not reach the alpha level set at lexical stress (χ 2(1) = 13.48, p = .000) and consonant type 0.05. Moreover, we observe a significant interaction be- (χ2(2) = 16.92, p = .000). Neither task, PTHN, PTHF, syl- tween pathology and syllabic frequency (χ2(1) = 7.57, p = .006), while syllabic frequency and pathology as main effects did not reach the significance threshold. The ex- Table 2: Confusion matrix for devoicing errors for both amination of this interaction reveals a different pattern for Broca’s and conduction aphasic subjects. the Broca’s and conduction aphasic group. In the Broca’s Perceived as: aphasic group the odds ratio of observing a devoicing er- Target ror are 1.18 times higher for frequent syllables, while in consonant /p/ /t/ /k/ /f/ /s/ /θ/ /x/ the conduction aphasic group the odds ratio of observing /b/ 87  2   4 7 1 0  0 such an error is 3.61 times lower the higher the syllabic /d/  1 33   6 0 2 5  1 frequency. This difference is statistically significant (p = .006). The results of the final model are detailed in /g/  9  4 113 5 2 1 23 Table 4. Table 3: Summary of the final mixed logistic regression model predicting the devoicing errors in word-initial stops. Predictors (reference category is given in italics): Pathology: Broca, conduction; Lexical stress: unstressed syllable, stressed syllable; Consonant type: /b/, /d/, /ɡ/. Predictor b SE z value (df) χ2 p-values* Pathology -2.93   1.13 -2.58  (1) = 6.67 0.010 Lexical stress -0.91  0.25  -3.67  (1) = 13.48 0.000 Consonant type (2) = 16.92 0.000  /d/ -0.83 0.36 -2.39  /g/ 0.58 0.27 2.18 *(Type II Wald Chi-square tests). Loquens, 2(2), July 2015, e023. eISSN 2386-2637 doi: http://dx.doi.org/10.3989/loquens.2015.023 Predicting segmental substitution errors in aphasic patients with phonological and phonetic encoding impairments • 9 Table 4: Summary of the final mixed logistic regression model predicting the devoicing errors in non-initial stops. Predictors (reference category is given in italics): Pathology: Broca, conduction; PTHN: numerical; Consonant type: /b/, /d/, /ɡ/; Syllabic frequency: numerical. Predictor b SE z value (df) χ2 p-values* Pathology -1.18 0.84 -1.41 (1) = 0.64 NS PTHN -0.47 0.21 -2.27 (1) = 5.17 0.023 Consonant type (2) = 15.13 0.000  /d/ -0.97 0.45 -2.14  /g/ 0.60 0.43 1.38 Syllabic frequency -0.93 0.24 -3.83 (1) = 14.67 NS Pathology* -1.45 0.53 -2.75 (1) = 7.57 0.006 syllabic frequency *(Type II Wald Chi-square tests). 3.2.4.  Predictors of occlusivisation errors level set at 0.05, however it approached this threshold (χ2(1) = 3.69, p = .055). In order to analyse occlusivisation errors we selected all Results indicate the lowest probability of occlusivisation the voiceless fricative of Spanish. We reported above that the error for alveolar /s/ (3%) and the highest for velar fricative rate of occlusivisation errors was significantly higher in the /x/ (28%), followed by labiodental fricative /f/ (17%), the Broca’s aphasic group than in the conduction aphasic group odds ratio for /f/ with respect to /s/ is 7.18 times higher and moreover, that such errors occurred almost exclusively (p = .004) while the odds ratio for /s/ with respect to /x/ is in the syllabic onsets. Hence, for the analyses reported below 13.02 lower (p = .001). All the other contrasts between con- we retained all the fricatives produced by the Broca’s group sonants groups did not reach the significance level. in any of the three phonetic contexts—word-initial, interson- Finally, the probability of observing an error of this orant and post-consonantal—corresponding always to the type is the highest in the word-initial position (17%), this syllabic onset. The confusion matrix for occlusivisation er- context being followed by the post-consonantal position rors in the Broca’s group in presented in Table 5. (15%) and it is the lowest in the intersonorant position The following predictors, which satisfied the non-colline- (6%). The odds ratio increases 3.10 times for the word-ini- arity requirement, were entered in the model: task, neighbour- tial position with respect to the intersonorant one (p = .017), hood frequency mean, word frequency, syllabic frequency, and it decreases 2.29 times for the intersonorant with re- consonant type (/f/, /s/, /θ/, /x/), phonetic context (only syllabic spect to the post-consonantal position (p = .048). onset in the initial, intersonorant and post-consonantal posi- Furthermore, the test revealed that the probability of tion) and lexical stress (stressed and unstressed syllable). occlusivisation errors increases (the odds ratio is 37.31 The PTHN variable was removed from for the predic- times higher, p = .055) for fricatives in stressed syllables tor candidates list since its VIF factor exceeded the than in unstressed ones (the probability of errors being threshold value 2.0. 