بهبود زبان بدون کلمات: شواهد نخست از زبان پریشی

The pervasiveness of word-finding difficulties in aphasia has motivated several theories regarding management of the deficit and its effectiveness. Recently, the hypothesis was advanced that instead of simply accompanying speech gestures participate in language production by increasing the semantic activation of words grounded in sensory-motor features, hence facilitating retrieval of the word form. Based on this assumption, several studies have developed rehabilitation therapies in which the use of gestures reinforces word recovery. Until now, however, no studies have investigated the beneficial effects of gesture observation in word retrieval. Here, we report whether a different modality of accessing action-motor representation interacts with language by promoting long lasting recovery of verb retrieval deficits in aphasic patients. Six aphasic participants with a selective deficit in verb retrieval participated in an intensive rehabilitation training that included three daily sessions over two consecutive weeks. Each session corresponded to a different rehabilitation procedure: (1) “action observation”, (2) “action observation and execution”, and (3) “action observation and meaningless movement”. In the four participants with lexical phonologically based disturbances, significant improvement of verb retrieval was found only with “action observation” and “action observation and execution”. No significant differences were present between the two procedures. Moreover, the follow-up testing revealed long-term verb recovery that was still present two months after the two treatments ended. In support of a multimodal representation of action, these findings univocally demonstrate that gestures interact with the speech production system, inducing long-lasting modification at the lexical level in patients with cerebral damage.

The hypothesis that gestures play an important role in lexical retrieval dates back to the beginning of the twentieth century (DeLaguna, 1927, Dobrogaev, 1929 and Mead, 1934). In the earliest published study, Dobrogaev (1929) reported that speakers instructed to inhibit facial expressions and gestural movements of the extremities found it difficult to produce articulate speech. More recently, Rimé (1982) and Rauscher, Krauss, and Chen (1996) showed that preventing gestures affected speech fluency adversely; in fact, the effects were similar to those found when word retrieval was prevented by other means (i.e., when subjects were requested to use rare or unusual words). More evidence supporting the interaction between gestures and lexical retrieval comes from studies of brain-damaged patients. Hadar, Wenkert-Olenik, Krauss, and Soroker (1998) reported that aphasics whose speech problems primarily concerned word retrieval tended to gesture more than both normal controls and other aphasics whose problems lay at a conceptual level. About 70% of the gestures of patients with word retrieval difficulties were associated with a hesitation or an erroneous production. Thus, viewed in relation to speech, it appears that aphasic patients have involved a compensatory strategy by increasing gesture production. According to these data, gestures and speech are two separate communication systems and gestures function as an auxiliary support when verbal expression is temporally disrupted or word retrieval is difficult (Hadar, 1989, Hadar et al., 1998 and Krauss and Hadar, 1999). Based on this assumption, several studies have proposed rehabilitation therapies in which the use of simple gestures or pantomime paired with verbal production improved word recovery (Code and Gaunt, 1986, Hanlon et al., 1990, Raimer et al., 2006, Rodriquez et al., 2006 and Rose et al., 2002). Code and Gaunt (1986) wanted to examine whether combining gesture and speech would improve or hindered the production of either or both systems in a patient with severe apraxia and aphasia. They found significant improvement in the patient's ability to produce a small range of useful hand signs, especially on those enhanced through word-cued gesture (i.e., where the word equivalent to the gesture was cued by the therapist and the gesture was required as response) and gesture-cued word (where the therapist gave the gesture and the patient was required to produce the word as response) facilitations. Additionally, there was some indication that access to single-word production was facilitated when the patient was cued with an appropriate hand sign, and access to hand signs was likewise facilitated when the patient was cued with an appropriate word. In the Hanlon et al.’s work (Hanlon et al., 1990) the effects of different unilateral gestural movements on naming to confrontation were examined. It was hypothesized that activating the hemiplegic right arm to execute a communicative but non-representational pointing gesture would have a facilitatory effect on aphasics’ naming ability. Results showed that gestures produced through activation of the proximal (shoulder) muscolature of the right paralytic limb facilitated naming performance. Gestures paired with verbal production have frequently been used to treat naming impairments in patients with aphasia (Pashek, 1998, Raimer and Thompson, 1991, Richards et al., 2002 and Rose et al., 2002). Rose and colleagues (Rose and Douglas, 2001 and Rose et al., 2002) noted that gestural treatment using pantomimes was more effective in individuals with a phonologically based word retrieval impairment than in those with semantically based word failure. Raimer et al. (2006) examined the effect of pantomime paired with verbal training for noun and verb retrieval in a group of aphasic participants. Effects were evaluated in spoken naming to pictured objects and actions. Results showed that naming improvements were present for trained nouns and verbs but not for untrained words. Contrary to the assumption of a functional separation between gestures and speech, another hypothesis suggests that the two systems are closely linked to the same conceptual representation (McNeill, 1992). In line with Martin et al.’s proposal (Martin, Wiggs, Ungerleider, & Haxby, 2000), it is assumed that the semantic representation of a concept is composed not only of stored information about the features defining that