The most prominent symptom of aphasia is difficulty

Transcription

1 AJSLP Research Article Effect of Verb Network Strengthening Treatment in Moderate-to-Severe Aphasia Lisa A. Edmonds a and Michelle Babb a Purpose: This Phase II treatment study examined the effect of Verb Network Strengthening Treatment (VNeST) on individuals with moderate-to-severe aphasia. Research questions addressed (a) pre- to posttreatment changes and pretreatment to treatment phase changes on probe sentences containing trained verbs (e.g., The carpenter is measuring the stairs ) and semantically related untrained verbs (e.g., The nurse is weighing the baby ); (b) lexical retrieval changes in single-word naming, sentence, and discourse measures; (c) functional communication by way of proxy and participant report; and (d) error evolution. Method: A multiple-baseline approach across participants was used. Effect sizes were calculated for pre- and posttreatment and maintenance probe responses. A C statistic was used to determine changes from the baseline to treatment phases. Results: One participant exhibited improvement on all generalization measures, whereas the other participant exhibited more limited generalization. Both participants showed improvement on the functional communication measure. Conclusions: As predicted, the participants did not show the same extent of improvement that was observed in participants with more moderate aphasia (Edmonds, Nadeu, & Kiran, 2009). Nonetheless, the findings suggest that VNeST may be appropriate for persons with moderate-to-severe aphasia, especially with a small adaptation to the treatment protocol that will be retained for future iterations of VNeST. Key Words: aphasia, verbs, treatment, VNeST, sentences The most prominent symptom of aphasia is difficulty retrieving words (Laine & Martin, 2006). The majority of word retrieval therapies reported in the literature have focused on picture naming and single-word noun retrieval (Kiran & Bassetto, 2008; Nickels, 2002). Overall, there has been limited generalization to untrained items and tasks, and without generalization, a treatment may be considered ineffective (Thompson, 2006). One exception is semantic feature analysis (SFA), which requires participants to retrieve features related to trained items (e.g., an apple is a fruit, sweet, and found on trees). Generalization to untrained semantically related items (e.g., orange) that share features of the trained items has been shown in a number of SFA studies (e.g., Boyle, 2004; Boyle & Coelho, 1995; Edmonds & Kiran, 2006; Kiran & Thompson, 2003; Law, Wong, Sung, & Hon, 2006). However, generalization beyond the word level has been negligible in SFA (see Nickels, 2002), even when items treated are selected from the discourse outcome measure (e.g., Rider, Wright, Marshall, & a The University of Florida, Gainesville Correspondence to Lisa A. Edmonds: edmonds@ufl.edu Editor: Laura Justice Associate Editor: Michelle Bourgeois Received April 26, 2010 Revision received September 8, 2010 Accepted February 25, 2011 DOI: / (2011/ ) Page, 2008) or are embedded during sentence production (Boyle, 2009). Improvement to noun retrieval in discourse has been observed when SFA was paired with the PACE (promoting aphasics communicative effectiveness) technique (Antonucci, 2009). In recent years, verb-centered treatments have emerged to address lexical retrieval deficits in sentence production. The governing assumption in these treatments is that verbs are central to sentence formulation and that an increase in verb retrieval will aid sentence production. Treatment methods have included sentence completion, cuing hierarchies, picture naming, SFA for verbs, and retrieving verbs (and sometimes nouns) within a sentence context (e.g., Edwards, Tucker, & McCann, 2004; Fink, Martin, Schwartz, Saffran, & Myers, 1992; Raymer et al., 2007; Raymer & Ellsworth, 2002; also, see Nickels, 2002). The results of these studies reveal that improvement in retrieval of trained verbs can result in improved sentence production with those verbs (Raymer & Ellsworth, 2002; Schneider & Thompson, 2003), but increased verb retrieval does not necessarily result in increased lexical retrieval in sentence production (Raymer et al., 2007; Schneider & Thompson, 2003; Wambaugh et al., 2001; also, see Nickels, 2002). There have also been mixed findings related to generalization to untrained verbs, sentences, and discourse (e.g., Edwards et al., 2004; Loverso, Prescott, & Selinger, 1988; Prescott, Selinger, & Loverso, 1982; Raymer & Ellsworth, 2002; Wambaugh & Ferguson, 2007). Theoretical work done to examine the semantic organization of verbs and their thematic roles serves as the foundation American Journal of Speech-Language Pathology Vol May 2011 A American Speech-Language-Hearing Association 131

2 for Verb Network Strengthening Treatment (VNeST), the focus of the current study. McRae and colleagues have investigated the interrelationship of verbs and their thematic roles and whether world knowledge related to thematic role concepts is verb-specific rather than part of distinct episodic or semantic memory systems. In a series of priming studies in young adults, they found a priming/facilitation effect for agent (the doer of the action) verb and patient (the receiver of the action) verb pairs when the agents or patients were presented after a related verb (e.g., visiting-tourist) as compared to an unrelated verb (e.g., visiting-prosecutor; Ferretti, McRae, & Hatherell, 2001). McRae, Hare, and Ferretti (2005) found that agents and patients primed related verbs, indicating that the priming effect was bidirectional. Bidirectional priming has also been observed for instruments (stirred-spoon; Ferretti et al., 2001; McRae et al., 2005), indicating that instrument representations are a part of event schemas that contain them. Verb-location pairs (e.g., skating-arena) resulted in priming from locations to verbs but not in the opposite direction. Additionally, other researchers have found priming among agent-verb-patient triads (e.g., wood + carpenter prime the verb cuts as compared to wood + carpenter with the verb eats; Khader, Scherag, Streb, & Rösler, 2003, Experiment 1). Based on their findings, McRae and colleagues have posited that memory is structured so that when a verb is read or heard, generalized situation knowledge is also activated (Ferretti et al., 2001, p. 523) and that an important part of this activated knowledge includes the types of agents and patients who participate in the activity. However, the authors do not claim that every instance of an event is brought to consciousness when a verb is encountered (McRae, Ferretti, & Amyote, 1997, p. 170). Rather, they conceptualize an organization of memory involving computation of a prototypical representation of the related thematic roles that is shaded by context (McRae et al., 1997, p. 170). Thus, they propose that word meaning and world knowledge is stored in the same form as episodes, and presumably constructed on-line from them (McRae et al., 1997, p. 141), suggesting no distinction between semantic and episodic memory (also see McKoon, Ratcliff, & Dell, 1986; Snowden & Neary, 2002). The protocol of VNeST was designed to require participants to generate explicit thematic roles related to trained verbs (the verb network) that represent relevant event schemas, thereby simultaneously activating world and semantic knowledge that would activate word/lemma level forms (e.g., Dell, Schwartz, Martin, Saffran, & Gagnon, 1997) within a basic sentence frame (subject/agent-verb-object/patient). Specifically, participants generate pairs of agents and patients for verbs, such that a single verb can result in many thematic role pairs that elicit various schemas (e.g., carpenter-lumber, chef-sugar, seamstress-fabric for measure) but which all reinforce, in part by the variety of possibilities, the target verb itself (see Figure 1 for an illustration). Additionally, participants answer wh-questions (where, when, and why) about one pair of words for each trained verb to further develop the schema and to make it more salient and relevant to the participant. For the example of a carpenter measuring lumber, a participant may recall when her kitchen was remodeled because it was outdated. Thus, treatment targeted at the verb is proposed to increase activation of the verb and related thematic roles in order to facilitate retrieval of the agent, verb, and patient in sentence production and potentially discourse. As previously mentioned, a treatment may not be considered effective without generalization (Thompson, 2006); therefore, extending improvement beyond the trained verb networks is a primary goal of VNeST. Improving lexical retrieval within sentences containing untrained verbs semantically related to trained verbs (e.g., measure/weigh) is based on the idea that engaging a verb in VNeST can potentially activate a semantically related verb via spreading activation (see Figure 1). This prediction is based on (a) evidence that verbs prime semantically related verbs (e.g., sweep-dust; Rösler, Streb, & Haan, 2001), (b) findings from previous semantic treatments for nouns (e.g., Kiran & Thompson, 2003) and verbs (e.g., Raymer & Ellsworth, 2002), and (c) previous VNeST findings (Edmonds, Nadeau, & Kiran, 2009). Furthermore, the activation of a semantically related verb would likely activate the verb s associated thematic roles. In this way, training the verb network for measure could facilitate production of sentences containing related verbs (e.g., The nurse is weighing the patient ). See Figure 1 for an illustration. The current investigation is the second study examining the effect of VNeST on lexical retrieval across various contexts. VNeST was first reported in a Phase I treatment study with four participants with moderate aphasia (two fluent and two nonfluent; Edmonds et al., 2009). The primary goals of the first treatment study were, and continue to be, to (a) increase the specificity of content words produced in sentences by persons with aphasia, (b) maximize generalization to untrained items across multiple contexts, and (c) engage and challenge participants with salient and relevant treatment items and protocol. All participants exhibited generalization to lexical retrieval of content words during sentence production for picture description of sentences containing trained verbs (e.g., The carpenter is measuring the stairs ) and semantically related untrained verbs (e.g., The nurse is weighing the baby ). Further improvement in sentence production was captured on the Northwestern Verb Production Battery (Thompson, 2002). Single-word noun and verb naming also improved for three of four participants, and discourse samples showed an increase in the use of complete utterances (relevant utterances containing subjects and verbs) for three participants. Thus, the preliminary findings of the first study showed encouraging generalization across participants to various lexical items (response generalization) and tasks (stimulus generalization), more than previous semantic and verb-centered treatments for lexical retrieval. The goals of the current Phase II study were to examine the effects of VNeST in a population that differed slightly from the first study (Robey, 2004), in this case in persons with more severe aphasia. Beeson and Robey (2006) suggested that Phase I and Phase II treatment studies should focus on change from pre- to posttreatment rather than performance during the treatment phase, which, they suggested, could be evaluated in later phases of treatment research. Thus, examination of effect sizes was our primary metric of change for probe sentences. Evaluation of baseline to treatment phase changes 132 American Journal of Speech-Language Pathology Vol May 2011