9% and 15% in unstressed and stressed syllable respec- The analyses of occlusivisation errors in syllabic on- tively), although, as mentioned, this effect failed to reach set positions in the Broca’s group identified the effect of the alpha level set at 0.05. consonant type (χ2(3) = 19.76, p = .000) and the effect of The results of the analyses are listed in Table 6. phonetic context (χ2(2) = 8.18, p = .017). Lexical stress did not predict the occlusivisation error at the significance 3.2.5.  Predictors of spirantisation errors Table 5: Confusion matrix for occlusivisation errors in Broca’s Spirantisation errors were observed in both the Bro- aphasic group. ca’s and the conduction aphasic group. Table 7 presents the confusion matrix for this type of error, in both groups Target Perceived as: and across all phonetic contexts. consonant /p/ /t/ /k/ However, as can be seen from the data in Table 8, the dis- tribution of spirantisation errors in onset and coda positions in /f/ 28  3  5 the syllabic structure is highly influenced by the aphasic pro- /s/  0  2  1 file. Indeed, while in the conduction aphasic group we observe /θ/  1 12  3 more errors in the coda position as compared to the syllabic onset, in the Broca’s group there are almost no errors in coda /x/  0  0 29 and syllabic onset is much more error-prone. Loquens, 2(2), July 2015, e023. eISSN 2386-2637 doi: http://dx.doi.org/10.3989/loquens.2015.023 10 • Anna Marczyk and Lorraine Baqué Table 6: Summary of the final mixed logistic regression model predicting occlusivisation errors in syllabic onsets fricatives produced by Broca’s aphasic subjects. Predictors (reference category is given in italics): Lexical stress: unstressed syllable, stressed syllable; Phonetic context: word- initial, intersonorant, post-consonantal; Consonant type: /f/, /s/, /θ/, /x/. Predictor b SE z value (df) χ2 p-values* Lexical stress 0.57 0.30 1.92 (1) = 3.69 0.055 Phonetic context (2) = 8.18 0.017  intersonorant -1.13 0.41 -2.74  post-consonantal -1.13 0.36 -0.35 Consonant type (3) = 19.75 0.000  /s/ -1.97 0.58 -3.37   /θ/ -0.48 0.40 -1.20  /x/ 0.59 0.38 1.54 *(Type II Wald Chi-square tests). Table 7: Confusion matrix for spirantisation errors in both odds ratio decreases 12.85 times for Broca’s group in aphasic groups and across all contexts. comparison with conduction aphasic group (p = .000). The difference between these two groups with respect to Target Perceived as: the syllabic onset is not significant (the probability of a consonant /f/ /s/ /θ/ /x/ spirantisation error being 2% and 0.7% for Broca’s and /p/ 28  3  5 0 conduction aphasic group respectively). Furthermore, we observe an effect of syllabic frequency (χ2(1) = 5.23, /t/  0  2  1 5 p  =  .022), which is independent of the aphasic profile. /k/  1 12  3 4 The probability of spirantisation errors decreases for /b/  0  0 29 0 more frequent syllables, the odds ratio decreases 1.59 times for every unit of increase in syllabic frequency. Fi- nally, the spirantisation errors were significantly predict- ed by consonant type (χ2(5) = 21.48, p = .001). The velar Table 8: Distribution of spirantisation errors depending on stop /ɡ/ triggers a spirantisation error with the highest syllable position. probability in comparison to all other voiced or voiceless stops (this probability being of 6% for /ɡ/, against the Broca conduction probabilities lower than 3% for all the other stops). With Onset 36 16 respect to /ɡ/, the odds ratio decreases 3.52 times for Coda  2 26 voiceless bilabial /p/ (p = .007), 3.51 for voiceless velar /k/ (p = .012) and 4.30 times for voiced dental /d/ (p = .017), the least vulnerable for this type of error. Oth- er contrasts are not significant. The results are summa- Therefore, we collapsed the spirantisation errors rised in Table 9. across all the phonetic contexts, and conducted one logis- tic regression mixed model where we entered the position in the syllable structure (onset and coda) as predictor 3.2.6.  