Apraxia of Speech Theory, Assessment, Differential Diagnosis, and Treatment: Past, Present, and Future

A 50-plus year history of research on apraxia of speech (AOS) has spawned a body of literature that has been always controversial, at times contradictory, and at other times enlightening and coherent. This chapter reviews briefly the phonetic theory and neurological substrates for the disorder. We discuss neurolinguistic models and experimental evidence that have helped to set diagnostic criteria for AOS and that separate it from aphasia and the dysarthrias. The important, but underspecified, methods and procedures for the assessment of AOS are discussed, and the current status, recent trends, and advances in AOS treatment are reviewed. Although much remains to be explored about the neuropsychological mechanisms of the disorder and about its clinical identification and management, it is apparent that AOS (a) is an entity that is theoretically and clinically separable from language and other sensorimotor speech disorders, (b) is treatable, and (c) provides a model for exploring the interface between language and speech movements. Finally, we have identified what we believe are likely future trends in theory, diagnosis, assessment, and treatment of AOS that expand on current research and clinical practice rather than produce a paradigm change in any one of these areas. The concept that there exists a distinct clinical entity that impairs speech planning and/or programming has remained stable since the 1960s when long held notions about “motor aphasia” were perturbed—most notably by Darley, Luria, and Wepman—and then mutated into what Darley (1969) called apraxia of speech (AOS). Since then, this underspecified 196 SPEECH MOTOR CONTROL IN NORMAL AND DISORDERED SPEECH core concept has been on a nonlinear, sometimes circular, but generally forward-moving theoretical and clinical journey.1 Theory and Neurological Underpinnings Darley’s conceptual and nosologic contributions2 set the stage for studies of AOS in subsequent decades that (a) described its features and the variables that influence their emergence and (b) began to map the theoretical and clinical borders between it and its bookend disorders—aphasia and dysarthria. Although there was little controversy about distinctions between dysarthria and AOS during the 1970s and 1980s, there was considerable debate about whether AOS was indeed separable from aphasia. Confusion was fueled by definitions of AOS as a phonological disorder (Wertz, LaPointe, & Rosenbek, 1984) and by efforts to address its phonemic variability (Johns & Darley, 1970), despite the investigators’ conclusions that AOS was a motor, not a linguistic, disorder. Relatedly, and more problematic, the inclusion of people with phonemic paraphasias who may or may not also have had AOS likely contaminated some experimental findings (McNeil, Doyle, & Wambaugh, 2000). Nevertheless, acoustic, physiologic, and narrow phonetic transcription studies eventually identified features not easily explained as phonological or neuromuscular in nature, at least relative to models at the time. The accumulating evidence led to further revision of the core characteristics of AOS as well as overlap features that can be phonological, dysarthric, or apraxic in nature. Today, the term “phonological” is nowhere to be found in most definitions of AOS, beyond its explicit exclusion as the disorder’s core disturbance. AOS is now broadly defined, for example, as reflecting “inefficiencies in the translation of well-formed and -filled phonological frames into previously learned kinematic information” (McNeil, Robin, & Schmidt, 2009, p. 264), or interference with “speech-specific mechanisms of the organization of vocal tract gestures” (Ziegler, 2009, p. 659). From this standpoint, the theoretical distinction between the two processes (phonologic and phonetic) and between the two clinical problems (phonologic paraphasias and AOS errors) has been clarified. Although a gray area may be reopening at the phonologic and phonetic encoding interface (discussed later), the related theoretical issues will likely be more clearly defined than they have been in the past. Clinical descriptions have also changed—migrating from AOS being considered an articulatory disorder (Darley, 1969) to an articulatory disorder with secondary prosodic compensations (Darley, Aronson, & Brown, 1975) or with primary or secondary prosodic abnormalities (Wertz et al., 1984)—to current acceptance that both articulatory and prosodic abnormalities are primary features for its identification. Today, after a long latent period, the concept and label of AOS have spread beyond the practice and lexicon of speech-language pathology and are increasingly recognized by other neuroscience disciplines. With the increasing integration