3 FIGURE 1. (a) Core Verb Network Strengthening Treatment (VNeST) steps that theoretically promote spreading activation to trained and semantically related untrained verb networks. (b) Theorized verb networks activated by treatment with proposed generalization possibilities. was secondary. Thus, our research questions (RQs) were as follows: RQ1. What is the magnitude of change as measured by effect size (d; Beeson & Robey, 2006) from baseline to posttreatment on probe sentences containing trained stimuli (e.g., The carpenter is measuring the stairs ) and semantically related untrained stimuli (e.g., The nurse is weighing the baby )? RQ2. What is the magnitude of change as measured by effect size (d) from baseline to maintenance on sentence probes with trained and untrained stimuli? RQ3. Is there a change of slope from the baseline to the treatment phase for probe sentences containing trained and semantically related untrained stimuli as measured by a C statistic (Tryon, 1982, 1984)? Using the results of Edmonds et al. (2009) as a guide, we predicted that effect sizes would be small to moderate. We also predicted an increase in slope during the treatment phase on both sets of probe sentences, though we predicted a shallower increase with more treatment sessions required as compared to the less impaired participants in Edmonds et al. (2009). To elucidate the results of our probe data and to evaluate potential generalization to a larger scope of lexical retrieval tasks and functional communication, we also asked the following questions: RQ4. Is there increased pre- to posttreatment performance on: a. single-word naming of objects and actions on the Object & Action Naming Battery (Druks & Masterson, 2000) as evaluated by a McNemar test? b. sentence production on the Argument Structure Production Test (ASPT) from the Northwestern Assessment of Verbs and Sentences (NAVS; Thompson, 2002) as evaluated by a McNemar test? c. production of correct information units (CIUs) in discourse (evaluated qualitatively due to the short length of samples)? d. ratings of functional communication on the Communicative Effectiveness Index questionnaire (CETI; Lomas et al., 1989) from participants and proxies as measured by a Wilcoxon signed-ranks test? Based on the results of Edmonds et al. (2009), an increase in single-word naming of nouns and verbs was expected, with perhaps more improvement to nouns. Improvement to discourse and sentence production on the NAVS was uncertain given the severity of the participants. Based on anecdotal Edmonds & Babb: Verb Network Strengthening Treatment 133

4 feedback from participants and families in Edmonds et al. (2009), we expected some evidence of functional improvement on the CETI. Finally, to understand the changes in lexical retrieval beyond accuracy, we analyzed errors on lexical retrieval of single words (Object & Action Naming Battery) and sentences (NAVS). Our RQ related to the error analysis was as follows: RQ5. Do errors on lexical retrieval evolve to reflect responses closer semantically and/or phonemically to the target? Error types were anticipated to shift toward responses semantically or phonemically closer to the target posttreatment. We evaluated errors qualitatively, not statistically. Method Experimental Design A multiple-baseline approach across participants was used with four phases: (a) baseline, (b) treatment of trained items with administration of generalization and control probes, (c) posttreatment probes, and (d) maintenance. The weekly probe task (associated with RQ1, RQ2, and RQ3) was sentence production for pictures depicting trained verbs ( The carpenter is measuring the stairs ) and untrained semantically related verbs ( The nurse is weighing the baby ). A weekly adjective control task was also administered for demonstration of experimental control. Details of stimuli development and procedures for administration of these measures are detailed below. Participants Two participants, P1 and P2, were recruited from the University of Florida Speech and Hearing Clinic. Participants met several inclusion criteria, including (a) diagnosis of aphasia with an aphasia quotient of <50 based on the Western Aphasia Battery (WAB; Kertesz, 1982), (b) monolingual English-speaking, (c) right-handedness prior to stroke, (d) considerable lexical retrieval deficits (approximately 50% on the Object & Action Naming Battery and NAVS ASPT), (e) negative history of diagnosed learning disorder or drug/ alcohol addiction, and (f ) no worse than a composite score of a mild deficit on the Cognitive Linguistic Quick Test (Helm-Estabrooks, 2001). P1 was a 42-year-old, right-handed woman with 16 years of education. Forty-nine months before beginning the study, P1 had a left hemisphere stroke. According to radiology reports, she sustained a massive left middle cerebral artery (MCA) distribution stroke extending from the striatocapsular territory through the operculum and involving the entire left frontal lobe from the base to the upper convexity and including the anterior cerebral artery territory. The infarct extended deep into the surface of the left lateral ventricle, involving the entire left periventricular white matter and much of the left superior temporal gyrus. Subsequent to her stroke, she was hospitalized for 3 months and underwent a decompressive craniotomy 48 hr after admission. During her hospitalization, she also sustained a small stroke in the high right pre- and postcentral gyrus and patchy infarction in the right lenticulostriate end zone resulting in relatively minimal infarction. Prior to her stroke, P1 worked as a lab technician and lived independently. At the time of the study, she was living independently close to her parents and received considerable support from them. P1 had received speech therapy previously, but she had discontinued it 2 years before participating in the current study. P1 passed a hearing screening at 30 db. P2 was a 49-year-old, right-handed woman with 12 years of education. Nine months prior to beginning the study, she had a large left MCA stroke right after the M1 segment involving almost the complete distribution, with mass effect upon the left cerebral hemisphere secondary to extensive parenchymal edema involving the temporoparietal region. According to the neurology report, the stroke involved considerable cortical gray, gray-white interface, and deep white matter that resulted in a significant right hemiparesis requiring the use of a wheelchair. Results of a full hearing evaluation at our speech and hearing clinic facility indicated a moderate bilateral high-frequency hearing loss (45 db at Hz) with good (>88%) word recognition. Hearing aids were recommended to help distinguish high-frequency speech sounds, but she did not get them during the course of the study. The hearing loss did not appear to affect P2 s performance in treatment, but the clinician was careful to produce words with higher frequency sounds (e.g., /s/ and /f/) in a loud voice to aid comprehension. Prior to her stroke, P2 was a choreographer living independently. At the time of the study, she was living with her parents for assistance with activities of daily living. P2 had received previous speech therapy, but she had discontinued this therapy a few months prior to enrollment in this study. Tests Administered Test used for inclusion/exclusion. The Cognitive Linguistic Quick Test (Helm-Estabrooks, 2001) was administered as a screen for cognitive impairments. To be included in the study, participants had to achieve a composite score of within normal limits (WNL) or mild deficit. Both participants had a composite score representing a mild deficit. Subscores for attention, executive function, and visuospatial skills were WNL. Both participants scored in the severe category for language and memory. The severe language results were consistent with their aphasia diagnoses; however, the severe memory results were inconsistent with our interactions with the participants. Upon investigation, it appeared that their memory scores were affected by components of their language performance, which were used to compute the memory subscore. Tests used to describe selected abilities. The WAB (Kertesz, 1982) was administered to determine aphasia severity. P1 scored an Aphasia Quotient of 45.2, and P2 scored Both participants were classified as having Broca s aphasia according to the WAB manual. In general, both were predominantly nonfluent with utterance lengths of one (P2) to one-to-two (P1) words, with relatively good comprehension and poor repetition (more detail on the participants linguistic abilities is presented below). 134 American Journal of Speech-Language Pathology Vol May 2011