Summary of results along with other factors including: task, phonological neighbourhood size, neighbourhood frequency mean, syl- The goal of the analyses reported in the preceding labic frequency, word frequency, stress, consonant type paragraphs was to identify factors predicting three differ- and pathology. Since we hypothesise that the pathology ent kinds of substitution errors—devoicing, occlusivisa- may affect how these factors influence the likelihood of tion and spirantisation—observed in the aphasic speech the appearance of spirantisation errors, we included inter- output. We will summarize below the main findings ac- actions in the model. cording to the predictor candidate across all the tests. The results of the model confirm that the syllable po- First, the analyses reveal a strong effect of pathology sition predicts the errors differently depending on the across different contexts and error types. The clinical pic- aphasic profile (χ2(1) = 21.23, p = .000). The probability ture of aphasia seems to exert a quantitative and—albeit to of such errors happening in the syllabic coda in the con- a lesser extent—qualitative influence on the likelihood of duction aphasic group is 13%, against 2% in the Broca’s different categories of substitution errors. This is evident in group in the same position. For the syllabic coda, the the higher number of both devoicing and occlusivisation Loquens, 2(2), July 2015, e023. eISSN 2386-2637 doi: http://dx.doi.org/10.3989/loquens.2015.023 Predicting segmental substitution errors in aphasic patients with phonological and phonetic encoding impairments • 11 Table 9: Summary of the final mixed logistic regression model predicting spirantisation errors. Predictors (reference category is given in italics): Pathology: Broca, conduction; Syllabic frequency: numerical; Syllabic position: coda, onset; Consonant type: /p/, /t/, /k/, /b/, /d/, /ɡ/. Predictor b SE z value (df) χ2 p-values* Pathology 2.55 0.94 2.72 (1) = 0.86 NS Syllabic frequency -0.46 0.20 -2.28 (1) = 5.23 0.022 Syllabic position 0.60 0.74 0.80 (1) = 44.33 0.000 Consonant type (5) = 21.48 0.001  /t/ 42 0.50 0.85  /k/ 0 0.43 0.01  /b/ 0.18 0.48 0.38  /d/ -0.20 0.49 -0.40  /g/ 1.26 0.36 3.46 Pathology* -3.73 0.81 -4.06 (1) = 21.23 0.000 syllabic position *(Type II Wald Chi-square tests). errors in the Broca’s aphasic group as compared to the con- phonetic factors including not only the lexical stress but duction aphasic group, with occlusivisation errors practi- also syllabic position and phonetic context. The effect of cally absent in the latter. Of the three syllabic contexts ex- stress was not observed for spirantisation errors, a pro- amined, the word-initial position was the most susceptible cess involving articulatory weakening. to errors for both groups and both types of errors. This Regarding phonological neighbourhood effects, the finding is consistent with evidence reported in previous neighbourhood size measured in the number of phono- studies showing that non-fluent aphasic subjects have par- logically similar words was found to be facilitating for ticular difficulty with the initiation of speech (e.g. Code, accuracy of production in the case of devoicing non- 1998; Kent & Rosenbek, 1983). word-initial errors. This effect was not influenced by the Second, our results show that consonant type has a aphasic profile. This result is consistent with findings for consistent effect on the likelihood of substitution errors aphasic speech in English (Gordon, 2002 for lexical ac- across all the tests except for spirantisation errors. Over- cess; Kurowski & Blumstein, 2016 for errors related to all, the analyses suggest that the coronal fricative /s/ and voicing). To our best knowledge there is no study for stop /d/ are the least vulnerable to errors of any kind, Spanish on the effect of this variable on aphasic produc- whereas velar fricative /x/ and stop /ɡ/ trigger more er- tion, and while there are studies on its effect on lexical rors, with these errors being occlusivisation in the case of retrieval, we are not aware of any work on the influence /x/ and spirantisation in the case of /ɡ/. This result is inde- of neighbourhood size on post-lexical encoding. No ef- pendent of the aphasic profile, as suggested by the ab- fect of lexeme frequency or PTHF was found in any of sence of significant interactions between consonant type the analyses reported in this study. Lack of PTHF effect is and pathology group. Another frequent error type was oc- consistent with previous research on aphasic speech (e.g. clusivisation of syllable onset /f/ (/f/à/p/), observed in Gordon, 2002). As for the word frequency, this effect may the Broca’s group speech output. It should be noted that be overshadowed by a more robust neighbourhood size /f/ is the only possible fricative constituent of a complex effect or factors intervening at phonological and phonetic syllabic onset in Spanish (as in frasco ‘jar’). encoding such as consonant type, syllabic position or lex- Third, a contrasting