of functional neuroimaging, computational modeling, and theories of language and speech motor control, AOS is seen 1 See Ziegler, Aichert, and Staiger (2012) for an in-depth cogent and critical review of current concepts and unresolved issues regarding the nature of AOS. 2 See Rosenbek (2001) for an insightful review of Darley’s contributions. AOS THEORY, ASSESSMENT, DIAGNOSIS, AND TREATMENT 197 as an important exemplar of how the interaction between language and speech production can break down (Ziegler, Aichert, & Staiger, 2012). Furthermore, in the last two decades, the emergence of better specified models of phonologic and phonetic encoding has provided theoretical support for more explicit and narrower diagnostic criteria for AOS (McNeil, Robin, & Schmidt, 1997, 2009; Wambaugh, Duffy, McNeil, Robin, & Rogers, 2006a, 2006b). These more explicit criteria help focus debate and investigations of their validity (e.g., Staiger, Finger-Berg, Aichert, & Ziegler, 2012), and they increase confidence in studies that claim to be studying the disorder (McNeil, Pratt, & Fossett, 2004). Neurobiological Underpinnings and Modeling Influences Attempts to localize motor speech programming and AOS have probably forever drifted away from the unfulfilled classical search for a single responsible structure. There now is general consensus that speech is a product of actions within a distributed interactional network of cortical and subcortical structures and pathways that include, at the least, left premotor cortex, posterior inferior frontal lobe, supplementary motor area, sensorimotor cortex, auditory cortex, insula, basal ganglia, thalamus, cerebellum, and the parietal-frontal dorsal pathway (Bohland, Bullock, & Guenther, 2009; Ghosh, Tourville, & Guenther, 2008; Poeppel, Emmorey, Hickok, & Pylkkänen, 2012; van der Merwe, 2009). It is very unlikely that lesions in any one of these areas can reliably cause AOS, and leading candidate areas (posterior inferior frontal lobe, precentral gyrus of the insula) have not yet produced a clear winner(s) (cf. Dronkers, 1996; Hillis et al., 2004). Even if damage to a single or several structures can be reliably associated with AOS, how such damage explains specific features of AOS has yet to be determined (Miller, 2002). There have been important changes in psycholinguistic and computational models of phonological encoding and speech programming, and increasingly sophisticated structural and functional neuroimaging are facilitating their biologic validation and refinement (Bohland et al., 2009). Many models are underspecified relative to articulation and prosody, but several models have contributed substantially to concepts of the speech production process—for example, the Nijmegen model/Word-form Encoding by Activation and VERification (WEAVER) model (Levelt, Roelofs, & Meyer, 1999; Roelofs, 1997), the Directions Into Velocities of Articulators (DIVA) and Gradient Order DIVA (GODIVA) models (Bohland et al., 2009; Peeva, Guenther, Tourville, Nieto-Castanon, & Anton, 2010), the state feedback control models (Hickok, 2012), and Ziegler’s (2009) nonlinear probabilistic phonetic code model. For example, DIVA, the most thoroughly specified and investigated adaptive computational model of speech control, has generated data consistent with a number of kinematic and acoustic attributes of speech (Guenther, Ghosh, & Tourville, 2006). The GODIVA model, which interfaces with DIVA, links data from functional imaging and lesion studies to specific linguistic and motor components of the model, thus helping to validate or refine its plausibility. These efforts can help frame studies of communication disorders, including AOS. For example, in the GODIVA model, the inaccurate articulation associated with AOS could reflect either damage to motor programs or defective selection of motor programs, with each alternative localized 198 SPEECH MOTOR CONTROL IN NORMAL AND DISORDERED SPEECH differently in the model’s neural architecture (Bohland et al., 2009). Such hypotheses are beginning to foster efforts to examine speech planning and preparation in people with AOS using online reaction time methods (e.g., Maas & Mailend, 2012; Strand, 1986). These model-to-clinical relationships are bidirectional because clinical analysis of AOS breakdowns may help test model validity and subsequent refinements. “The reasoning about the relationship of abstract representations of word forms to their phonetic counterparts is currently undergoing a substantial change” (Ziegler et al., 2012, p. S1486). Models associated with articulatory phonology (such as those proposed by Browman & Goldstein, 1992; Goldstein, Pouplier, Chen, Saltzman, & Byrd, 2007) propose an interactive, boundary-blurring interface between phonological and phonetic encoding processes (Goldrick & Blumstein, 