5 Semantic processing was tested using the pictures in the Pyramids and Palm Trees test (Howard & Patterson, 1992) for nouns and the Kissing and Dancing test (Bak & Hodges, 2003) for verbs. On each test, participants pointed to one of two items that went best with the target item (e.g., eyes for eyeglasses rather than ears). P1 scored WNL for verbs with a mild deficit in noun processing. P2 scored WNL on both tests. The Apraxia Battery for Adults, Second Edition (Dabul, 2000) was administered to determine the presence and/or extent of speech, oral, or limb apraxia. To be included in the study, participants had to achieve an overall profile rating of no or mild deficit. Both participants had profile scores representing an overall mild impairment. Subscores for P1 revealed mild oral apraxia and mild deficits in all other tasks except for limb apraxia, which was WNL. P2 s subscores were WNL for oral and limb apraxia, with mild deficits in all other tasks except the diadochokinetic rate subtest, which revealed a moderate deficit. Additionally, characteristics of mild apraxia of speech were also observed (based on McNeil, Robin, & Schmidt, 2009). P1 exhibited a slow rate with inability to increase rate while maintaining phonemic integrity, successive self-initiated trials to repair error with responses no closer to the target, phoneme distortions, including distorted sound substitutions, and consistent type and location of errors within an utterance. P2 exhibited an inability to increase rate while maintaining phonemic integrity, phoneme distortions, and prolonged vowels. Pre- to posttesting measures for RQ4. The tests in this section were analyzed from pre- to posttreatment to address RQ4. Single-word lexical retrieval was evaluated with the Boston Naming Test (BNT; Goodglass, Kaplan, & Weintraub, 1983) and Object & Action Naming Battery (Druks & Masterson, 2000). The BNT contains 60 pictures of objects decreasing in frequency (e.g., bed to abacus), and the Object & Action Naming Battery contains pictures of objects (n = 162) and actions (n = 100). The objects and action stimuli are divided into two forms (A and B) that are balanced for psycholinguistic (e.g., frequency and age of acquisition) and nonlinguistic (e.g., imageability and visual complexity) stimulirelated variables. Both participants were given Form A. Both participants named nouns better than verbs, and P1 scored higher overall. Given P2 s poor spoken naming, she was also tested on written naming. She performed markedly higher on written naming as compared to spoken naming (see Table 1 for details). Lexical retrieval in sentences was evaluated with the ASPT from the NAVS (Thompson, 2002). The NAVS consists of 50 pictures of common situations with different people and animals. Verbs included one-, two-, and three-place verbs (e.g., The dog is barking, The boy is climbing a tree, and The woman is putting the box on the shelf ). The NAVS TABLE 1. Scores for tests and questionnaires administered pre- and posttreatment. Participant 1 (P1) Participant 2 (P2) Spoken Spoken Written Tests and questionnaires Pretreatment Posttreatment Pretreatment Posttreatment Pretreatment Posttreatment Aphasia severity Western Aphasia Battery a Aphasia Quotient Information Fluency Comprehension Repetition Naming Semantic processing Pyramids and Palm Trees b (n = 52) 45 (86.5%) 49 (94.2%) 49 (94.2%) 47 (90.4%) Kissing and Dancing c (n = 52) 48 (92.3%) 51 (98.1%) 47 (90.4%) 50 (96.2%) Lexical retrieval Boston Naming Test d (n = 60) 21 (35%) 22 (36.6%) 5 (8.3%) 16 (26.6%)** Object & Action Naming Battery e Nouns (n = 81) 37 (45.7%) 53 (65.4%)* 13 (16%) 39 (48.1%)** 63 (77.7%) 69 (85.2%)** Verbs (n = 50) 28 (56%) 25 (50%) 6 (12%) 21 (42%)** 33 (66%) 43 (86%)* Northwestern Assessment of Verbs and Sentences f Sentence production without verb provided (n = 36) 9 (25%) 9 (25%) 0 (0%) 8 (22.2%)** Functional communication P1 s mom (respondent) P2 (respondent) P2 s mom (respondent) Communicative Effectiveness Index g Average SD * * * a Kertesz (1982). b Howard and Patterson (1992). c Bak and Hodges (2003). d Goodglass et al. (1983). e Druks and Masterson (2000). f Argument Structure Production Test (Thompson, 2002). g Lomas et al. (1989). *p <.05 from pre- to posttreatment. **p <.001 from pre- to posttreatment. Edmonds & Babb: Verb Network Strengthening Treatment 135

6 protocol requires showing and reading the verb to the participant. We modified the protocol so that the verb was not shown to or heard by the participants. Furthermore, the test contains two- and three-place optional verbs in optional and obligatory contexts (e.g., The man is sweeping and The man is sweeping the dirt ), but in our adaptation only the more complex sentence was shown to avoid multiple presentations. Thus, a total of 36 of 50 pictures were administered (oneplace, n= 8; two-place, n= 17; three-place, n= 11). Participants received credit if all required lexical items of the sentence were present, regardless of word order or inflection of the verb. P1 achieved 22% accuracy, and P2 produced no sentences (see Table 1). The CETI questionnaire (Lomas et al., 1989) was completed by the participants and their mothers to get insight into the participants abilities in functional communication interactions. The responses to the questions are represented on a visual analogue scale, which participants mark from Not at all able to As able as before stroke along a 100-mm line. The higher on the line that the people mark, the more as able as before the stroke the communication is rated. P1 had difficulty understanding the questionnaire, so her responses could not be evaluated. P1 s mother s responses revealed an average score across the 16 questions of (SD = 15.95). Relative strengths included tasks not requiring spoken output (e.g., giving appropriate yes/no answers and communicating without words). P2 had no difficulty with the questionnaire and rated herself with an average of (SD =29.41). Relative strengths included understanding writing and responding without words, indicating that she understood what was being said to her. Questions related to spoken output were scored much lower (e.g., having a one-to-one conversion and saying someone s name). P2 s mother rated her much higher than P2 herself, with an average of (SD = 17.51), though the relative strengths she indicated were similar to those P2 indicated. Connected speech. To gauge lexical retrieval in connected speech, participants were asked to respond to all 10 elicitation materials in Nicholas and Brookshire (1993), which included procedural, personal, single-picture, and sequential picture stimuli. For P1, 93.9% (n = 79) of her responses were real words, and 6.1% (n = 6) were neologisms. CIUs were coded and used to determine what percentage of real words was related to the target (Nicholas & Brookshire, 1993). For P1, 67.87% of the pretreatment words were CIUs. For P2, 71.42% (n =35) of all lexical items were real words, and 28.58% (n = 14) were neologisms. Of the real words, 67.56% were CIUs. Error Analysis An error analysis was conducted (RQ5) on individual words for pre- and posttreatment single-word naming on the BNT and Object & Action Naming Battery (see Table 2). TABLE 2. Pre- to posttreatment error types for single naming of nouns and verbs. Nouns (Boston Naming Test and Object & Action Naming Battery) Verbs (Object & Action Naming Battery) Participant Type of error Pretreatment Posttreatment Pretreatment Posttreatment P1 P2 P2 Spoken n Semantic 51.8% 60.6% 9.1% (Same) 36% (Same) 9.1% (Diff) 20% (Diff) 50% phonemes 4.8% 13.6% 0% 4% Mixed 2.4% 3.1% 4.5% 12% Unrelated 12.1% 13.6% 27.3% 28% Neologism/<50% phonemes 8.4% 3.1% 9.1% 0% IDK/NR/unintelligible 20.5% 6.0% 40.9% 0% Spoken n Semantic 0.8% 3.5% 2.3% (Same) 3.4% (Same) 2.3% (Diff) 3.4% (Diff) 50% phonemes 8.1% 13.9% 4.5% 13.8% Mixed 1.6% 2.3% 2.3% 6.9% Unrelated 11.4% 3.5% 4.5% 13.8% Neologism/<50% phonemes 41.5% 75.6% 9.1% 58.7% IDK/NR/unintelligible 36.6% 1.2% 75% 0% Written n Semantic 33.3% 41.7% 23.5% (Same) 14.3% (Same) 29.4% (Diff) 42.8% (Diff) 50% graphemes 16.7% 41.7% 0% 0% Mixed 5.6% 0% 0% 0% Unrelated 11.1% 8.3% 35.3% 14.3% Neologism/<50% graphemes 11.1% 0% 0% 14.3% IDK/NR/unintelligible 22.2% 8.3% 11.8% 14.3% Note. Same = error within the same grammatical class; Diff = error within a different grammatical class; IDK/NR = I don t know /no response. 136 American Journal of Speech-Language Pathology Vol May 2011