pattern of results was observed ical stress. for lexical stress. Our results show that lexical stress has a Finally, we observed two kinds of interactions with facilitating effect for word-initial devoicing errors, more aphasic profile, one involving syllabic frequency and an- frequent in unstressed syllables in comparison with other syllabic position. In the case of non-initial devoic- stressed ones, independently of the aphasic profile (a sim- ing errors, the conduction aphasic group proved more ilar effect was found for omission errors by Nickels & sensitive to syllabic frequency than the Broca’s group, in Howard, 1999). On the other hand, for the analyses per- the sense that conduction aphasic patients produced sig- formed on occlusivisation errors in the Broca’s aphasic nificantly fewer errors in frequent syllables as opposed to group, stressed syllables were found to trigger more er- infrequent. These results reinforce the hypothesis that rors than unstressed ones. This finding may suggest that phonetic encoding follows a dual route, where one of the occlusivisation errors in this clinical group may originate routes consists in accessing ready-made motor programs from a lack of an appropriate level of articulatory tension, for frequent syllables stored in a syllabary at the interface especially when the articulatory tension is reinforced by between phonological and phonetic encoding, and the Loquens, 2(2), July 2015, e023. eISSN 2386-2637 doi: http://dx.doi.org/10.3989/loquens.2015.023 12 • Anna Marczyk and Lorraine Baqué second and indirect route involves on-line assembling post-lexical encoding levels, including the nature of motor programs for infrequent or new syllables (Levelt & ‘acoustic traces’. Wheeldon, 1994). Moreover, the hypothesis proposed by The second conclusion is related to the existence of Varley and Whiteside (1998, 2001), which posits a loss of common effects, either promoting or constraining errors, access to the mental syllabary in apraxia of speech (Bro- which are independent of aphasic profiles. Our findings ca’s aphasia), seems congruent with our findings for lend support to the hypothesis, embraced by connection- Spanish-speaking aphasic subjects. Interestingly, howev- ists, that errors are sensitive to some universal aspects of er, this result suggests that conduction aphasic patients speech production (see the classical study by Blumstein, have difficulty with a strictly phonetic task, namely, com- 1973). These commonalities play themselves out in the posing motor programmes for articulation on-line. existence of preferential tendencies and variable effects With respect to the syllabic position effect, we found in both aphasic groups and suggest that aphasic syn- that syllable onsets were overall more sensitive to errors dromes may be better described with the notion of ‘spec- in the Broca’s group. The difference between groups con- trum of disorders’ rather than clear-cut categories. cerned spirantisation errors, where we observed more er- Finally, our third conclusion tones down second one. rors in the coda position in conduction aphasia and in the Though our findings indicate that not only disfluent but onset position in Broca’s aphasia. While the pattern of er- also fluent aphasic patients exhibit a phonetic disorder, rors in the conduction aphasic group can be explained by they also suggest that these phonetic deficits are not en- articulatory weakening in the syllabic coda, consistent tirely similar in the two groups. Particular factors, such as with the phonological processes in normal speech, the difficulty in initiating speech or control of articulatory finding for the Broca’s group calls for an entirely differ- tension, may reinforce the phonetic impairment in Bro- ent interpretation, which again may be related to inappro- ca’s aphasia but play no role in conduction aphasia. Fi- priate articulatory tension. nally, different factors may promote accuracy depending on the clinical profile. Indeed, our results show that high frequency of syllables may facilitate correct production 4.  