Within- and Cross-Language Generalization in Narrative Production of Bilingual Persons with Aphasia following Semantic Feature Analysis Therapy

The main goal of language therapy for persons with aphasia (hereafter referred to as PWAs) is to improve their functional communication and connected speech, while the methods used in therapy primarily aim to enhance impaired language skills [1‒3]. Lexical access and naming difficulties are among the most pronounced symptoms of aphasia [4, 5]. Additionally, there is evidence that narrative production in PWAs is also impaired. Findings from studies focusing on the narrative production of PWAs suggest that narratives produced by PWAs can be distinguished from those produced by neurotypical individuals in several measures, such as informativeness, verbal efficiency, syntactic complexity, grammaticality, verb diversity, and fluency [6‒13]. To assess the language abilities of PWAs, alongside standard language tests that evaluate separate language skills, the examination of narrative production, which reflects natural speech, has been increasingly utilized in recent years [6, 7, 9‒16]. Given that naming therapy is one of the most common and widely used approaches in aphasia treatment [4, 5], recent research in monolingual PWAs has focused significantly on how single-word retrieval therapy generalizes to narrative production [17‒19]. Studies on this topic in bilingual persons with aphasia (hereafter referred to as BiPWAs) are scarce and are mostly confined to case studies [16, 20‒22]. The current study aimed to investigate generalization after Semantic Feature Analysis (SFA) therapy [17, 19] to narrative production in a group of Russian-Hebrew-speaking BiPWAs. This is the first study involving multiple participants that includes both experimental and control groups of BiPWAs.

SFA is an organized method for facilitating semantic network activation, using the structured generalization or verification of semantic features of trained words. It is widely used in the treatment of single-word naming difficulties in aphasia [17, 23, 24]. SFA is based on the Spreading Activation Model of lexical-semantic processing [25]. The effectiveness of SFA has been demonstrated in the treatment of both monolingual and bilingual PWAs [17, 23, 26]. By automatically activating the conceptual/semantic network associated with a specific word meaning, SFA facilitates the activation of related meanings, potentially leading to within-language generalization, i.e., improvements in naming semantically related words in the treated language [23, 26‒29]. There is evidence that SFA leads to improvement in single-word naming in the untreated language (i.e., cross-language generalization) in BiPWAs. Since bilingual individuals have a shared conceptual system for both languages, lexical and semantic activation in one language can trigger activation in the other (see the Revised Hierarchical Model proposed by Kroll and Stewart [30], for a review).

Several factors may underlie the potential generalization of SFA to narrative production. First, when PWAs describe the semantic features of target items, these feature representations are reinforced, and their connections to associated lexemes are strengthened. The strongly activated representations spread activation to other related nodes in the lexical-semantic network, potentially improving the naming of semantically related items by making their shared semantic features more accessible. This mechanism suggests that SFA enhances the efficiency and effectiveness of the lexical system overall, thereby improving narrative production [31]. Second, it could be suggested that the improvement of lexical retrieval for nouns and verbs following SFA, along with the enhancement of verb argument structures and semantic roles, can positively impact the efficiency, informativeness, and grammaticality of narrative production in PWAs [21, 22, 32]. Some authors suggest that generating semantic features during SFA and learning to use descriptions when there are difficulties in word retrieval may improve narrative output because PWAs use this strategy [33]. Additionally, repeatedly working in a structured and methodical manner, as used during SFA, can automate the process of using semantic features for word retrieval, regardless of the target word. Any generalization to untrained words may result from improved procedural memory for word retrieval processes, which could potentially extend to more complex levels of narrative [34].

There is evidence that individuals with less severe aphasia symptoms tend to show stronger generalization effects following language treatment [17, 29, 31, 35‒37]. It is suggested that cross-language generalization results from improvements in language processes in both the treated and untreated languages. This effect is associated with a balance between the activation of the conceptual/semantic network for a specific word and the inhibition of the non-target language [38, 39]. In their systematic review of 14 treatment studies that included 45 BiPWAs, Faroqi-Shah et al. [40] reported cross-language generalization in approximately half of all cases. Most studies investigating the generalization effects of SFA for naming in treated and untreated languages in BiPWAs have reported both within- and cross-language generalization [26, 28, 39]. However, the most recent systematic review by Lee and Faroqi-Shah [35], which analyzed 17 studies including data on the generalization effects of SFA on naming in 39 bilingual and multilingual PWAs, found only a marginally small effect of within-language generalization and no effect of cross-language generalization.