7 Incorrect responses were assigned to one of six error categories: (a) semantic (subdivided into semantic errors of the same grammatical class, such as noun for a noun, and different grammatical class, such as noun for verb), (b) 50% of the phonemes/graphemes correct, (c) mixed (a combined semantic and phonemic error), (d) unrelated (a real word not related to the target), (e) neologism, and (f ) I don t know / no response (IDK/NR). Perseveration was also a category, but neither participant produced perseverative errors. An error analysis was also conducted on the ASPT. Errors for each word in the sentence (subject, verb, direct object, and indirect object) were categorized as semantic, phonemic, grammatical (e.g., her for she), unrelated, neologism, or omission (see Table 3 for details). Hypothesized levels of impairment. There were no specific criteria related to the proposed level of impairment in lexical retrieval for participants. However, it is informative to examine outcomes and errors within a model of lexical retrieval (e.g., Kay, Lesser, & Coltheart, 1992). P1 was more impaired on verbs than nouns. She appeared to have a relatively intact semantic system given her semantic testing scores and relatively good comprehension. She primarily produced semantic errors in single-word naming, indicating a disruption between semantic and phonological levels (Dell et al., 1997). Additionally, the presence of phonemic errors close to the target and impaired repetition suggest additional disruption at the level of the phonologic output lexicon (POL; Kay et al., 1992). If P2 exhibited more difficulty with verbs than nouns, it was subtle, as her noun and verb scores were similar. However, one third of her semantic errors in verb naming were noun productions, indicating that noun production may have been somewhat more preserved. P2 exemplified a more intact semantic system than P1, with WNL scores on semantic testing and better auditory and reading comprehension. P2 s errors across spoken tasks were largely neologistic, with impaired repetition, indicating a primary impairment at the phonological level. Written errors were largely semantic and graphemic, indicating disruption between the semantic system and the orthographic output lexicon. See Tables 2 and 3 for error analysis results. Stimuli Development Sentence elicitation pictures. The details of stimuli development are described in Edmonds et al. (2009). However, a brief review is included here along with discussion of additional stimuli. For the first study, 24 pictures were developed for baseline and treatment probes. All pictures elicited an agent, verb, and patient. The agents and patients had specific titles in most cases (e.g., nurse or carpenter) to promote specific language instead of general words (e.g., lady or man). Sentences were divided into two sets, so that each verb in one set was semantically related to a verb in the other set (e.g., measure/weigh). For the current study, three additional sentences/pictures were created: (a) The patient is brushing his teeth, (b) The soldier is shaving his face, and (c) The teacher is erasing the board. The verbs brush and shave are semantically related grooming verbs. The sentence with erase was created for P2 to be a semantic pair with scrub ( The maid is scrubbing the floor ) to replace a picture depicting a zookeeper, which was problematic due to low naming agreeability. Control task. A single-word adjective retrieval task was developed as a control task that would rule out the possibility that improvements occurring during treatment reflected a nonspecific effect on semantic knowledge. Participants were expected to complete a sentence by providing a synonym for the adjective provided in the sentence. The adjectives were matched to the 24 verbs included in the probe stimuli on frequency. See Edmonds et al. (2009) for details. Procedures for Collecting Probe Data (RQ1, RQ2, and RQ3) and Administering Treatment Probe and control measures. During baseline, 20 probe pictures and 10 items for the adjective control task items were TABLE 3. Pre- to posttreatment error types for the Argument Structure Production Test for both participants. Pretreatment Posttreatment Participant Type of error Sub Verb DO IO Sub Verb DO IO P1 P2 Note. n Semantic 0% 27.27% 0% 33.33% 50.00% 34.78% 33.33% 0% Phonemic 0% 4.55% 0% 0% 0% 8.70% 0% 0% Grammatical 0% 0% 0% 0% 0% 0% 0% 0% Unrelated 0% 9.09% 9.09% 0% 50.00% 26.09% 11.11% 0% Neologism 0% 4.55% 0% 0% 0% 0% 0% 0% Omission 0% 54.55% 90.91% 66.67% 0% 30.43% 55.56% 100% n Semantic 18.18% 2.86% 4.35% 0% 37.50% 19.05% 9.09% 0% Phonemic 0% 2.86% 4.35% 0% 0% 4.76% 0% 11.11% Grammatical 22.73% 0% 0% 0% 0% 0% 0% 0% Unrelated 0% 2.86% 0% 0% 12.50% 4.76% 9.09% 0% Neologism 4.55% 14.29% 0% 0% 25.00% 33.33% 27.27% 33.33% Omission 54.55% 77.14% 91.30% 100% 25.00% 38.10% 54.44% 55.56% Sub = subject; DO = direct object; IO = indirect object. Edmonds & Babb: Verb Network Strengthening Treatment 137

8 administered at the beginning of each session. During administration, probe pictures were presented pseudorandomly with semantically related verbs (e.g., bake/fry) in nonsequential order. For each picture, participants were instructed to Make a sentence and include him/her, the action, and this (while pointing to the agent [carpenter], verb [measure], and patient [stairs]). Prompts were not provided unless the participant produced a general word for the target (e.g., cut instead of slice or man instead of carpenter), for which a prompt for a more specific word was given. Given P2 s severe spoken output deficits with relatively preserved writing, weekly probes were administered in spoken and written modalities (on different days) to assess potential improvement in both modalities. For the control task, participants were asked to complete sentences using a synonym for the provided adjective (e.g., Someone who is sick is also said to be [target = ill]). In the event of multiple attempts, the adjective closest to the target was scored. Treatment. The treatment protocol is detailed in Edmonds et al. (2009), but a summary is provided here. Stimuli consisted of (a) 10 cards containing the names of the 10 trained verbs (Verb Set 1; e.g., measure); (b) six to eight cards for each verb containing three to four agents and three to four patients that formed three to four pairs related to each verb (e.g., chef/sugar, carpenter/lumber, surveyor/land, and designer/room for the verb measure) chosentorepresenta range of possibilities; (c) five cards containing the words who, what, where, when, and why; and (d) 12 sentences used for semantic judgment (heard but not seen by participants). The 12 sentences contained the target verb broken into four categories: (a) correct ( The designer measures the room ), (b) inappropriate agent ( The infant measures the lumber ), (c) inappropriate patient ( The chef measures the television ), and (d) thematic reversal ( The room measures the designer ). VNeST was administered two times per week for 2-hr sessions. The first hour of the second session each week was dedicated to probes for P1. For P2, probes were presented during the beginning of each session (written or spoken). During treatment, participants were asked to produce orally three to four thematic role pairs (e.g., carpenter and lumber) for a provided verb (e.g., measure). When they were unable to produce a word, written options on cards were provided (some appropriate and some foils). Participants were encouraged to provide at least one personal pair (e.g., dad/boat for drive), and responses could change from week to week. In the original protocol (i.e., Edmonds et al., 2009), after generating three to four appropriate agent-patient pairs, participants read each agent-patient pair aloud (the verb was not read aloud) and chose one to answer wh-questions about it (e.g., when, where, or why); this is the protocol P1 followed. However, it was apparent after P1 completed the study that reading the verb with the nouns might be important, so a modification was made for P2 so that she read the agent, verb, and patient. An additional modification was made to Step 1 to accommodate P2 s sparse and neologistic speech. Following the protocol, she was asked to orally produce thematic role pairs for a provided verb; however, when she was unable to provide an appropriate verbal response, she was allowed to write her response. Oral reading of her correct written response always followed. This adaptation allowed P2 to express her own ideas rather than being provided potentially less relevant words by the clinician. Criterion for ending treatment was met when participants produced a minimum of 24 relevant agent-patient pairs (80% accuracy) during treatment Step 1 (e.g., for measure, acceptable pairs could include chef/sugar, wife/windows, or designer/room). Since there were 10 treatment items and three opportunities for agent-patient pairs, a total of 30 pairs in 1 week was possible. Thus, 24 correct pairs (80%) in 1 week met treatment termination criterion. P1 achieved criterion for ending treatment on Week 9, but we continued the treatment to 15 weeks. P2 did not achieve criterion for treatment termination solely in the spoken modality; rather, the combination of her spoken and written responses produced in Step1enabledhertomeetthecriteriononWeek12.After subtracting the time taken for probes and breaks (a 15-min break was given during each session), P1 had 45 hr of treatment, and P2 had Scoring Language measures. Accuracy of participant responses on the WAB, BNT, Object & Action Naming Battery, and NAVS was determined according to guidelines within each test manual, though responses consisting of one phonemic error per lexical item were considered correct for all lexical retrieval tasks except for those on the WAB. In the event of multiple attempts, the best response for the target was accepted. Sentence probes. For probe responses, the scoring system evaluated each content word within sentences. Credit (1 point) was given to responses that matched the target, allowing for one phonemic error (e.g., carfenter for carpenter). However, participants were not penalized for grammatical or morphological errors because these were not targeted in treatment. In the event of multiple attempts, the best response for the target was accepted. Reliability Reliability was determined for a number of measures by the authors, trained research assistants in the University of Florida Aphasia and Bilingualism Lab, and/or speech-language pathologists with experience with aphasia. Reliability was determined by dividing the number of responses agreed upon by the total number of responses scored. Any disagreements in scoring were discussed and resolved by consensus among those involved in the reliability process. Pre- to posttreatment language measures. Interrater reliability was conducted on pre- to posttreatment language measures. One hundred percent reliability was conducted on the WAB, BNT, and NAVS, and 50% reliability was completed on the Object & Action Naming Battery. Scoring agreement was 97%. Treatment. To ensure consistency in execution of the treatment protocol, a trained research assistant observed 25% of the sessions live. The student was provided with a form that listed all required treatment steps for each item being 138 American Journal of Speech-Language Pathology Vol May 2011