concept, such as its typical form, color and motion but also of the motor movement associated with its use. Semantic representation of word concepts can be encoded in both propositional and non-propositional formats, and words whose retrieval is facilitated by gestures are more likely to be analogically encoded in sensory-motor features (Krauss et al., 2000 and Krauss and Hadar, 1999). In the embodied cognition view, there is “no language module” and the representation of a concept is crucially dependent upon sensory-motor processes related to that concept (Barsalou, 1999, Gallese and Lakoff, 2005 and Rizzolatti and Craighero, 2004). Several lines of evidence have already demonstrated a strong connection between language and action, particularly with regard to language comprehension. Words mediating actions performed with different motor districts (e.g. the feet ‘kick’, the hands ‘pick’ and the mouth ‘licks’) enhance the same neural substrates involved in executing those actions (Binkofski and Buccino, 2006, Fadiga et al., 2002, Hauk and Pulvermuller, 2004, Pulvermuller et al., 2005 and Rizzolatti et al., 2001). Similarly, in a behavioural study Sato, Mengarelli, Riggio, Gallese, and Buccino (2008) found slower responses with the right hand when subjects had to categorize hand-action-related verbs semantically than when the task involved foot-action-related verbs. Conversely, it has been showed that gesture execution influences word comprehension and production also when subjects are simply asked to observe the performed action (Bernardis and Gentilucci, 2006, Gentilucci and Dalla Volta, 2008 and Gentilucci et al., 2008). These results are in accordance with the hypothesis of a shared motor representation for the execution and observation of actions (the so-called “mirror neuron” theory) (Rizzolatti and Arbib, 1998 and Rizzolatti et al., 1999). This motor representation, by matching observation with execution, makes it possible for individuals to recognize and understand the meaning of actions performed by others (Gallese et al., 1996 and Rizzolatti et al., 1996). Accordingly, brain-imaging studies have indicated that Broadmann's area 44 (BA44) which is located in the pars opercularis of the inferior frontal gyrus, together with the superior temporal sulcus and the inferior parietal lobule, may serve as a core neural network for action understanding (Binkofsky et al., 1999, Buccino et al., 2001, Fadiga et al., 1995, Rizzolatti et al., 2000 and Zadeh et al., 2006). This fronto-parietal network has reciprocal connection in the underlying white matter located in the superior longitudinal fasciculus (SFL). The most inferior branch of SFL originates from the rostral portion of the inferior parietal lobule (Broadmann's area 40) and terminates in ventral area 6, area 44 and are 9/46 (Petrides & Pandia, 2002). In Gentilucci et al.’s works (Bernardis and Gentilucci, 2006, Gentilucci and Dalla Volta, 2008 and Gentilucci et al., 2008), the execution of meaningful gestures modified the voice spectra of words that had the same meaning, but not of meaningless words (i.e., pseudo-words). Moreover, observing a meaningful gesture affected verbal responses in the same way as executing the same gesture. The authors concluded that the spoken word and the symbolic gesture are coded as a single signal by a unique communication system. Nevertheless, it is still an open question to what extent this interaction works and at which level of the language production system gestures might exert their influence. The more traditional view has suggested that gestural information might contribute to the construction of the speaker's communicative intention and might affect lexical retrieval only indirectly (Hadar and Butterworth, 1997, Hadar et al., 1998 and Hanlon et al., 1990); more recent works, however, have indicated that gestures and language production closely interact at least at a motor/articulatory level (Bernardis and Gentilucci, 2006, Gentilucci and Dalla Volta, 2008 and Gentilucci et al., 2008). In this study, we investigated whether observing gestures exert its influence in the language production system also at a lexical level by promoting long-lasting recovery of word retrieval deficits in aphasic patients. As far as we know, no other studies have previously addressed this issue. In most of the previous treatments, gestures were combined with a verbal cue (Pashek, 1998, Raimer and Thompson, 1991, Richards et al., 2002 and Rose et al., 2002) and when they were used as the only facilitation, they were not semantically related with the action (Hanlon et al., 1990). With regard to gesture observation, while the studies univocally addressed their crucial role for language comprehension, no studies have been reported on the relationship between gestures and lexical retrieval. Specifically, we were interested in exploring whether “the observation of semantically congruent actions” and/or “the observation and execution of semantically congruent actions” would improve verb-finding difficulties in a group of anomic patients. It is well known that in aphasic patients word-finding difficulties are the most pervasive symptom of language breakdown and that naming disorders lead to a wide variety of errors because of damage to different stages of name processing. Generally, anomic difficulties are due to inability to retrieve either the semantic word representation or the phonological word form (Basso et al., 2001, Howard et al., 1985, Levelt and Meyer, 2000 and Marangolo and Basso, 2006). Semantic impairments lead to difficulties in both word comprehension and production, whereas lexical phonological disturbances result in spoken word retrieval impairments with preserved word comprehension (Lambon Ralph et al., 2002 and Wilshire and Coslett, 2000). To further evaluate the proposal of Rose et al. (2002) that gestural facilitation effects are greater for individuals with phonologic than semantic word retrieval failures, we contrasted the effect of treatments found in two semantically word retrieval impaired participants with the results obtained in four participants with lexical phonological disturbances. To measure long-lasting beneficial effects, three follow-up sessions were carried out one week, one month and two months after the end of each treatment condition.