Several studies have found improvements in naming abilities in the stronger pre-stroke language following single-word naming treatment conducted in the weaker pre-stroke language [33, 38, 41‒43]. This direction of cross-language generalization is associated with the fact that, in bilinguals, the weaker language accesses this unified system through the stronger language [30]. Activation of the conceptual system through therapy in the weaker language can spread to the stronger language, thereby improving lexical retrieval in the stronger language. However, there are also studies reporting the opposite direction of cross-language generalization – namely, improvements in naming in the weaker pre-stroke language following therapy in the stronger pre-stroke language in BiPWAs. This opposite direction of cross-language generalization is suggested to be attributed to deficits in cognitive control mechanisms [26, 44‒47].

A fraction of the available research, most of which comprises case studies and case series, has assessed the generalization of SFA to narrative production in monolingual PWAs at both sub-acute and chronic stages of aphasia [17‒19, 27, 48, 49]. To analyze this generalization effect, the authors employed various measures tapping into the micro- and macrostructure of narrative production, such as the number of tokens, the number/percentage of different types of nouns and verbs, the number/percentage of correct information units (CIUs), and the number of tokens or CIUs per minute, among others. The reported generalization effect of SFA to narrative production in monolingual PWAs is mixed. Some authors have reported improvements in all or some of the narrative measures in certain PWAs [19, 48‒50], while others have shown no improvement in narrative production [17, 27]. It should be noted that the lack of a generalization effect of SFA to narrative production at the group level may be explained by the high variability of the results of individual participants [18]. In their systematic review of SFA studies, which overviewed data from 55 PWAs (51 monolingual and four bilingual), Efstratiadou et al. [31] concluded that improvement in narrative measures was present in approximately 40% of reported PWAs. However, another meta-analysis conducted by Oh et al. [51], which included 32 monolingual PWAs, noted that the wide variation in discourse production measures across available studies makes it challenging to determine SFA generalization to narrative production.

Several studies, primarily case studies, have investigated the influence of different single-word naming treatment methods, including SFA, on narrative production in BiPWAs [16, 20‒22, 32, 33]. Similar to monolingual studies, the methodology, assessment measures, individual participant factors, and therapy outcomes vary widely across these studies. Some authors reported a positive influence of the treatment on several narrative measures in the treated language, regardless of whether it was L1 or not, and attributed this effect to a general improvement in language skills in the treated language [20‒22, 32, 33]. Others did not find any within-language generalization effect on narrative production following treatment provided in L1 [21] or in non-L1 [16]. This lack of generalization was suggested to be linked to the difficulty of the narrative elicitation task for the participant [16] or to limited improvement in the language that was not the language of the environment at the time of the provided treatment [21].

Most studies found improvements in several narrative measures in the untreated language following treatment in either L1 or non-L1 and attributed this improvement to cross-language facilitation [16, 20‒22, 33]. However, some studies reported decreases in narrative measures in the untreated languages, attributing these declines to language inhibition due to impairments in language control [21, 22, 33]. Table 1 summarizes four case studies that investigated the generalization effect of SFA on narrative production in BiPWAs [21, 22, 32, 33]. It should be noted that generalization effects on narrative production following single-word naming treatment are not always accompanied by improvements in naming, which emphasizes that the generalization effects of language treatment in BiPWAs may manifest differently across tasks and languages [16].

The aim of the present study was to investigate the generalization effects of SFA on narrative production and their relationship to SFA gains within a group of L1-Russian-L2-Hebrew-speaking BiPWAs. Since SFA targets the conceptual/semantic level of word representation and extends to semantic networks, this study focused exclusively on examining the generalization of SFA at the lexical level of narrative production. Given the sparse and inconsistent data in this area, we expected to find partial generalization of SFA on several narrative measures at the group level, along with high inter-subject variability. Specifically, we addressed the following questions.

• 1.

  Does SFA lead to within- and cross-language generalization to narrative production in BiPWAs compared to a control group?

• 2.