2006) in which “abstract” phonemes have physical properties (gestural scores) that are associated with speech movement patterns (Ziegler, 2009). Thus, the frequent perceptual challenge of separating AOS from phonological errors may partly reflect the complexities of “an integrated hierarchical action system whose higher order components have emerged from and are rooted in the physical conditions of speaking” (Ziegler et al., 2012, p. S1498). The implications of this line of thought for explanations of errors in AOS have already received attention (e.g., Laganaro, 2012). In the last half-century, theories about the nature of AOS and its neurological substrates have evolved. Testable hypotheses generated by neurobiologically valid models of normal speech production, and a better specified definition and model of AOS and its core behavioral features, will converge in the next decade (Ballard, Tourville, & Robin, 2014). We expect this convergence to yield a more refined theory of the disorder that is consistent with its clinical manifestations and its underlying structural and functional substrates. Diagnosis—The Search for Unique Patterns of Behavior or Neurobiology The behaviors, biology, and physiology by which AOS can be identified have in large measure been influenced by Darley’s initial characterization of the disorder as an “apraxia” in the tradition of Liepmann (1900). Liepmann (1900, 1913) and many others since have used the criteria that the diagnosis of an apraxia could be made only when (1) the basic sensorimotor integrity of the system was shown to be intact, (2) the underlying knowledge or representations for the action were shown to be intact, (3) the intention to move was clearly demonstrated, and (4) the action could be performed automatically while being impaired when produced volitionally. These criteria have had a profound influence and have perhaps constrained research on AOS since its identification by Darley in 1969. Remnants of this influence are especially evident from the 1970s and 1980s. During these two decades, studies were conducted on oral-sensory functions (Rosenbek, Wertz, & Darley, 1973), visual and auditory perception (Halpern, Keith, & Darley, 1976; Johns & Darley, 1970; Shewan, Leeper, & Booth, 1984), the underlying linguistic knowledge, and differences between automatic and volitional speech (Deal & Darley, 1972; Johns & Darley, 1970). Other researchers attempted to describe the behaviors that united this group, and the search for the underlying mechanisms of the movement disorder began in earnest. Darley’s clinical intuition was that this speech production deficit impaired individuals AOS THEORY, ASSESSMENT, DIAGNOSIS, AND TREATMENT 199 who were neither aphasic nor dysarthric. However, without clear a priori criteria for participant selection, group assignment for study relied on the theoretical supposition about the nature of phonological errors in persons with aphasia (PWA) as much as it did about the underlying nature of the movement disorder. That is, it was assumed that PWA would show comparable deficits in comprehension and in production, and if they did not (especially if no phonological comprehension deficits could be demonstrated), then the participants were indeed apraxic as long as they did not have dysarthria. Models of phonological encoding such as those of Dell (1988) or Nadeau (2001) that could account for speech sound-level production errors in the absence of comparable errors on the input side were not yet discussed, and evidence for a language-level disorder required the demonstration of comparable deficits in both perception/comprehension and production. This led to selecting participants for the study of AOS who were, without a doubt, likely aphasic as apraxic—leading to a database that described phonological behaviors in addition to those that could be legitimately attributed to motor planning/programming impairments. Circumstantial, corroborative evidence for a confounded database in the AOS literature is apparent in Halpern et al.’s (1976) study. Designed to investigate the phonological production errors in PWA, participants were excluded if they showed groping, off target highly inconsistent articulatory errors—primarily substitutions, additions, prolongations and repetitions—in attempting to target words in the context of islands of fluent speech, these errors being especially evident on repetition tasks and increasing incidence with increase in length of word. (Halpern et al., 1976, p. 366) They found that the great majority (93%) of their PWA did not demonstrate phonological errors, and those who did were accounted for by word-level errors (75%). As discussed by McNeil et al. (2004), abundant evidence is available that PWA, especially those labeled as having “conduction aphasia,” do make frequent phonemic