Given the small number of fluent patients (N = 2) and the fact that for both of them we had only two time points (baseline and after 2 weeks), we used a non-parametric approach to evaluate their increase in response's accuracy for the three treatments. In particular, we conducted a series of McNemar's tests (i.e., a non-parametric test used to compare paired proportions; Seagle & Castellana, 1988) on the proportion of correct responses for each participant by treatment and time of assessment. As shown in Table 2, neither of the two patients benefited from the treatments. In all experimental conditions, there was no increase in response's accuracy after two weeks from the end of the treatments. Table 2. Proportion of correct responses for each fluent participant by treatment and time of assessment and McNemar's test results. Control Treatment 1 Treatment 2 Treatment 3 Baseline 2 Weeks McNemar's p-value Baseline 2 Weeks McNemar's p-value Baseline 2 Weeks McNemar's p-value Baseline 2 Weeks McNemar's p-value P.A. .07 .03 1 .00 .07 .500 .03 .07 1 .07 .03 1 V.F. .00 .03 1 .13 .06 .687 .00 .06 .500 .10 .06 1 Notes: Number of observations (i.e., number of verbs) per cell: N = 29 for P.A. and N = 31 for V.F. In italics statistically significant p-values at the .01 level, indicating a significant increase in the proportion of correct responses. Table options In the nonfluent group, statistical analyses were performed in three steps.1 First, for each patient we conducted descriptive analyses (see Table 3 and Fig. 1) on response accuracy by type of treatment and time. As shown in Fig. 1, all patients showed an improvement in response accuracy for treatment 1 (based on “action observation” and treatment 2 (based on “action observation and execution”), which still persisted two months after the end of the two treatments. Table 3. Proportion of correct responses for each nonfluent participant by treatment and time of assessment. U.P. M.B. R.M. M.P. Contr. Treat1 Treat2 Treat3 Contr. Treat1 Treat2 Treat3 Contr. Treat1 Treat2 Treat3 Contr. Treat1 Treat2 Treat3 Baseline .10 .05 .14 .14 .12 .12 .12 .18 .13 .00 .00 .09 .27 .00 .18 .18 2 Weeks .33 .62 .67 .38 .24 .53 .47 .24 .30 .61 .65 .35 .45 .82 .73 .64 F/U 1 Week .29 .57 .81 .33 .35 .47 .47 .35 .35 .57 .74 .35 .55 .91 .73 .64 F/U 1 Month .19 .43 .52 .33 .24 .71 .65 .35 .30 .48 .52 .30 .45 .82 .73 .64 F/U 2 Months .29 .43 .57 .33 .18 .76 .71 .24 .35 .52 .61 .35 .45 .82 .73 .64 Note: Total number of observations: N = 420 (21 per cell) for U.P., N = 340 (17 per cell) for M.B., N = 460 (23 per cell) for R.M., N = 220 (11 per cell) for M.P. Legend: Contr. = control list, Treat = treatment, F/U = follow-up Table options Full-size image (78 K) Fig. 1. Percentage of correct responses for each nonfluent subject as a function of treatment (treat 1 = action observation, treat 2 = action observation and execution, treat 3 = action observation and meaningless movement) and time (F/U (follow-up) 1 wk (1 week), F/U 1 mo (1 month), F/U 2 mos (2 months). Figure options Second, a generalized mixed model approach (Baayen, 2008, Jaeger, 2008 and Pinheiro and Bates, 2000) was used to evaluate the effect of treatment on participants’ responses. As each patient was administered different items and as each treatment included different items, we conducted a series of logistic mixed models for each participant using the item as random effect. From a theoretical perspective, the rationale for conducting separate models is that each patient represents a single case study; from a statistical perspective, treating the patients in this study as random effects in a global model could have led to distortions in the results because of the small sample size (n = 4). Thus, to test our hypotheses we estimated four nested logistic mixed models. In Model 1 (M1), the dependent variable was accuracy of response (0 = incorrect, 1 = correct); the fixed effects were accuracy of response at baseline, time point (2 weeks