  Are the generalization effects of SFA on narrative production related to naming gains?

A total of 20 BiPWAs took part in this study. The sample in the current study overlaps with the samples in Bihovsky et al. [26, 52]. A study by Bihovsky et al [26] investigated the effects of SFA on naming ability and general language skills in the treated and untreated languages, as well as the factors predicting the success of SFA. Another study by Bihovsky et al. [52] addressed the frequency, patterns, and predictors of language mixing [LM] in BiPWAs. The current study focuses on generalization to narrative production following SFA. All the participants were L1-Russian and L2-Hebrew speakers, predominantly right-handed prior to their brain damage. The group consisted of 11 females and nine males, all in the chronic stage of aphasia. To obtain information about pre- and post-stroke language proficiency, the assessment included a language history and background questionnaire (based on Anderson et al. [53] and Li et al. [54]). All participants had immigrated to Israel from the former Soviet Union and identified Russian as their dominant language pre-stroke (t(38) = 5.98, p < 0.001). They also reported high levels of pre-stroke proficiency in L2-Hebrew in terms of both comprehension and production (see Table 2). The inclusion criterion for participation in the study evaluating SFA efficacy was an accuracy level lower than 60% in either of the two languages on at least one of the following BAT subtests: Pointing, Synonyms, Antonyms, Naming, and Semantic Opposites [55, 56]. They all retained the ability to understand simple phrases and questions, and could engage in basic conversation through simple sentences and phrases. Notably, none of the participants received additional language therapy while being involved in this study. Among the participants, 18 had aphasia resulting from a stroke, while the other two (E201 and E212) developed aphasia after having a meningioma removed. The type and severity of aphasia in both L1-Russian and L2-Hebrew were determined based on the Bilingual Aphasia Test (BAT) performance and clinical judgment (Table 3).

The experimental group in the study was composed of ten participants who received SFA. The control group consisted of ten participants who opted to delay treatment due to the COVID-19 pandemic restrictions, but consented to participate in the assessment phases. The results of their language assessments, carried out 2 months apart, served as a baseline for comparison to determine the effect of SFA on narrative production of the experimental group. No significant differences between the experimental and control groups were observed in terms of age (t(18) = 0.12, p = 0.90), age of onset of L2-Hebrew acquisition (t(18) = 0.07, p = 0.94), exposure to L2-Hebrew (t(18) = 0.10, p = 0.92), education (t(18) = 1.68, p = 0.11), and post-onset time (t(18) = 0.12, p = 0.84), or the manifestations of aphasia (L1-Russian: t(18) = 0.33, p = 0.74; L2-Hebrew: t(18) = 0.30, p = 0.77). Both self-reported data from the questionnaire and data from the BAT demonstrated that Post-stroke proficiency in L2-Hebrew was lower than that of L1-Russian in the experimental and the control groups (see Tables 2, 3).

The protocol for the current study was developed based on the work of Kiran and Roberts [23] and Wambaugh et al. [19] (see [26] for more details). It included seven steps for the experimental group (see Fig. 1a) and two steps for the control group (see Fig. 1b). We will not re-describe the entire procedure in detail but will focus solely on the aspects directly relevant to this study. The full battery of assessment and treatment was conducted by the first author, a certified Russian-Hebrew-speaking SLP, who has clinical experience with SFA.

All participants in this study (n = 20) completed a baseline pre-treatment assessment. Participants in the experimental group underwent post-treatment assessments after each SFA treatment block (4 participants received one block of SFA, and 10 participants received two blocks of SFA). Participants in the control group were assessed a second time 2 months after the initial assessment.

The assessment battery included an evaluation of:

• 1.

  Language skills, using Part B of the short version of the Bilingual Aphasia Test (BAT) [56] in L1-Russian [55] and in L2-Hebrew [56];

• 2.

  Narrative production in L1-Russian and in L2-Hebrew, including the description of series of pictures and narratives about personal events (for similar procedure see [14, 16, 20]);

• 3.

  Cognitive functions, using a battery of ten non-verbal and verbal cognitive tests. In this article, we do not include cognitive assessment data from the participants as it is not pertinent to the focus of this study. These data have been reported in previously published studies [26, 52].