9 treated. As she watched the treatment session, she marked off whether each step was conducted correctly. A percentage of reliability was calculated by dividing the total number of steps that were conducted correctly by the total number of steps observed. The treatment protocol was followed with a reliability of 99%. Sentence probes. To ensure consistency in probe scoring, interrater reliability for 100% of weekly probes was conducted. A point-to-point evaluation was conducted, and agreement in responses and scoring was 95%. Error analysis. Interrater reliability was conducted by a trained speech-language pathologist with experience working with persons with aphasia on 70% of all error analyses for preto posttreatment language measures. A point-to-point evaluation was conducted, and agreement in scoring was 96%. Results To evaluate RQ1 and RQ2 (magnitude of change from baseline to posttreatment and baseline to maintenance), effect sizes (d; Beeson & Robey, 2006) were calculated to get an index of durability (Beeson & Robey, 2006, p. 167) at different time points after cessation of treatment. Effect sizes were calculated by dividing the difference of the baseline average from the posttreatment average by the standard deviation of the baseline. For RQ2, the equation was the same, though the maintenance average was used in place of the posttreatment average. A meta-analysis for lexical retrieval treatment studies by Robey and Beeson (2005) resulted in benchmarks of 4.0, 7.0, and 10.1 for small, medium, and large effect sizes, so these numbers are used to aid interpretation of the results below. Additionally, in the following section, we compare our results to those observed in Edmonds et al. (2009). RQ3 evaluates potential changes from the baseline phase to the treatment phase using a C statistic (Tryon, 1982, 1984). There has been concern expressed regarding use of the C statistic for this type of analysis (e.g., Blumberg, 1984; see Tryon, 1984, for a response), so we are conservative in our use and interpretation of the statistic. We first conducted a C statistic on baseline probes. If no significant upward trend was observed ( p <.05), the baseline was deemed stable, and the C statistic was conducted on the baseline and treatment phases as described in Tryon (1982; but see Blumberg, 1984, for some corrections in the equations). If a significant upward trend in the baseline was observed ( p.05), the baseline was deemed unstable (though visual analysis also aided in determining baseline stability), and the modified C statistic that evaluates residuals from trended baseline data and the treatment data was used (Tryon, 1982, pp ). The adjective control task data were treated the same. We used a significance alpha of.01 rather than.05 for our C statistics to defend somewhat against the possibility of a Type I error. Results for RQ1, RQ2, and RQ3 The results for RQ1, RQ2, and RQ3 are described below. See Figures 2 and 3 for visual representation of the data and Table 4 for the raw data. Participant 1 RQ1 and RQ2. We calculated effect sizes from the baseline to posttreatment (1 month posttreatment) and from baseline to maintenance (5 months posttreatment) to evaluate magnitude of change from pre- to posttreatment. For both time points, the effect size was 5.73 for sentence probes containing trained stimuli. For the probes containing semantically related, untrained stimuli, the effect size was 2.98 at 1 month posttreatment and 3.86 at the 5-month maintenance testing. The effect sizes of the adjective control task were considerably lower than for the probe tasks (1.10 at posttreatment and 2.01 at 5-month maintenance). RQ3. There was a slight rise for P1 s trained baseline phase, which was marginally significant (C = 0.58, Z = 1.65, p =.05). A review of the last baseline, which ascends slightly, revealed that no new lexical items were produced in that last baseline probe that were not produced in the third and fourth baseline probes. Nonetheless, we evaluated the baseline to treatment phase as recommended by Tryon (1982) FIGURE 2. Results for Participant 1 s baseline (B), treatment (P), posttreatment (Post), and maintenance (M) probes for sentences containing trained stimuli (TR) and untrained semantically related stimuli (UT) as well as the adjective control task (Cont). Edmonds & Babb: Verb Network Strengthening Treatment 139

10 FIGURE 3. (Top) Results for Participant 2 s (P2 s) baseline, treatment, posttreatment, and maintenance probes for sentences containing trained stimuli and untrained semantically related stimuli as well as the adjective control task in the spoken modality. (Bottom) Results for P2 s baseline, treatment, posttreatment, and maintenance probes for sentences containing trained stimuli and untrained semantically related stimuli in the written modality. when an ascending baseline is observed. Using this approach, there was no significant difference between baseline and treatment (C = 0.56, Z = 1.58, p =.94). According to the C statistic, there was a rise in P1 s untrained baseline phase (C = 0.62, Z = 1.74, p =.04). This rise reflected an increase in production from Baseline Point 2 to 3, which flattened after that point and thus, based on visual inspection, had stabilized. However, to be conservative, we calculated the C statistic assuming a rising baseline and found no significant difference between the slopes of the baseline and the treatment phase (C = 0.18, Z = 0.51, p =.31). The baseline of the adjective control task was stable as indicated by the C statistic (C = 0.48, Z = 1.35, p =.09), with no statistical increase from the baseline to the treatment phase (C = 0.24, Z = 1.11, p =.13). These results, in addition to the low effect sizes of the control task, indicate that experimental control was established. Participant 2 RQ1 and RQ2 spoken probes. For P2 s spoken probes, the effect size for pre- to posttreatment was 10, indicating a large effect. This number dropped to 5.61, indicating a small effect, at 1-month maintenance. For the probes containing untrained stimuli, the effect size posttreatment was 5.66 (small effect), and the effect size dropped slightly to 4.24 (small effect) at 1-month maintenance. The effect sizes for the adjective control task were considerably lower (1.04 posttreatment and1.79at1-monthmaintenance). RQ3 spoken probes. Both of P2 s spoken probe baselines were stable as indicated by the C statistic (trained: C = 0.42, Z = 1.20, p =.12; untrained: C = 0.25, Z = 0.71, p =.24), so we used the conventional C statistic (Tryon, 1982). The results showed a significant difference between baseline and treatment for probes containing trained (C =0.63,Z =2.99, p =.001) and untrained stimuli (C =0.72,Z =3.37,p =.000). The baseline for the adjective control task was stable prior to treatment (C = 0.37, Z = 1.10, p =.14), though there was a significant increase from baseline to treatment (C = 0.57, Z = 2.70, p =.003). This result appears to be due to a decreasing trend in the baseline (ones to zeroes) with an ascending trend in treatment (scores of many zeroes to scores of mostly ones), not due to an actual increase in performance. Thus, we assert that experimental control was maintained. RQ1, RQ2, and RQ3 written probes. Baseline to posttreatment and baseline to maintenance effect sizes were also calculated for the written probes. For trained stimuli, the effect sizes were large (18.17 posttreatment and 13.5 at 1-month 140 American Journal of Speech-Language Pathology Vol May 2011