DISCUSSION of consonants in conduction aphasia but not in Broca’s aphasia. The evidence reported in this paper can be discussed in relation to two major issues disclosed in the introduc- tion. The first of them concerns the classical dichotomy of REFERENCES phonetic vs. phonological errors in aphasic speech and its relation to the clinical dichotomy of Broca’s and conduc- Alajouanine, T., Ombredane, A., & Durand, M. (1939). Le syndrome tion aphasia. The second concerns the seriality/interactiv- de désintégration phonétique dans l’aphasie. Paris: Masson. ity axis. The present investigation contributes to these Alameda, J., & Cuetos, F. (1995). Diccionario de frecuencias de las unidades lingüísticas del castellano. Oviedo: Servicio de Publi- debates by providing evidence from Spanish-speaking caciones de la Universidad de Oviedo. aphasic subjects. Baayen, H. R. (2001). Word frequency distributions. Dordrecht: Kluwer Three main conclusions can be drawn from the anal- Academic Publishers. http://dx.doi.org/10.1007/978-94-010-0844-0. yses reported above. First, processing at the phonetic Baqué, L. (2015). Velocidad de articulación y estructuración del es- pacio vocálico en función del acento en la afasia: un estudio and phonological levels of encoding, and consequently preliminar. In A. Cabedo Nebot (Ed.), Normas. Perspectivas any impairment of these processes, is influenced by actuales en el análisis fónico del habla: tradición y avances en higher stage processes, such as competition between la fonética experimental (Vol. 7, pp. 297–308). Retrieved from phonological neighbours, and this influence is similar— http://ampercan.webs.ull.es/sites/default/files/Libro_Foneti- although possibly robust to varying degrees—in Broca’s ca_2015.pdf. Baqué, L., Estrada, M., Nespoulous, J.‑L., Le Besnerais, M., Rosas, and conduction aphasia. More specifically, the neigh- A., & Marczyk, A. (2006). Protocole Initial du projet Cognifon. bourhood size plays a facilitating role and promotes ac- Barcelona. curacy in aphasic speech in patients with a phonological Baqué, L., Estrada, M., Nespoulous, J.‑L., Le Besnerais, M., Rosas, and phonetic level deficit. This result is consistent with A., & Marczyk, A. (2008). Corpus léxico del proyecto COGNI- FON. Barcelona [Unpublished technical document]. that reported by Kurowski and Blumstein (2016), a Baqué, L., Marczyk, A., Rosas, A., & Estrada, M. (2015). Disabili- study that asks very different research questions and is ty, repair strategies and communicative effectiveness at the also very different methodologically. Kurowski and phonic level: Evidence from a multiple-case study. In C. Astés- Blumstein (2016) examine the acoustic characteristics ano & M. Jucla (Eds.), Neuropsycholinguistic perspectives on of devoicing and voicing errors of fricative consonants language cognition (pp. 144–165). Cambridge: Routledge Tay- lor & Francis. /s/ and /z/ in English, while ours is a perceptual study Baum, S. R., Blumstein, S. E., Naeser, M. A., & Palumbo, C. L. aimed at identifying the predictors of productions per- (1990). Temporal dimensions of consonant and vowel pro- ceived as segmental substitutions. Thus, while our study duction: An acoustic and CT scan analysis of aphasic indicates that interactions between lexical and phonetic speech. Brain and Language, 39(1), 33–56. http://dx.doi. org/10.1016/0093-934X(90)90003-Y. level exist, it cannot confirm or reject the ‘traces hy- Baum, S. R., & Slatkovsky, K. (1993). Phonemic false evalua- pothesis’. A crossed instrumental and perceptual as well tion?: Preliminary data from a conduction aphasia patient. as cross-linguistic study on these effects could provide Clinical Linguistics & Phonetics, 7(3), 207–218. http://dx.doi. further insights into the interaction between lexical and org/10.3109/02699209308985558. Loquens, 2(2), July 2015, e023. eISSN 2386-2637 doi: http://dx.doi.org/10.3989/loquens.2015.023 Predicting segmental substitution errors in aphasic patients with phonological and phonetic encoding impairments • 13 Béland, R., & Valdois, S. (1989). Les perturbations phonétiques et Duñabeitia, J. A., Cholin, J., Corral, J., Perea, M., & Carreiras, M. phonémiques: nouvelles perspectives. Langages, 24e Année, (2010). SYLLABARIUM: An online application for deriving 96, 44–63. http://dx.doi.org/10.3406/lgge.1989.1558. complete statistics for Basque and Spanish orthographic sylla- Blumstein, S. E. (1973). A phonological investigation of aphasic bles. Behavior Research Methods, 42(1), 118–125. http:// speech. The Hague: Mouton. dx.doi.org/10.3758/BRM.42.1.118. Blumstein, S. E., Cooper, W. E., Goodglass, H., Statlender, S., & Fox, J., & Weisberg, S. (2011). An {R} Companion to Applied Re- Gottlieb, J. (1980). Production deficits in aphasia: A voice-on- gression (2nd ed.). Thousand Oaks, CA: Sage. set time analysis. Brain and Language, 9(2), 153–170. http:// Garrett, M. (1980). Levels of processing in sentence production. In dx.doi.org/10.1016/0093-934X(80)90137-6. B. Butterworth (Ed.), Language production Vol. 1: Speech and Buchwald, A., & Miozzo, M. (2012). Phonological and motor errors talk (pp. 177–220). in