after baseline, 1 week follow-up, 1 month follow-up, 2 months follow-up), treatment (0 = control, 1 = treatment 1, 2 = treatment 2, 3 = treatment 3 (action observation and meaningless movement)), type of verb (transitive vs. intransitive), and two- and three-way interactions among time, treatment and type of verb; the random effect was the item. Model 2 (M2) was identical to M1, except for the three-way interaction, which was removed. Model 3 (M3) included only the main effects of time, treatment, type of verb, and the covariate (i.e., accuracy of response at baseline). Model 4 (M4) included only the main effects of treatment and the covariate. The covariate was always considered in order to remove the effects of participants’ baseline level of performance. The best-fitting model was selected using the BIC criteria ( Raftery, 1995 and Wagenmakers, 2006), i.e., the model with the smallest BIC is considered the most appropriate model for reproducing the observed data. As can be seen in Table 4, M4 was the best-fitting model for all participants, indicating that time and type of verb had no effect on participants’ performance after controlling for baseline level. Table 5 presents the four final models. Table 4. BIC values for models tested. Models U.P. M.B. R.M. M.P. M1 524.90 469.00 534.00 338.10 M2 499.90 426.20 496.10 314.10 M3 425.80 360.20 413.10 242.30 M4 414.70 342.90 396.40 224.10 Note: Numbers in italics the minimum value of BIC, indicating the best-fitting model, for each subject. Table options Table 5. Summary statistics for the best-fitting model by subject. Predictors U.P. M.B. R.M. M.P. χ2 (3) B Z χ2 (3) B Z χ2 (3) B Z χ2 (3) B Z Accuracy at baseline .71 .84 .53 .70 2.48 1.38 .89 1.07 Treatment 14.24** 15.47** 13.77** 8.97* Treatment 1 1.78 2.38* 2.45 3.28** 2.91 2.43* 2.66 2.94** Treatment 2 2.60 3.44*** 2.14 2.89** 3.91 3.23** 1.60 1.96* Treatment 3 .42 .56 .35 .45 .20 .87 1.99 1.22 Note: Control list was used as baseline category. * p < . 05. ** p < .01. *** p < .001. Table options Overall, treatment had a significant effect on accuracy of response across all participants. In particular, treatments 1 and 2 were significantly and positively associated with an improved performance, whereas treatment 3 had no significant effect. For each participant, planned comparisons were performed to assess whether treatment 1 and 2 had differential effects on participants’ performance. Results indicated that the two treatments did not differ from each other in terms of their effect on accuracy of response (all ps > .20). Third, we adopted a meta-analytic approach (Borenstein, Hedges, Higgins, & Rothstein, 2009) to obtain a global measure of the effect size of each type of treatment (i.e., a summary effect). Given the small number of participants, we performed a fixed-effect meta-analysis, as suggested by Borenstein et al. (2009). Analyses were conducted only on the 2-month follow-up data in order to obtain the most conservative evaluation of treatment efficacy. Using the procedure suggested by Borenstein et al. (2009), we calculated the summary odds ratio, its level of significance and the associated confidence interval for each treatment (see Table 6 for summary of results and Appendix). As can be seen in Table 6, only treatments 1 and 2 had a positive significant effect on performance, confirming the results obtained in step 2. Table 6. Results of the meta-analysis by treatment. Treatments Summary odds ratio p-Value 95% Confidence interval Lower limit Upper limit 1 3.37 <.001 1.69 6.71 2 4.02 <.001 2.00 8.07 3 1.30 .470 .64 2.65 Table options In summary, results clearly show that, for the nonfluent aphasics verb production improved to the same degree by “observing” and by “observing plus executing the action” and for both rehabilitation procedures this improvement was long-lasting and still present also at two months after the end of the treatment. Both the two fluent patients did not benefit from the treatments.