errors. It is likely that this finding is the result of having included the phonological group with the AOS group that was so carefully eliminated from the study. Indeed, the elimination criteria are those that characterize individuals with phonological paraphasia (McNeil et al., 2009). Martin’s (1974) objection to the term AOS (not to its existence) as applied to the groups being studied raised this possibility and motivated a number of studies designed to demonstrate that the groups being studied were indeed not aphasic (e.g., Square, Darley, & Sommers, 1981; Square-Storer, Roy, & Hogg, 1990). It also prompted a search for and study of groups that did not have co-occurring pathologies (McNeil & Adams, 1990; McNeil, Weismer, Adams, & Mulligan, 1990; Square et al., 1981) from which defining characteristics could be derived. Studies of speech production impairments that preceded Darley’s identification of AOS as a separate syndrome from aphasia and dysarthria undoubtedly included participants who were apraxic—though labeled as aphasic (e.g., Shankweiler, Harris, & Taylor, 1968). Johns and Darley (1970) conducted the first study published with participants diagnosed 200 SPEECH MOTOR CONTROL IN NORMAL AND DISORDERED SPEECH with AOS. In this study, normal control and dysarthric groups were compared with a group with AOS. The selection criteria for the group with AOS were described as follows: “In each case the identification of the communication problem as apraxia of speech was made by one of the staff consultants in speech pathology after thorough examination of speech and language functions” (Johns & Darley, 1970, p. 559). Although no identification criteria were specified, the study compared perception with production (more production errors were evident in the participants with AOS and dysarthria than in the control participants), and better automatic than volitional production was reported. Additionally, the apraxic group performed with greater error inconsistency and unpredictability of error pattern than the dysarthric group. They also reported that unlike the dysarthric group, the “apraxic participants as a group made markedly fewer articulatory errors and gained in overall intelligibility when they read faster” (Johns & Darley, 1970, p. 579). Perhaps most revealing about the nature of the impairment in the AOS group was their characterization of their connected speech: These subjects as a group did a creditable job of miming secondary stutterers, both acoustically and behaviorally. They sometimes circumlocuted and substituted words. One 21-year-old man said, “When I was a june, when I was a ju-june, when I was a june-in the eleventh grade . . . .” They anticipated difficulty, made false starts, repeated, and blocked. They perseverated on phonemes, syllables, words and phrases. (Johns & Darley, 1970, p. 581) This description implicates phonological, lexical, and sentence-level language mechanisms that are not easily accounted for by any motor planning, programming, or execution model of speech production. These and other characteristics of the 10 individuals identified as apraxic in this initial investigation became the descriptors and criteria for group membership, formalized first by the Apraxia Battery for Adults test by Dabul (1986). From the inventory of articulation characteristics of AOS from this battery, only two characteristics appear to be unique to AOS (abnormal prosody and the intrusive schwa between syllables or within consonant clusters), and some features appear to be unique to phonemic paraphasia (i.e., phonemic anticipatory, perseverative, and transposition errors). Dabul’s (1986, 2000) credible list of observable behaviors that characterize AOS provide criteria for identifying phonological paraphasias and for identifying AOS (see McNeil et al., 2004, for additional discussion of this issue). Subsequent to Dabul’s influential AOS battery, Wertz et al. (1984) provided a comprehensive summary and interpretation of the literature on AOS to that point in time, and they identified the four most salient features of AOS: (1) effortful, trial-and-error groping of articulatory movements and attempts at selfcorrection; (2) dysprosody unrelieved by extended periods of normal rhythm, stress, and intonation; (3) frequent articulatory errors (predominated by sound substitutions); and (4) articulatory variability. Between 1984 and 1997, these identified characteristics (and those derived from Dabul’s, 1986, battery) became the primary criteria for the majority of studies selecting AOS participants for study and for treatment. AOS THEORY, ASSESSMENT, DIAGNOSIS, AND TREATMENT 201 McNeil et al. (1997; and, subsequently, McNeil et al., 2004, 2009) proposed a different set of behaviors that better differentiate AOS from phonemic paraphasia. Importantly, it is