individuals with acquired impairment. Journal of Speech, Goldrick, M., & Blumstein, S. E. (2006). Cascading activation from Language, and Hearing Research, 55(5), S1573–S1586. http:// phonological planning to articulatory processes: Evidence from dx.doi.org/10.1044/1092-4388(2012/11-0200). tongue twisters. Language and Cognitive Processes, 21, 649– Buchwald, A., Rapp, B., & Stone, M. (2007). Insertion of discrete 683. http://dx.doi.org/10.1080/01690960500181332. phonological units: An ultrasound investigation of aphasic Goldrick, M., Folk, J. R., & Rapp, B. (2010). Mrs. Malaprop’s speech. Language and Cognitive Processes, 22(6), 910–948. neighborhood: Using word errors to reveal neighborhood struc- http://dx.doi.org/10.1080/01690960701273532. ture. Journal of Memory and Language, 62, 113–134. http:// Buckingham, H. W. (1986). The scan-copier mechanism and the po- dx.doi.org/10.1016/j.jml.2009.11.008. sitional level of language production: Evidence from phonemic Gordon, J. K. (2002). Phonological neighborhood effects in paraphasia. Cognitive Science, 10, 195–217. http://dx.doi. aphasic speech errors: Spontaneous and structured contexts. org/10.1207/s15516709cog1002_4. Brain and Language, 82, 113–145. http://dx.doi.org/10.1016/ Buckingham, H. W. (1992). The mechanisms of phonemic parapha- S0093-934X(02)00001-9. sia. Clinical Linguistics & Phonetics, 6(1–2), 41–63. http:// Gordon, J. K., & Dell, G. (2001). Phonological neighborhood ef- dx.doi.org/10.3109/02699209208985518. fects: Evidence from aphasia and connectionist modeling. Buckingham, H. W., & Christman, S. (2008). Disorders of phonet- Brain and Language, 79, 21–23. ics and phonology. In B. Stemmer & H. A. Whitaker (Eds.), Jescheniak, J. D., & Levelt, W. (1994). Word frequency effects Handbook of the Neuroscience of Language (pp. 127–136). in speech production: Retrieval of syntactic information and London: Academic Press Elsevier. http://dx.doi.org/10.1016/ of phonological form. Journal of Experimental Psychology: B978-0-08-045352-1.00012-4. Learning, Memory, and Cognition, 20(4), 824–843. http:// Buckingham, H. W., & Yule, G. (1987). Phonemic false evaluation: dx.doi.org/10.1037/0278-7393.20.4.824. Theoretical and clinical aspects. Clinical Linguistics & Phonetics, Kent, R. D., Kent, J. F., & Weismer, G. (2000). What dysarthrias 1(2), 113–125. http://dx.doi.org/10.3109/02699208708985007. can tell us about the neural control of speech. Journal of Pho- Canter, G. J., Trost, J. E., & Burns, M. S. (1985). Contrasting netics, 28, 273–302. http://dx.doi.org/10.1006/jpho.2000.0122. speech patterns in apraxia of speech and phonemic parapha- Kent, R. D., & Rosenbek, J. C. (1983). Acoustic patterns of apraxia sia. Brain and Language, 24(2), 204–222. http://dx.doi. of speech. Journal of Speech and Hearing Research, 26(2), org/10.1016/0093-934X(85)90131-2. 231–49. http://dx.doi.org/10.1044/jshr.2602.231. Cholin, J., Dell, G., & Levelt, W. (2011). Planning and articulation Kurowski, K., & Blumstein, S. E. (2016). Phonetic basis of phonemic in incremental word production: Syllable-frequency effects in paraphasias in aphasia: Evidence for cascading activation. Cor- English. Journal of Experimental Psychology: Learning, Mem- tex, 75, 193–203. http://dx.doi.org/10.1016/j.cortex.2015.12.005. ory, and Cognition, 37(1), 109–22. http://dx.doi.org/10.1037/ a0021322. Laganaro, M. (2015). Paraphasies phonémique et/ou phonétiques? Code, C. (1998). Models, theories and heuristics in apraxia of Des raisons et des difficultés de cette distinction. Revue de Neu- speech. Clinical Linguistics and Phonetics, 12(1), 47–65. http:// ropsychologie, 7, 27–32. http://dx.doi.org/10.3917/rne.071.0027. dx.doi.org/10.3109/02699209808985212. Laganaro, M., & Alario, F. X. (2006). On the locus of syllable fre- Cuetos, F., Glez-Nosti, M., Barbón, A., & Brysbaert, M. (2011). quency effect. Journal of Memory and Language, 55, 178–196. SUBTLEX-ESP: Spanish word frequencies based on film subti- http://dx.doi.org/10.1016/j.jml.2006.05.001. tles. Psicológica, 32, 133–143. Laganaro, M., & Zimmermann, C. (2010). Origin of phoneme substitu- Davis, C. J., & Perea, M. (2005). BuscaPalabras: A program for tion and phoneme movement errors in aphasia. Language Cognitive deriving orthographic and phonological neighborhood statis- Processes, 25, 1–37. http://dx.doi.org/10.1080/01690960902719259. tics and other psycholinguistic indices in Spanish. Behavior Lecours, A. R., & Lhermitte, F. (1969). Phonemic paraphasias: Lin- Research Methods, 37, 665–671. http://dx.doi.org/10.3758/ guistic structures and tentative hypotheses. Cortex, 5, 