11 TABLE 4. Raw data for sentence probes containing trained (TR) and untrained (UT) semantically related stimuli and adjective control task for both participants. Session P1 maintenance). For untrained stimuli, the effect sizes were less than the benchmark for a small effect (2.86 posttreatment and 1.22 at 1-month maintenance). Both of P2 s written probe baselines were stable as indicated by the C statistic (trained: C = 0.33, Z = 0.91, p =.82; untrained: C = 0.50, Z = 1.37, p =.91), and there were significant changes in slopes from baseline to treatment for probe sentences containing trained and untrained stimuli (C = 0.68, Z = 3.13, p=.001; C = 0.59, Z = 2.79, p =.003, respectively). Results for RQ4a d The results for the dependent measures related to RQ4 are described below. Holm s sequential correction (Holm, 1979) was applied to the noun and verb naming (BNT and Object & Action Naming Battery) and sentence production (NAVS) results, for which a McNemar test was conducted. Using this, the most significant of the three p values has to be less than.017 (.05/3) to be significant, the second most significant has to be less than.025 (.05/2), and the third most significant has to be less than.050. Participant 1 RQ4a, b. P1 s performance on noun retrieval on the BNT remained constant from pre- to posttreatment (35% to 36.6%), while performance on nouns on the Object & Action Naming Battery increased from 45.7% to 65.4% ( p =.024), though P2 Spoken Spoken Written TR UT Control TR UT Control TR UT B B B B B P P P P P P P P P P P P P P P Post Post Post M Note. B = baseline; P = treatment; Post = posttreatment; M = maintenance. not significantly with the Holm s correction. No increase in single verb naming was observed on the Object & Action Naming Battery (56% to 50%; p =.571). P1 s ability to produce sentences on the NAVS remained unchanged, with 25% correct pre- and posttreatment (see Figure 3). RQ4c. On the Nicholas and Brookshire (1993) connected speech tasks, P1 produced more words from pre- to posttreatment (79 to 100), increased her production of CIUs (67.87% to 83.16%), and decreased production of neologisms from 8.78% to 0%. Statistical analyses were not conducted on these data due to the low number of lexical items in the sample. RQ4d. Examination of pre- to posttreatment changes on P1 s mother s responses on the CETI was conducted using a Wilcoxon signed-ranks test. She was not shown her responses to the pretreatment questionnaire while completing the posttreatment questionnaire. The posttreatment responses indicated a significant increase in ratings of functional communication (Z = 2.045, p =.041). Participant 2 RQ4a, b. In the spoken modality, P2 improved in her ability to retrieve nouns with a significant increase from 8.3% to 26.6% on the BNT ( p =.007) and from 16% to 48.1% on the Object & Action Naming Battery ( p =.000). Verb retrieval improved significantly from 12% to 42% correct on the Object & Action Naming Battery ( p =.000). P2 s overall ability to produce sentences on the NAVS improved pre- to posttreatment from 0% to 22.2% correct ( p =.008). In the written modality, noun retrieval did not significantly improve on the Object & Action Naming Battery (77.7% to 85.2%; p =.180), but verb retrieval did (66% to 86% accuracy; p =.006; see Table 1). RQ4c. P2 s Nicholas and Brookshire (1993) language samples increased from 35 words pretreatment to 117 words posttreatment, and her use of neologisms decreased from 40.02% to 26.17%. However, her CIUs dropped from 67% to 57%, though the decrease might have been the result of inflated pretreatment CIUs relative to the few words produced. Statistics were not conducted on these results due to the short length of the samples. RQ4d. Examination of pre- to posttreatment changes on the CETI for P2 and her mother was conducted using a Wilcoxon signed-ranks test. Their responses from the pretreatment questionnaire were not shown to the participants when they completed the posttreatment questionnaire. The posttreatment responses indicated a significant increase in ratings of functional communication (P2: Z = 2.28, p =.023; P2 s mother: Z = 3.47, p =.001; see Table 1 for details). Results for RQ5: Error Analyses A general description of error analysis trends for each participant for single-word naming (from the BNT and Object & Action Naming Battery) and on sentences from the NAVS is provided below (see Tables 2 and 3 for details). Participant 1. The most notable changes for noun naming (BNT and Object & Action Naming Battery combined) for P1 were a reduction in IDK/NR (20.5% to 6.0%) and neologistic responses (8.5% to 3.1%), with a corresponding increase in semantic and phonemic errors. Pretreatment, half Edmonds & Babb: Verb Network Strengthening Treatment 141

12 (51%) of the verb productions were IDK/NR or neologistic responses, while posttreatment there were no IDK/NR or neologistic responses. This change corresponded with a large increase in semantic errors (18.2% to 56%), with a moderate increase in phonemic (0% to 4%) and mixed (4.5% to 12%) errors. We analyzed each word in the sentences on the NAVS, but we focused our attention on verbs and direct objects due to a low number of errors on subjects and indirect objects. Despite no increase in accuracy or decrease in the number of errors, P1 showed some increase in verb processing, with a reduction in omissions and a moderate increase in unrelated, semantic, and phonemic errors. A more dramatic increase away from omissions toward semantic errors was observed with direct objects. Participant 2. In noun naming, P2 exhibited a notable reduction in IDK/NR errors (37% to 1%) with an increase in neologisms from 41% to 76%. Increases to semantic and phonemic errors were modest. For verbs, there was a dramatic drop of IDK/NR errors in verb naming from 75% to 0%. Similar to her noun production, there was a corresponding increase of neologistic errors (9% to 59%) with modest increases in phonemic, unrelated, and mixed errors. Overall P2 s written naming responses were more accurate and different in error type than her spoken responses. Pretreatment, the predominant error type for nouns and verbs was semantic (33% and 53%, respectively). Graphemic errors were observed for nouns (17%) but not for verbs. Posttreatment, noun errors were split equally among semantic and graphemic errors (41.7% for each), whereas verb errors were still primarily semantic (up slightly to 57%). P2 did not produce any correct sentences on the NAVS before treatment, and her primary error for all words was omission. Posttreatment, there was an increase in accuracy (to 25%) with a corresponding reduction in omissions in incorrect sentences. Similar to her single-word naming, a decrease in omissions corresponded to an increase in neologisms, though increases in semantic errors were observed across subjects (18% to 37.5%), verbs (3% to 19%), and direct objects (4% to 9%). Discussion The first study with VNeST (Edmonds et al., 2009) reported encouraging results in four persons with moderate aphasia. The purpose of the current Phase II study was to examine the effect of VNeST on persons with more severe aphasia. We predicted that the participants would exhibit the same trend of generalization to sentence probes as observed in the first study but that the improvement would require more treatment sessions and be less pronounced. On additional lexical retrieval generalization measures, we predicted an increase in single-word noun and verb naming, but we did not have specific predictions for sentence production or discourse given the severity of the participants. Based on anecdotal reports of improved communication from participants and their families in Edmonds et al. (2009), we also predicted increased functional communication for the participants as indicated by responses to the CETI (Lomas et al., 1989). The overall results showed different responses to treatment across participants, with P1 exhibiting limited effects and P2 exhibiting more widespread effects. P1 s pre- to posttreatment effect sizes (RQ1) for trained items were small, with a negligible effect for untrained items. Maintenance of gains was observed 5 months posttreatment (RQ2). P2 s effect sizes were greater, with large effects for trained and untrained probe stimuli, with maintenance scores still exhibiting a small effect for both probe sets. Maintenance results were promising, but further studies are needed across more participants to evaluate maintenance within the context of participant (e.g., severity) and treatment (e.g., length of treatment) variables. In general, the trends indicated by the effect sizes are consistent with both participants C-statistic results (RQ3), with the exception of the C-statistic results for P1 s trained condition. However, the C-statistic analysis, which evaluates potential change in slope from baseline to treatment, addresses a different question than the pre- to posttreatment effect sizes. Though the results of the C statistic generally support that P2 showed more treatment effect than P1 from baseline to treatment, we interpret the results cautiously, as the significant findings could result from an increase in power from adding a large number of points to the series in the treatment phase. Thus, as indicated in our introduction, and suggested by Beeson and Robey (2006), we use effect sizes and evidence of experimental control as our primary indicators of a treatment effect in this early phase study. Sentence probes represent the dependent measure most theoretically related to treatment. Presumably, an increase in probes containing trained stimuli indicates that the repeated activation and retrieval of agent-patient pairs around a verb during treatment facilitate increased activation and retrieval of the target items in the more constrained picture description task. Increased lexical retrieval to sentences containing untrained semantically related verbs suggests spreading activation from the trained verb to the untrained verb and its associated thematic roles. Thus, the probe findings suggest that the treatment effects were limited to trained verb networks for P1, whereas P2 showed the expected generalization to both probe sets. VNeST involves activation of large semantic networks that can potentially result in generalization to lexical retrieval beyond probe stimuli (Edmonds et al., 2009). Theoretically, generalization to untrained probes should occur if other, more distantly related items are also to improve. This appears to be mostly the case for P1, as she did not improve on verb naming or sentence production on the NAVS (RQ4). However, noun naming did increase by 20% on the Object & Action Naming Battery. Noun improvement in the absence of verb improvement follows logically from the treatment protocol, where participants retrieve approximately six nouns for each verb. Unlike P1, P2 exhibited fairly dramatic generalization to all lexical retrieval generalization measures. It is unclear why P1 did not improve to the extent P2 did, though we can exclude several possibilities. First we can rule out aphasia severity and motor programming impairments, since P2 had a lower Aphasia Quotient than P1 and both exhibited a mild apraxia of speech. P1 had slightly lower scores on semantic testing and comprehension on the WAB, but 142 American Journal of Speech-Language Pathology Vol May 2011