the specific perceptually derived cluster of behaviors that is claimed to differentiate AOS from its clinical neighbors because it is recognized that many of the single characteristics overlap with both phonological paraphasia and with dysarthria. Nonetheless, it was proposed that the presence of (1) sound distortions (including distorted sound substitutions), (2) extended segment durations (realized as slow speech with lengthened consonants and vowels), (3) extended intersegment durations (realized as sound, syllable, and word segregation), and (4) prosodic deficits that may be at least in part the result of extended segment and intersegment durations—along with some of the characteristics shared with aphasia (e.g., trial-to-trial variability, trial-and-error articulatory searching or groping behavior, increased errors with increased word length, and increased speech demands such as that required with increased rate)—yield a pattern of speech that is unique and perceptually identifiable as AOS. In 2006, a consensus group convened to analyze treatment studies in AOS (Wambaugh et al., 2006a, 2006b). In their evaluation of 59 treatment studies, the group first examined the evidence supporting the diagnosis of AOS for each study using a five-level descriptive scale. Level 1 specified that all primary characteristics consistent with the definition and criteria adopted from the core features of AOS identified by McNeil et al. (1997) were described. Level 2 described all primary characteristics consistent with the definition of AOS and some speech characteristics that may have been attributable to aphasia or dysarthria. Level 3 described most of the primary characteristics consistent with these criteria in addition to the identification of appropriate exclusionary behaviors. Level 4 evidence provided incomplete/inadequate description of the discriminative characteristics of AOS, without exclusionary behaviors. Level 5 evidence provided only the label, without a description of characteristics consistent with that diagnosis and/or with a description of behaviors that were contradictory to the criteria for AOS. With these criteria for participant identification, the authors concluded that only one of the studies provided Level 1 evidence in the characterization of participants. Nine studies were rated with Level 2 evidence, and the rest of the studies (83%) were judged to have provided inadequate justification for the diagnosis of AOS. This carefully considered evaluation of the literature adds to the evidence that the literature on AOS is likely constructed on an admixture of participants with AOS and with other speech production pathologies. Subsequent to these guideline documents, McNeil et al.’s (1997) description of AOS has been adopted frequently by researchers, at least in the United States, and continues to be frequently used for AOS diagnosis. It should be noted that they have not been adopted universally and that they have not been subjected to rigorous scientific study. Indeed, the kernel characteristics for the diagnosis of AOS (identified by Wertz et al., 1984) that have been argued to be inadequate for AOS diagnosis continue to be used to select AOS participants in some studies (Dronkers, 2004; Richardson, Fillmore, Rorden, LaPointe, & Fridriksson, 2012). Josephs, Duffy, and colleagues are currently exploring additional criteria for the identification of AOS that co-occur or are the sole speech/language feature in degenerative neurologic disease. In one important study, Josephs et al. (2012) characterized primary 202 SPEECH MOTOR CONTROL IN NORMAL AND DISORDERED SPEECH progressive AOS by listing the prevalence of 16 signs derived from 12 participants who met their extensive neurological and behavioral inclusion and exclusion criteria for isolated primary progressive AOS. Table 9.1 shows the hierarchy of perceptually identified speech production characteristics observed in these participants. They identified (1) slow overall speech rate, (2) lengthened intersegment durations, (3) increased sound distortions and distorted sound substitutions with increasing target length, (4) syllable segmentation within words greater than one syllable in length, (5) sound distortions, and (6) syllable segmentation across words in phrases/sentences as the six most frequently occurring features. Importantly, they noted that many of the 16 features can also occur in other sound-level speech production pathologies, including aphasia and spastic dysarthria, and five of the six most frequently occurring signs were also identified as perceptual signs of spastic dysarthria. This hierarchy of observed signs not only provides converging evidence Table 9.1. Prevalence (highest to lowest) of speech characteristics identified in progressive apraxia of speech (Josephs et al., 2012, p. 2).