193–228. BF03192738. http://dx.doi.org/10.1016/S0010-9452(69)80031-6. Dell, G., Chang, F., & Griffin, Z. M. (1999). Connectionist models Levelt, W. (1989). Speaking: From intention to articulation. Cam- of language production: Lexical access and grammatical encod- bridge, MA: MIT Press. ing. Cognitive Science, 23, 517–542. http://dx.doi.org/10.1207/ Levelt, W. (1999). Models of word production. Trends in Cogni- s15516709cog2304_6. tive Sciences, 3(6), 223–232. http://dx.doi.org/10.1016/ Dell, G., & Gordon, J. K. (2003). Neighbors in the lexicon: Friends S1364-6613(99)01319-4. or foes? In N. O. Schiller & A. S. Meyer (Eds.), Phonetics and Levelt, W., & Wheeldon, L. R. (1994). Do speakers have access to a phonology in language comprehension and production: Differ- mental syllabary? Cognition, 50, 239–269. http://dx.doi. ences and similarities (pp. 9–38). New York: Mouton de org/10.1016/0010-0277(94)90030-2. Gruyter. http://dx.doi.org/10.1515/9783110895094.9. Lüdecke, D. (2015). sjPlot: Data Visualization for Statistics in So- Dell, G., Juliano, C., & Govindjee, A. (1993). Structure and content cial Science [software package]. Retrieved from https://mran. in language production: A theory of frame constraints in phono- microsoft.com/package/sjPlot/. logical speech errors. Cognitive Science, 17(2), 149–195. http:// Marian, V., Bartolotti, J., Chabal, S., & Shook, A. (2012). CLEAR- dx.doi.org/10.1207/s15516709cog1702_1. POND: Cross-Linguistic Easy-Access Resource for Phonologi- Dell, G., & O’Seaghdha, P. G. (1992). Stages of lexical access in cal and Orthographic Neighborhood Densities. PLoS ONE, language production. Cognition, 42, 287–314. http://dx.doi. 7(8), e43230. http://dx.doi.org/10.1371/journal.pone.0043230. org/10.1016/0010-0277(92)90046-K. Nespoulous, J.-L., Baqué, L., Rosas, A., Marczyk, A., & Estrada, Dell, G., Schwartz, M. F., Martin, N., Saffran, E. M., & Gagnon, M. (2013). Aphasia, phonological and phonetic voicing within D. A. (1997). Lexical access in aphasic and nonaphasic speak- the consonantal system: Preservation of phonological opposi- ers. Psychological Review, 104(4), 801–838. http://dx.doi. tions and compensatory strategies. Language Sciences, 39, org/10.1037/0033-295X.104.4.801. 117–125. http://dx.doi.org/10.1016/j.langsci.2013.02.015. Loquens, 2(2), July 2015, e023. eISSN 2386-2637 doi: http://dx.doi.org/10.3989/loquens.2015.023 14 • Anna Marczyk and Lorraine Baqué Nespoulous, J.-L., Joanette, Y., Ska, B., Caplan, J. D., & Lecours, Shallice, T. (1988). From neuropsychology to mental structure. A. R. (1987). Production deficits in Broca’s and conduction Cambridge, UK: Cambridge University Press. http://dx.doi. aphasia: Repetition versus reading. In E. Keller & M. Gopnik org/10.1017/CBO9780511526817. (Eds.), Motor and sensory processes of language (pp. 53–81). Tesak, J., & Code, C. (2008). Milestones in the history of aphasia: Hillsdale, NJ: Lawrence Erlbaum. Theories and protagonists. London: Psychology Press. Nespoulous, J.-L., Lecours, A. R., & Joanette, Y. (1983). La dichot- Thomas, M. S. C., & McClelland, J. L. (2008). Connectionist models omie phonétique/phonologique a-t-elle une valeur nosologique? of cognition. In R. Sun (Ed.), Cambridge handbook of computa- In P. Messerli, P. M. Lavorel & J.‑L. Nespoulous (Eds.), Neu- tional psychology (pp. 23–58). Cambridge, UK: Cambridge Uni- ropsychologie de l’expression orale (pp. 71–91). Paris: Editions versity Press. http://dx.doi.org/10.1017/CBO9780511816772.005. du C.N.R.S. Tuller, B., & Seider Story, R. (1988). Anticipatory and carryover coar- Nickels, L., & Howard, D. (1999). Effects of lexical stress on apha- ticulation in aphasia: An acoustic study. Cognitive Neuropsycholo- sic word production. Clinical Linguistics and Phonetics, 13, gy, 5(6), 747–771. http://dx.doi.org/10.1080/02643298808253281. 269–294. http://dx.doi.org/10.1080/026992099299086. Valdois, S., & Nespoulous, J.-L. (1998). Perturbation du traitement Plaut, D. C. (1995). Double dissociation without modularity: Evi- phonologique et phonétique du langage. In X. Séron & M. dence from connectionist neuropsychology. Journal of Clinical Jeannerod (Eds.), Traité de Neuropsychologie Humaine (pp. and Experimental Neuropsychology, 17(2), 291–321. http:// 360–374). Liège: Mardaga. dx.doi.org/10.1080/01688639508405124. Varley, R. A., & Whiteside, S. P. (1998). A reconceptualisation of Pouplier, M., & Hardcastle, J. W. (2005). A re-evaluation of the na- apraxia of speech: A synthesis of evidence. Cortex, 34, 221– ture of speech errors in normal and disordered speakers. Pho- 231. http://dx.doi.org/10.1016/S0010-9452(08)70749-4. netica, 