To avoid variability in the production of narratives caused by lexical-semantic effects and personal preferences of the participants, 17 participants (E204 – C220) in the present study were asked to produce six narratives at each assessment stage – three narratives based on a series of pictures and three narratives about personal events – in both L1-Russian and L2-Hebrew. Three additional participants (E201 – E203), who participated in a pilot study, produced one personal story and described one series of pictures in each language at each assessment stage. To elicit a narrative based on a series of wordless pictures, we used a series of freely available comic images by Herluf Bidstrup that have a subtext or hidden meaning. These picture series have been previously used with monolingual Russian-speaking PWAs [57]. In the present study, we used sequences consisting of 11–13 images, representing neutral situations, without any religious or other sensitive contexts. In different assessments, various series of pictures were used to elicit a narrative in L1-Russian and L2-Hebrew. Those series that were used for narrative assessment in L1-Russian after the first SFA block, were also used for narrative assessment in L2-Hebrew after the second SFA block, and vice versa (Table 4). It should be noted that narrative analysis was carried out immediately after therapy, without a follow-up assessment.

In this study, we analyzed 488 narratives produced by the participants (248 by the experimental group and 240 by the control group) across all assessment stages. All narratives were audio-recorded and subsequently transcribed. We randomly selected 24 narratives (12 in each language), representing 5% of the total narrative data, to be re-transcribed by a highly proficient Russian-Hebrew bilingual coder. The inter-rater reliability for transcriptions was 98.6% (SD = 2.3%; Range: 90–100%).

We followed the guidelines for coding narrative measures outlined in previous studies [58], and we formulated additional coding guidelines to account for cross-linguistic differences between Russian and Hebrew (see [52], online suppl. Appendix E; for all online suppl. material, see https://doi.org/10.1159/000542477). In narrative analysis, we used only the measures that could potentially improve as a result of SFA. For each narrative, we calculated the total number of verbal units produced in both languages (TVUs). We distinguished between verbal units produced in the target language and those in the non-target language. Subsequently, we counted CIUs, defined as words in the target language that were “accurate, relevant, and informative relative to the eliciting stimuli” (Nicholas and Brookshire, 1993, p. 340 [58]). We applied lexical and discourse measures used in previous research for narrative analysis [1, 18, 20, 33, 58]. Lexical measures were determined by calculating the ratio of noun and verb types produced in the target language to the total verbal units produced in that same language. Discourse level measures included speech rate (TVUs per minute), informativeness (CIUs per minute), and verbal efficiency (CIUs/TVUs).

The semantic feature verification variant of SFA was applied, as suggested by Edmonds and Kiran [28] and Kiran and Roberts [23] for noun treatment, and the approach by Wambaugh et al. [19] for action treatment. Six stimulus sets of 12 words (eight nouns and four verbs) for treated and untreated languages were individually constructed for each participant (based on [23]). Set 1, treated during the first SFA block, included words that the participant could not correctly name in either language during the three baseline sessions conducted prior to treatment and that were familiar to the participant before the stroke, according to self-report. During these baseline sessions, participants were asked to name 150 pictures (90 objects and 60 actions) that were sampled from a Russian database [http://neuroling.ru/en/db.htm] in both L1-Russian and L2-Hebrew [59‒61]. Set 2 included 12 semantically related words to the words from Set 1 and was used to evaluate within-language generalization during the SFA therapy. Set 3 consisted of 12 semantically unrelated words to the words from Set 1 and was used to control the general improvement of naming abilities in the treated language. For the untreated language, Sets 1, 2, and 3 were translations of their counterparts in the treated language and were used to indicate cross-language generalization during SFA therapy. It is worth noting that the stimulus sets included words from different semantic groups, which may have provided a wider generalization effect of the SFA on different types of narratives. At the same time, the selection of stimuli was determined by the individual difficulties of the participants in retrieving specific words, rather than by the lexical and semantic content of the narratives included in the assessments.

Each participant received three treatment sessions per week, with session durations varying from 27 to 60 min (M = 45, SD = 8.8). During each treatment session, the semantic features of 12 words, including nouns and verbs, were targeted as proposed by Edmonds and Kiran [28] and Kiran and Roberts [23], and Wambaugh et al. [19]. At the beginning of every third session, participants were asked to name pictures from the sets constructed for them. Each treatment block was stopped when 80% of the pictures from the experimental set (Set 1) were named correctly in two of three consecutive treatment sessions (for a similar procedure see [19, 62]). Otherwise, the treatment block stopped after 12 sessions.