13 these differences were not dramatic and would not explain the differences in findings. Further, the participants had similar cognitive profiles as measured by the Cognitive Linguistic Quick Test. However, we did not do in-depth testing of various cognitive functions such as attention, working memory, and other processing abilities that could affect response to treatment. P1 was more chronic than P2, but persons who were more chronic than P1 showed improvement in Edmonds et al. (2009). Finally, spontaneous recovery for P2 can be ruled out due to lack of improvement on the adjective control task and a number of production subtests on the WAB. One strong possibility for the difference in generalization is the slight difference in treatment protocol for P1 and P2. Recall that P1 followed the protocol of Edmonds et al. (2009), where only the agent and patient were read/repeated in Step 2. The protocol was adjusted for P2 to require her to retrieve the agent, verb, and patient during Step 2 of treatment. Thus, P1 would have read aloud the word pairs carpenter-lumber, chef-sugar, and surveyor-land when working on the verb measure, whereas P2 would have read the agent, verb, and patient that composed the sentence (carpenter-measure-lumber, etc.). Given P1 s proposed disruption to the connection between the semantic system and the POL and to the POL itself, this lack of practice at retrieving the verb may have limited generalization to probes with trained stimuli and noun naming, whereas the extra practice P2 received may have contributed to her greater generalization to verbs and sentences containing untrained stimuli. Although P1 showed limited generalization, there were indications of clinical/functional improvements. First, she made clinically meaningful improvements on the WAB (Katz & Wertz, 1997). Second, on the connected speech task, she produced more words from pre- to posttreatment, increased her production of CIUs, and decreased production of neologisms. Given the short length of the language samples, we interpret these results cautiously, but the findings suggest an increase in lexical retrieval of relevant words and intelligibility in tasks less constrained than naming and single-sentence production to pictures. We do not know whether these skills extended beyond the clinic into more functional situations, but the results of her mother s responses to the CETI suggest that P1 did improve in her functional communication abilities. The types of tasks on which P1 s mother indicated improvement (getting someone s attention, getting involved in group conversations about herself, giving appropriate yes/no answers, and communicating emotions) were consistent with those we observed in the clinic. However, little or no improvement was reported on more complex tasks such as having coffee-time visits with friends and neighbors or participating in conversation with strangers, as would be expected. P1 s mother also mentioned anecdotally that P1 started attending more family functions and socializing more with family and friends. There are indications that P2 made some clinical/functional gains in addition to her improvements on generalization measures. Like P1, she showed clinical improvement on her WAB score (Katz & Wertz, 1997). She also showed signs of improvement in the connected speech task, with a large increase in words and a decrease in neologisms. Additionally,P2andhermotherbothreportedimprovedfunctional communication based on their responses to the CETI, indicating that the improvement P2 exhibited on all generalization measures may have transferred to functional communication. Like P1 s mother, P2 and her mother indicated improvements on the simpler communication tasks, which were consistent with our observations. Since a primary goal of VNeST is to promote participantgenerated responses (as opposed to clinician-generated), we allowed P1 to write responses that she was unable to produce orally when generating agents and patients in Step 1 of treatment, though she was always required to read/repeat the word afterward. We then evaluated the effect of treatment on written probes, which showed large effects that were well maintained for sentences containing trained stimuli. Though there was statistical significance on untrained probes, based on visual inspection and effect sizes, we assert that there was no generalization to these sentences. We did not evaluate writing in discourse or functional situations, so we do not know if there was generalization beyond the trained stimuli. The writing results suggest that there may be potential for cross-modal generalization with VNeST, though future studies must examine this possibility in a more controlled manner. By extending VNeST to writing, it may be possible to increase communication for participants who use writing as part of their daily communication, including persons with severe apraxia of speech. Despite differences in generalization, both participants posttreatment error patterns showed an evolution of responses that suggest increased processing during lexical retrieval. Across all tasks, P1 made more attempts at production, and her posttreatment errors were more semantically and phonologically related to the target. P2 showed reductions in omissions with increases in neologisms, consistent with her hypothesized POL impairment, though some increases in semantic and phonemic errors were observed. Consistent with the overall generalization patterns, P1 showed only modest indications of improvement on the NAVS, while P2 showed more marked evolution of errors, including fewer verb errors overall. We hypothesized that both participants had a relatively intact semantic system with disruption to the connection between the semantic system and POL for both participants and disruption to the POL itself (worse for P2). Thus, it may seem counterintuitive that P2 exhibited such widespread gains given that VNeST is fundamentally a semantic treatment. However, there are a number of factors that would predict potential improvement in persons with largely phonological impairments. First, while VNeST s theoretical foundation is rooted in semantic theory, the treatment materials and paradigm include access to the orthography and phonology of the trained words. Second, interactive models of lexical retrieval posit bidirectional spreading activation between the semantic, lexical, and phonological levels during word production (Dell, 1986), so all of these levels are theoretically activated during treatment. Finally, there is precedent for similar findings. In her review of the literature, Nickels (2002) noted that the majority of individuals with impaired retrieval of phonological forms seem to benefit from tasks that combine semantic and phonological activation. The individuals who benefit can have impairments at a Edmonds & Babb: Verb Network Strengthening Treatment 143

14 semantic level or in retrieval of the phonological form (in the absence of a semantic impairment) (p. 954). Thus, participants do not necessarily have to have primarily semantic level deficits to benefit from VNeST, though more research is needed to define more precisely the types of participants who might most benefit. An interesting finding on posttreatment probe errors was nominalization of verbs with their instruments (e.g., sew Y needle; slice Y knife). Verbs prime not only agents and patients but also instruments (Ferretti et al., 2001; McRae et al., 2005), so the concept of the instrument is inherent to the verb. Thus, it follows that the instrument would be activated during VNeST. For example, when training verbs such as slice, measure, sew, or write, the instruments of knife, tape (measure)/ruler, needle, and pen/pencil are theoretically activated. Both participants produced these instrument names in place of verbs in posttreatment probes, which supports the assumption of spreading activation within and beyond the trained verb network. It is possible that explicit naming of an instrument during treatment is required to distinguish it from the action being trained, especially for persons with selective verb impairments. This possibility needs to be explored in further research. As predicted, the participants in the current study took longer to achieve treatment criterion, roughly twice as long, compared with less impaired participants. Effect sizes were not reported in Edmonds et al. (2009), but we calculated them here for two of the participants in that study (effect sizes for the other two participants could not be calculated due to zeros in all their baselines, but their improvement was roughly similar to that of the other two participants). One, who had nonfluent aphasia, had an effect size of 15 for probes with trained and untrained stimuli, and the other, who was fluent, had effect sizes of 19 and 25 for trained and untrained, respectively. Thus, as predicted, the effect sizes for both moderately impaired participants were larger than those for the current study, especially considering the probes for the first study evaluated sentences, and those in the present study evaluated lexical retrieval within sentences. Though the findings from this study and Edmonds et al. (2009) are preliminary, there are some clinical implications that are beginning to emerge. First, VNeST may be appropriate for persons with moderate and moderate-to-severe aphasia, though those with more severe aphasia will likely need more hours of therapy to yield less improvement than persons with more moderate impairments. With respect to the protocol, the findings from the current study suggest that it may be crucial to require participants to retrieve the verb during Step 2 in order to maximize generalization, and future research with VNeST will adopt this change. Further, the protocol of VNeST is systematic and offers opportunity for repetition while encouraging personal and salient responses from the participant, both important factors in aphasia rehabilitation (Raymer et al., 2008). Finally, VNeST may result in functional gains and cross-modal generalization, though more research is needed. Despite the encouraging findings related to VNeST, more research is warranted. Some issues we are pursuing include refinement of our understanding of the semantic relationships between verbs and their thematic roles, including instruments and locations, so we can make more precise predictions about generalization. Related to this is defining what treatment steps are crucial so the protocol can be streamlined. Treatment intensity and duration also need to be examined as well as the relationship between participant factors (e.g., pretreatment cognitive and linguistic skills, and access to communication partners) and outcome. Acknowledgments We would like to thank the research participants and their families for their enthusiasm and motivation throughout the treatment process. We would also like to extend our thanks to the following students in the University of Florida Aphasia and Bilingualism Lab who assisted with this study: Liz Owens, Amanda Napolitano, and Megan Ferguson. Finally, we would like to thank Will Hula for helpful discussions regarding statistical analysis. References Antonucci, S. M. (2009). The use of semantic feature analysis in group aphasia treatment. Aphasiology, 23, Bak, T. H., & Hodges, J. R. (2003). Kissing and dancing a test to distinguish the lexical and conceptual contributions to noun/ verb and action/object dissociation. Preliminary results in patients with frontotemporal dementia. Journal of Neurolinguistics, 16, Beeson, P. M., & Robey, R. R. (2006). Evaluating single-subject treatment research: Lessons learned from the aphasia literature. Neuropsychology Review, 16, Blumberg, C. J. (1984). Comments on A simplified time-series analysis for evaluating treatment interventions. Journal of Applied Behavior Analysis, 17, Boyle, M. (2004). Semantic feature analysis treatment for anomia in two fluent aphasia syndromes. American Journal of Speech- Language Pathology, 13, Boyle, M. (2009, September). Semantic feature analysis treatment for aphasic word retrieval impairments: Variations on a theme. Paper presented at the 4th Annual Eleanor M. Saffran Cognitive Neuroscience Conference, Philadelphia, PA. Boyle, M., & Coelho, C. A. (1995). Application of semantic feature analysis as a treatment for aphasic dysnomia. American Journal of Speech-Language Pathology, 4(4), Dabul, B. (2000). Apraxia Battery for Adults, Second Edition. Austin, TX: Pro-Ed. Dell, G. S. (1986). A spreading-activation theory of retrieval in sentence production. Psychological Review, 93, Dell, G. S., Schwartz, M. F., Martin, N. M., Saffran, E. M., & Gagnon, D. A. (1997). Lexical access in aphasic and nonaphasic speakers. Psychological Review, 104, Druks, J., & Masterson, J. (2000). An Object & Action Naming Battery. Hove, United Kingdom: Psychology Press. Edmonds, L. A., & Kiran, S. (2006). Effect of semantic naming treatment on crosslinguistic generalization in bilingual aphasia. Journal of Speech, Language, and Hearing Research, 49, Edmonds, L., Nadeau, S., & Kiran, S. (2009). Effect of Verb Network Strengthening Treatment (VNeST) on lexical retrieval of content words in sentences in persons with aphasia. Aphasiology, 23, Edwards, S., Tucker, K., & McCann, C. (2004). The contribution of verb retrieval to sentence construction: A clinical study. Brain and Language, 91, Ferretti, T. R., McRae, K., & Hatherell, A. (2001). Integrating verbs, situation schemas, and thematic role concepts. Journal of Memory and Language, 44, American Journal of Speech-Language Pathology Vol May 2011