62, 227–243. http://dx.doi.org/10.1159/000090100. Varley, R. A., & Whiteside, S. P. (2001). What is the underlying im- Pouplier, M., Marin, S., & Waltl, S. (2014). Voice onset time in con- pairment in acquired apraxia of speech. Aphasiology, 15, 39–49. sonant cluster errors: Can phonetic accommodation differenti- Vijayan, A., & Gandour, J. (1995). On the notion of a “subtle pho- ate cognitive from motor errors? Journal of Speech, Language, netic deficit” in fluent/posterior aphasia. Brain and Language, and Hearing Research, 57, 1577–1588. http://dx.doi. 48(2), 106–119. http://dx.doi.org/10.1006/brln.1995.1004. org/10.1044/2014_JSLHR-S-12-0412. Vitevitch, M. S. (1997). The neighborhood characteristics of mala- R Development Core Team. (2011). R: A Language and Environ- propisms. Language and Speech, 40(3), 211–228. http://dx.doi. ment for Statistical Computing. Vienna, Austria. Retrieved org/10.1121/1.415242. from http://www.r-project.org. Vitevitch, M. S. (2002). The influence of phonological similarity Rapp, B., & Goldrick, M. (2000). Discreteness and interactivity in neighborhoods on speech production. Journal of Experimental spoken word production. Psychological Review, 107(3), 460– Psychology: Learning, Memory, and Cognition, 28(4), 735– 499. http://dx.doi.org/10.1037/0033-295X.107.3.460. 747. http://dx.doi.org/10.1037/0278-7393.28.4.735. Romani, C., Olson, A., Semenza, C., & Granà, A. (2002). Patterns Vitevitch, M. S., & Stamer, M. K. (2006). The curious case of competi- of phonological errors as a function of a phonological versus an tion in Spanish speech production. Language and Cognitive Pro- articulatory locus of impairment. Cortex, 38(4), 541–567. cesses, 21, 760–770. http://dx.doi.org/10.1080/01690960500287196. http://dx.doi.org/10.1016/S0010-9452(08)70022-4. Ziegler, W. (2002). Psycholinguistic and motor theories of apraxia RStudio Team (2015). RStudio: Integrated Development Environ- of speech. Seminars in Speech and Language, 23, 231–243. ment for R. Boston, MA: RStudio, Inc. http://dx.doi.org/10.1055/s-2002-35798. Ryalls, J. (1986). An acoustic study of vowel production in Ziegler, W. (2008). Neurophonetics. In M. J. Ball, N. Müller, M. R. Per- aphasia. Brain and Language, 29, 48–67. http://dx.doi. kins & S. Howard (Eds.), Handbook of Clinical Linguistics. Oxford: org/10.1016/0093-934X(86)90033-7. Blackwell. http://dx.doi.org/10.1002/9781444301007.ch31. APPENDIX Table 1. Linguistic description of patients. Aphasia Native Permanent residence Permanent residence Bilingual Other Patient Age type tongue (>10 years) (>10 years) (catalan) languages FNG 68 B Spanish Barcelona NA yes Italian FRG 61 B Spanish Barcelona Alicante no NA MFB 45 B Spanish Barcelona NA yes English JLLV 40 B Catalan Barcelona NA yes English CPB 48 C Spanish Barcelona NA yes NA LFC 57 C Spanish Barcelona NA no NA JMC 50 C Spanish Barcelona NA no English JAOF 57 C Spanish Barcelona NA yes NA Loquens, 2(2), July 2015, e023. eISSN 2386-2637 doi: http://dx.doi.org/10.3989/loquens.2015.023 Table 2. Results of the language assessment. Key: APH.: Aphasia type, B=Broca, C=conduction; TpO: Months post onset; Et.: Etiology; G.I.: Guided interview; C.O.: Oral comprehension : C.Om.: Oral comprehension of words, C.Oph: Oral comprehension of sentences; C.E.: Written comprehension: C.Em.: Written Comprehension of words, C.Eph.: Written comprehension of sentences; R.: Repetition; L.: Reading; D.: Picture Naming; L.A.: Automatic language; P.B.F: Bucco-facial praxia; D.A.: Stress discrimination task (number of errors out of 10); D.F.: Phonetic discrimination task (out of 12); S.S.: Syllabic segmentation (out of 12); Mtv.: Working memory (out of 10); T.T; Token Test. Level of deficit: +: mild ; ++: moderated ; +++ : severe ; SA: no deficit. Patient Sex APH. TpO Et. G.I. C.Om. C.Oph. C.Em. C.Eph. R L. D. L.A. P.B.F D.A. D.F S.S Mtv T.T. FNG H B  6 CVA ++ S.A + S.A. + ++ +++ S.A. S.A. S.A. 4 S.A. S.A. 4 – FRG F B  8 CVA ++ S.A. + S.A. + + + + S.A. + 3 S.A. S.A. 3 26 MFB F B  6 CVA + S.A. + + + + S.A. S.A. S.A. S.A. 3 1 S.A. 3 – JLLV H B 36 CVA ++ S.A. + + ++ +++ +++ +++ ++ + S.A. S.A. S 5 – 2 out CPB H C 17 CVA ++ S.A. + S.A. S.A. ++ + ++ + + 1 1 2 28 of 10 LFC H C  8 Tumor + S.A. + S.A. + ++ + + S.A. S.A. S.A. S.A. S.A. 2 20 JMC H C  7 CVA + S.A. S.A. ++ S.A. + + + S.A. S.A. S.A. S.A. S.A 3 24 JAOF H C  1 CVA + S.A. S.A. S.A. S.A. + S.A. S.A. S.A. S.A. S.A. S.A. S.A. 7 32 Loquens, 2(2), July 2015, e023. eISSN 2386-2637 doi: http://dx.doi.org/10.3989/loquens.2015.023 Predicting segmental substitution errors in aphasic patients with phonological and phonetic encoding impairments • 15