Six participants (E201, E203, E204, E206, E208, and E210) experienced word retrieval difficulties in both languages during baseline sessions and therefore received two SFA blocks – one in L1-Russian and another in L2-Hebrew. The order of the blocks was counterbalanced between participants. For these participants, a translation of Set 2 was treated in the second SFA block (see [23] for a similar procedure). Four participants (E202, E207, E209, and E212) received therapy only in L2-Hebrew since they were able to name objects and actions in L1-Russian but faced difficulties in L2-Hebrew during baseline sessions. Each participant received six to 12 sessions (M = 10, SD = 2.4), in each treatment block. All treatment sessions were audio-recorded. To monitor the fidelity of the treatment, the recordings of 15 out of 152 sessions (10%) were checked by a second researcher, a clinical linguist with extensive experience in assessment and therapy plan design for bilingual individuals. Treatment fidelity was 93%, calculated as the percentage of steps in the SFA protocol that were administered correctly (for a similar procedure see [16]). To evaluate the maintenance of improvements from the SFA, all participants from the experimental group were asked to name the pictures from their individual sets 2 months after completing the treatment (follow-up).

To determine the generalization effect of SFA on narrative production, we evaluated changes in ratio scores for five measures (speech rate, informativeness, verbal efficiency, as well as noun and verb types produced) in L1-Russian and L2-Hebrew narrative production following the last SFA block for each participant in the experimental group. For each participant in the control group, we calculated the delta between the second and first assessments for each narrative measure in both languages. We then compared the results between the two groups. Positive values indicate an increase in narrative measures, while negative values indicate a decrease.

To evaluate within- and cross-language generalization effects of SFA, we calculated the differences in narrative measures in L1-Russian and L2-Hebrew for each participant following SFA treatment provided in L1-Russian and L2-Hebrew separately. We then compared the changes in narrative measures in the experimental group to those in the control group for each language separately. To perform this comparison, we applied the nonparametric Mann-Whitney test with a 0.05 (5%) significance level using SPSS23. To assess the links between SFA gains and SFA generalization to narrative production, we performed correlational analyses between the post- and pre-treatment deltas of correctly named pictures in all sets, separately for treated and untreated languages, and the post- and pre-treatment deltas of all narrative measures in both treated and untreated languages following each treatment block, using SPSS23.

Figure 2 provides a visual comparison of the changes in narrative measures in L1-Russian and L2-Hebrew between the experimental and control groups. Despite the high variability of results observed in both groups, a positive trend was noted in the experimental group compared to the control group. Most narrative measures increased in L1-Russian and L2-Hebrew in the experimental group after the completion of SFA. In contrast, no positive changes were observed in L1-Russian and L2-Hebrew in the control group between the second and first assessments. Instead, a decrease in most narrative measures was found in the control group.

When we compared the changes in narrative performance of the experimental group following SFA provided in L1-Russian and L2-Hebrew with the changes in narrative performance of the control group, we found significant positive changes in speech rate and verbal efficiency in treated L1-Russian following the SFA provided in L1-Russian. Additionally, speech rate in untreated L2-Hebrew narratives improved following the SFA in L1-Russian (see Table 5). Following SFA provided in L2-Hebrew, an increase in informativeness and in various types of nouns was observed in untreated L1-Russian, but no positive changes were found in any narrative measures in treated L2-Hebrew (see Table 5).

Our results indicated a significant decline in verbal efficiency in L2-Hebrew narrative production following block 2 of SFA (r = −0.675, p = 0.016). No links were found between the post- and pre-treatment deltas in naming for the experimental sets in treated and untreated languages and the post- and pre-treatment deltas in narrative measures in both L1-Russian and L2-Hebrew within the experimental group. Individual deltas between post- and pre-treatment narrative and BAT scores, along with treatment gains for participants in the experimental group, are shown in online supplementary Tables S1 and S2. Individual differences in narrative and BAT scores between Assessments 2 and 1 for control group participants are displayed in online supplementary Table S3.

The current study investigated the generalization effects of SFA on narrative production in Russian-Hebrew-speaking BiPWAs. We examined measures addressing both the lexical and discourse levels of narrative production in both languages of BiPWAs who received SFA. To the best of our knowledge, this is the first multiple-participant study to investigate the generalization effects of SFA on narrative production in BiPWAs. Another novel aspect of this study is the inclusion of both experimental and control groups, as well as the examination of the relationship between SFA generalization to narrative production and SFA gains.

The comparison of changes in narrative measures between the experimental and control groups indicates a positive trend in L1-Russian and L2-Hebrew for the experimental group, and a decrease in most narrative measures in both languages for the control group. These results suggest the existence of a generalization effect of SFA on narrative production in BiPWAs. The findings of the current study are consistent with previous studies that reported a positive influence of single-word naming treatment on narrative production in monolingual [19, 48‒50] and bilingual PWAs [16, 21, 22, 32, 33]. We observed high variability in the generalization effects within the current group of BiPWAs. This variability is not surprising and is typically attributed to the wide range of language profiles and varying severity of language impairment, which can differentially influence treatment effects in a group of PWAs [18, 31, 51]. In the group of participants in this study, variability in results may also be due to bilingualism factors such as age of language acquisition, language dominance, and pre-stroke language use. These factors may have different effects on the generalization of therapy (for review, see [35]).

Several factors might explain the decline in narrative production observed in the control group. First, compensating for language difficulties in aphasia often requires significant mental effort, which can lead to fatigue and, consequently, poorer performance in subsequent assessments. Second, the day-to-day variability in aphasic symptoms, influenced by factors such as stress or other underlying conditions, might contribute to decreased performance between the first and second assessments. Lastly, performance could be influenced by motivational or emotional factors, including the level of cooperation with the researcher [63].

The results of the current study indicated an increase in narrative measures such as speech rate, verbal efficiency, informativeness, and the use of various types of nouns in both treated and untreated languages following SFA. Our findings support the previous suggestion that single-word naming treatment targeting the semantic level of language processing may influence the lexical level of narrative production. It can be suggested that the strengthening of connections between semantic representations and their associated lexemes during SFA enhances lexical access to content words, leading to more efficient and faster verbal production [31]. This may explain the observed increase in speech rate following SFA. The enhancement in the effectiveness of the lexical system, along with the possible use of description as a strategy to convey meaning, may also improve the verbal efficiency and informativeness of narrative production [21, 22, 31‒33]. Finally, the improvement in lexical retrieval for nouns, which is a primary goal of SFA, can increase the variety of nouns used in narrative production [21, 22, 32].

According to the results of the current study, within-language generalization on narrative production occurred in the current group of BiPWAs following SFA provided in L1-Russian, the pre- and post-stroke dominant language for most participants. We tend to attribute this observed within-language generalization in L1-Russian to an increase in language activation following treatment [21]. In contrast, no within-language generalization effects were found following SFA in L2-Hebrew. Our results align with previous studies that reported within-language generalization in some treated languages – whether L1 or non-L1 [16, 20, 21]. However, we cannot fully apply previous explanations to the current study. Since we conducted a joint analysis of different types of narratives for our group of participants, we cannot, like Lerman et al. [16], attribute limited generalization to difficulties associated with a specific type of narrative task. Similarly, we cannot, like Goral et al. [21], attribute the lack of generalization to the fact that L2-Hebrew was not the language of the environment during treatment as all participants in our study used L2-Hebrew in their daily lives prior to the stroke. We suggest that the limited within-language generalization observed in L1 in the current study may support the view that treatment generalization manifests differently across languages [16]. Cross-language generalization on several narrative measures was observed following SFA provided in both L1-Russian and L2-Hebrew. Our findings are in line with previous studies that reported improvements in several narrative measures in the untreated language following treatment in either L1 or non-L1, attributing these improvements to cross-language facilitation [16, 20‒22, 33].

Although SFA positively affected narrative production in the current group of BiPWAs, as reported in this study, treatment gains in single-word naming were observed in our previous study [26], no associations between treatment gains in naming and improvements in narrative measures were found. While this finding is surprising – given that improvements in the lexical measures of narrative production are typically associated with enhanced lexical access [31] – it is consistent with the results of several previous studies. For example, Lerman et al. [16], who applied Verb Network Strengthening Treatment in the L2 of a BiPWA case and evaluated its effects on both single-word naming and narrative production, found an improvement in naming in the treated L2 but no generalization to narrative production. Conversely, in the untreated L1, significant improvements were observed in narrative production without corresponding treatment gains in naming [16]. At the same time, we are not inclined to attribute the lack of association between naming gains and improvements in narrative production in the current group of BiPWAs solely to the different manifestations of treatment effects across tasks, as suggested by Lerman et al. [16]. Instead, we propose that the relatively small sample size of the group and the high variability in the results may have reduced the sensitivity of the statistical analysis to detect this association.

The results of the current study indicated a significant decline in verbal efficiency in L2-Hebrew narrative production following block 2 of SFA. Interestingly, we previously reported a decline in general language skills in the untreated language (both L1 and L2) after block 2 of SFA in the same group of BiPWAs (for review, see [26]). Taken together, these findings suggest that SFA treatment provided in either L1 or L2 of BiPWAs can lead to pronounced cross-language generalization across various language skills, including connected speech, due to cross-language facilitation (for review, see [21, 46]). However, shifting treatment to the alternate language may lead to inhibition of the untrained language and, consequently, result in a negative effect of SFA on the untreated language.

While the study makes considerable advancements in our understanding on within- and cross-language generalization in narrative production, it also has some limitations which should be acknowledged. First, the size of the experimental group is relatively small, and the treatment procedures varied between participants (some received one SFA block in L2, while others received two SFA blocks in L1 and L2, with the order of blocks differing across participants). This variability in treatment procedures may have increased the variability of the observed generalization effects. Second, we did not consider the influence of background, language-related, and cognitive factors on the generalization effects. Third, the control group did not receive an active intervention protocol, which may have influenced the results. It is important to acknowledge that simply interacting with a clinician in the control group could have had a motivational effect.

To determine whether the observed generalization effects reported in this study are specifically caused by SFA, future studies should employ a more controlled design by comparing SFA with another active intervention, such as cognitive training or another non-language-related intervention. This would help isolate the impact of SFA and provide a clearer understanding of its effectiveness. Future research should aim to investigate the effects of SFA on narrative production in larger groups of BiPWAs with varying degrees of aphasia severity and different types of aphasia. Additionally, it would be valuable to assess whether naming therapy generalizes to narrative production in other language pairs and whether this depends on the linguistic distance between the languages. Finally, further research is needed to explore the contribution of background, language-related, and cognitive factors to the generalization effects of SFA in BiPWAs.

Based on the results of the present study, it can be concluded that semantic-based single-word treatment methods in BiPWAs have the potential to generalize to narrative production. However, the therapy gains in naming following the treatment did not consistently align with generalization patterns, highlighting the importance of assessing narrative production alongside other language skills to evaluate treatment effectiveness in PWAs. It can be suggested that the narrative measures used in this study may serve as sensitive indicators of the influence of semantic-based single-word treatment on the narrative production of PWAs. The generalization effects of treatment manifest differently across languages in BiPWAs and may be associated with a balance between cross-language facilitation and inhibition. From a clinical perspective, it may also be suggested that single-word naming treatment provided in L1 of BiPWAs has greater within- and cross-language generalization potential than treatment provided in L2.

We express our gratitude to our participants and their families for dedicating their time to participate in our study.

This study protocol was reviewed and approved by the Institutional Review Board of Bar-Ilan University on 10 July 2019. Before participating in the study, all participants signed an informed consent form.

The authors have no conflicts of interest to declare.

No specific funding was received for this research. There were no sponsors involved in the study design, execution, data analysis, manuscript preparation, or the decision to publish.

Alina Bihovsky designed the study, collected data, analyzed and interpreted the data, and drafted the manuscript. Michal Ben-Shachar collaborated on drafting and interpretation and performed revisions. Natalia Meir collaborated on the design, data analysis and interpretation, drafting, and editing of the manuscript.

Raw data and supplementary materials are uploaded to the OSF repository: https://osf.io/nkdb9/?view_only=5b381d848b0c4fb9a4d03e149739204a. Further inquiries can be directed to the corresponding author.