15 Fink, R. B., Martin, N., Schwartz, M. F., Saffran, E. M., & Myers, J. L. (1992). Facilitation of verb retrieval skills in aphasia: A comparison of two approaches. Clinical Aphasiology, 21, Goodglass, H., Kaplan, E., & Weintraub, S. (1983). Boston Naming Test. Philadelphia, PA: Lea & Febiger. Helm-Estabrooks, N. (2001). Cognitive Linguistic Quick Test. San Antonio, TX: The Psychological Corporation. Holm, S. (1979). A simple sequentially rejective multiple test procedure. Scandinavian Journal of Statistics, 6(2), Howard, D., & Patterson, K. (1992). The Pyramids and Palm Trees Test. London, England: Harcourt Assessment. Katz, R. C., & Wertz, R. T. (1997). The efficacy of computerprovided reading treatment for chronic aphasic adults. Journal of Speech, Language, and Hearing Research, 40, Kay, J., Lesser, R., & Coltheart, M. (1992). Psycholinguistic Assessments of Language Processing in Aphasia. New York, NY: Psychology Press. Kertesz, A. (1982). Western Aphasia Battery. Austin, TX: Pro-Ed. Khader, P., Scherag, A., Streb, J., & Rösler, F. (2003). Differences between noun and verb processing in a minimal phrase context: A semantic priming study using event-related brain potential. Cognitive Brain Research, 17, Kiran, S., & Bassetto, G. (2008). Evaluating the effectiveness of semantic-based treatment for naming deficits in aphasia: What works? Seminars in Speech and Language, 29(1), Kiran, S., & Thompson, C. K. (2003). The role of semantic complexity in treatment of naming deficits: Training semantic categories in fluent aphasia by controlling exemplar typicality. Journal of Speech, Language, and Hearing Research, 46, Laine, M., & Martin, N. (2006). Anomia: Theoretical and clinical aspects. New York, NY: Psychology Press. Law, S., Wong, W., Sung, F., & Hon, J. (2006). A study of semantic treatment of three Chinese anomic patients. Neuropsychological Rehabilitation, 16, Lomas, J., Pickard, L., Bester, S., Elbard, H., Finlayson, A., & Zoghaib, C. (1989). The Communicative Effectiveness Index: Development and psychometric evaluation of a functional communication measure for adult aphasia. Journal of Speech and Hearing Disorders, 54, Loverso, F. L., Prescott, T. E., & Selinger, M. (1988). Cueing verbs: A treatment strategy for aphasic adults (CVT). Journal of Rehabilitation Research, 25(2), McKoon, G., Ratcliff, R., & Dell, G. S. (1986). A critical evaluation of the semantic-episodic distinction. Journal of Experimental Psychology: Learning, Memory, and Cognition, 12, McNeil, M. R., Robin, D. A., & Schmidt, R. A. (2009). Apraxia of speech. In M. R. McNeil (Ed.), Clinical management of sensorimotor speech disorders (2nd ed., pp ). New York, NY: Thieme. McRae, K., Hare, E., & Ferretti, T. R. (2005). A basis for generating expectancies for verbs from nouns. Memory and Cognition, 33, McRae, K., Ferretti, T. R., & Amyote, L. (1997). Thematic roles as verb-specific concepts. Language and Cognitive Processes, 12(2/3), Nicholas, L. E., & Brookshire, R. H. (1993). A system for quantifying the informativeness and efficiency of the connected speech of adults with aphasia. Journal of Speech and Hearing Research, 36, Nickels, L. (2002). Therapy for naming disorders: Revisiting, revising, and reviewing. Aphasiology, 16, Prescott, T. E., Selinger, M., & Loverso, F. L. (1982). An analysis of learning, generalization, and maintenance of verbs by an aphasic patient. In R. H. Brookshire (Ed.), Clinical Aphasiology Conference Proceedings (pp ). Minneapolis, MN: BRK. Raymer, A. M., Beeson, P., Holland, A., Kendall, D., Maher, L. M., Martin, N.,... Rothi, L. J. G. (2008). Translational research in aphasia: From neuroscience to neurorehabilitation. Journal of Speech, Language, and Hearing Research, 51, S259 S275. Raymer, A. M., Ciampitti, M., Holliway, B., Singletary, F., Blonder, L. X., Ketterson, T.,... Rothi, L. J. G. (2007). Semantic-phonologic treatment for noun and verb impairments in aphasia. Neuropsychological Rehabilitation, 17(2), Raymer, A. M., & Ellsworth, T. A. (2002). Response to contrasting verb retrieval treatments: A case study. Aphasiology, 16, Rider, J. D., Wright, H. H., Marshall, R. C., & Page, J. L. (2008). Using semantic feature analysis to improve contextual discourse in adults with aphasia. American Journal of Speech- Language Pathology, 17, Robey, R. R. (2004). A five-phase model for clinical-outcome research. Journal of Communication Disorders, 37, Robey, R. R., & Beeson, P. M. (2005, November). Aphasia treatment: Examining the evidence. Paper presented at the Annual Convention of the American Speech-Language-Hearing Association, San Diego, CA. Rösler, F., Streb, J., & Haan, H. (2001). Event-related brain potentials evoked by verbs and nouns in primed lexical decision task. Psychophysiology, 38, Schneider, S. L., & Thompson, C. K. (2003). Verb production in agrammatic aphasia: The influence of semantic class and argument structure properties on generalisation. Aphasiology, 17, Snowden, J. S., & Neary, D. (2002). Relearning of verbal labels in semantic dementia. Neuropsychologia, 40, Thompson, C. K. (2002). The Northwestern Verb Production Battery/The Northwestern Assessment of Verbs and Sentences. Unpublished manuscript. Thompson, C. K. (2006). Single subject controlled experiments in aphasia: The science and the state of the science. Journal of Communication Disorders, 39, Tryon, W. W. (1982). A simplified time-series analysis for evaluating treatment interventions. Journal of Applied Behavior Analysis, 15, Tryon, W. W. (1984). A simplified time-series analysis for evaluating treatment interventions : A rejoinder to Blumberg. Journal of Applied Behavior Analysis, 17, Wambaugh, J. L., & Ferguson, M. (2007). Application of semantic feature analysis to retrieval of action names in aphasia. Journal of Rehabilitation Research and Development, 44, Wambaugh, J. L., Linebaugh, C. W., Doyle, P. J., Martinez, A. L., Kainyak-Fliszar, M., & Spencer, K. A. (2001). Effects of two cueing treatments on lexical retrieval in aphasic speakers with different levels of deficit. Aphasiology, 15, Edmonds & Babb: Verb Network Strengthening Treatment 145

16 Effect of Verb Network Strengthening Treatment in Moderate-to-Severe Aphasia Lisa A. Edmonds, and Michelle Babb Am J Speech Lang Pathol 2011;20; ; originally published online Mar 8, 2011; DOI: / (2011/ ) The references for this article include 8 HighWire-hosted articles which you can access for free at: This information is current as of May 2, 2011 This article, along with updated information and services, is located on the World Wide Web at: