Rehabilitating an attrited language in a bilingual person with aphasia

ABSTRACT Language difficulties can arise from reduced exposure to any given language (e.g. attrition) or after brain damage (e.g. aphasia). The manifestations of attrition and aphasia are often similar so differentiating between their effects on language loss and recovery is challenging. We investigated treatment effects for an English-Hebrew bilingual person with stroke-induced aphasia who had minimal contact with his Hebrew for over 14 years. We asked whether his attrited language could be rehabilitated, how effects of attrition and aphasia can be dissociated, and how such dissociation aids our understanding of the mechanisms involved in language recovery in aphasia. We administered a verb-based semantic treatment in Hebrew three times a week for six weeks (totalling 29 hours of therapy) and assessed changes in both Hebrew and English comprehension and production abilities across a variety of language tasks. Quantitative analyses demonstrated improvement in Hebrew production across language tasks, including those involving lexical retrieval processes that were trained during treatment. Improvement to English occurred in these same lexical retrieval tasks only. We interpret these results as indicating that the participant’s attrited language (Hebrew) could be rehabilitated with both specific treatment and general exposure to Hebrew contributing to improvement. Furthermore, treatment effects transferred to the untreated English. Qualitative analyses indicated that an interaction among aphasia, incomplete mastery of Hebrew pre-stroke, and attrition contributed to the participant’s language difficulties post-stroke. We conclude that partially shared underlying mechanisms of attrition and aphasia drive language processing and changes to it with treatment.

Language difficulties can arise from reduced exposure to any given language (e.g. attrition) or after brain damage (e.g. aphasia). The manifestations of attrition and aphasia are often similar so differentiating between their effects on language loss and recovery is challenging. We investigated treatment effects for an English-Hebrew bilingual person with stroke-induced aphasia who had minimal contact with his Hebrew for over 14 years. We asked whether his attrited language could be rehabilitated, how effects of attrition and aphasia can be dissociated, and how such dissociation aids our understanding of the mechanisms involved in language recovery in aphasia. We administered a verb-based semantic treatment in Hebrew three times a week for six weeks (totalling 29 hours of therapy) and assessed changes in both Hebrew and English comprehension and production abilities across a variety of language tasks. Quantitative analyses demonstrated improvement in Hebrew production across language tasks, including those involving lexical retrieval processes that were trained during treatment. Improvement to English occurred in these same lexical retrieval tasks only. We interpret these results as indicating that the participant's attrited language (Hebrew) could be rehabilitated with both specific treatment and general exposure to Hebrew contributing to improvement. Furthermore, treatment effects transferred to the untreated English. Qualitative analyses indicated that an interaction among aphasia, incomplete mastery of Hebrew pre-stroke, and attrition contributed to the participant's language difficulties post-stroke. We conclude that partially shared underlying mechanisms of attrition and aphasia drive language processing and changes to it with treatment.

Introduction
Attrition of a language due to prolonged lack of use or reduced exposure has similar symptoms to some characteristics of aphasia. When one language of a multilingual person with aphasia has attrited, it can be difficult to differentiate between the effects of attrition and of aphasia and to understand how each contributes to observed language difficulties. We investigated a bilingual participant who sustained a stroke 14 years prior to the study interference from any non-attrited language(s), or both (e.g. Köpke, 2004;Köpke & Schmid, 2004), with the level of inaccessibility varying across different aspects of language (Higby et al., 2019). Similarly, theories relating to aphasia also attribute reduced language processing to mechanisms of activation, interference and interference control (e.g. Goral & Lerman, 2020;Kiran et al., 2013;Paradis, 2007), although depending on the characteristics of the brain lesion, certain language processing skills may be irrecoverable rather than inaccessible.
The Activation Threshold Hypothesis (ATH; Paradis, 1985Paradis, , 2007, for example, posits that lexical items and/or linguistic structures used less frequently require more activation in order to pass the threshold for processing than items or structures used more frequently. Thus, not only would a less-used language require more activation for successful and timely processing, but within that language, more-frequent lexical items and/or linguistic structures would require less activation than less-frequent ones, and therefore be more resistant to attrition (e.g. Köpke, 2004;Köpke & Schmid, 2004). Furthermore, aphasia treatment has been hypothesised to lower the activation threshold of targeted language processes and, in multilingual people with aphasia, the threshold may be lowered in untreated languages as well (e.g. Goral, 2012;Kiran et al., 2013;Lerman, et al., 2022).
A complementary theory to the ATH concerns competition among limited cognitive resources (e.g. Green, 1998;Köpke, 2004). As one language falls out of use, difficulty in controlling interference from the more frequently-used language(s) increases when multilingual people attempt to communicate in the less-used language (Köpke, 2004). This increased difficulty controlling interference consumes processing resources, which are then unavailable for other levels of language processing (Green, 1998), resulting in reduced accessibility of that language. Similarly, cognitive-resource deficits appear to play a substantial role in aphasia (see, Monetta & Joanette, 2003). In particular, inhibition mechanisms including interference control are often impaired in aphasia, resulting in a slowing of language processing (Szöllősi & Marton, 2016).
Another theory pertinent to decline in lexical retrieval (a deficit common to both attrition and aphasia), is the Transmission Deficit Hypothesis (TDH; Burke & College, 1997;Burke et al., 1991) which posits that nodes connecting the semantic network to phonological information weaken, resulting in reduced priming transmission and delayed or impaired word production. Such weakening of nodes is partly related to frequency of use for specific lexical items. This hypothesis is more commonly linked to the decline in language processing in monolingual older adults, but it has also been linked to attrition in multilingual people (Goral, 2004). The TDH does not contradict the ATH, rather it specifies that the connection between semantic and phonological information is necessary for successful processing, for lexical production more than comprehension, as Goral (2004) observes. The TDH may also be compatible with explaining some characteristics of aphasia such as word retrieval difficulty and phonological paraphasias. As hypothesised for older adults' production and for certain aspects of aphasia, semantic information is relatively spared and can be accessed, but only part, if any, of the phonological information can be accessed in lexical-retrieval breakdown (e.g. Friedmann et al., 2013;Meier et al., 2016).

Recovery of an attrited language after a stroke
There are several published cases of multilingual individuals with aphasia in whom one language has potentially attrited, either before or after the stroke (e.g. Filiputti et al., 2002;Goral, 2012;Goral et al., 2013;Knoph et al., 2015;Meinzer et al., 2007). Indeed, we must not only consider changes in the post-stroke language environment (e.g. after giving up work) that may result in attrition of one or more languages, but also attrition as a direct result of the stroke. For example, a non-parallel impairment, that is, different degrees of impairment in the two or more languages of a multilingual person, could result in attrition of the more-impaired language if it is used less for daily communication after the stroke than a better-spared language, as Lerman et al. (2019) proposed.
Recovery of an attrited language after a stroke may, therefore, be the result of re-exposure to that language, spontaneous recovery in the months immediately after the stroke, and/or a direct result of aphasia treatment. The underlying mechanisms of recovery may also be similar for language loss due to attrition and to aphasia. For example, the balance between increased spreading activation through treatment and controlling interference from a non-target language has been described in detail in the bilingual aphasia literature (e.g. Conner et al., 2018;Kiran et al., 2013), and may also apply to recovering an attrited language; in both instances, activation thresholds would be lowered by re-exposure and interference from other languages would be more easily controlled.

The current study
Similarities between the characteristics and potential underlying mechanisms of attrition and aphasia motivated our research questions. First, we asked whether a language rarely used in 14 years prior to a stroke could be successfully rehabilitated after the stroke, and whether any treatment effects generalise to an untreated language. Second, we asked how the effects of attrition and aphasia can be dissociated in such a case and how this dissociation adds to our understanding of the mechanisms involved in language recovery.
We hypothesised that the language skills of our participant, EH02, 1 would improve after treatment in the attrited language. If aphasia is a key underlying cause of his pre-treatment language difficulties, recovery should be specific to the treatment provided, and potentially generalise to the untreated language. Conversely, if attrition due to lack of use is a key underlying cause of his difficulties in the rarely-used language, recovery should be more general, and only in the treated language. We also hypothesised that it should be possible to differentiate among factors underlying EH02ʹs language skills by conducting a qualitative analysis of his responses during pre-and post-treatment assessment, although we were concerned that his pre-attrition, pre-stroke moderate language proficiency might be a confounding factor. Finally, we hypothesised that the qualitative analysis together with our quantitative analysis should indicate the mechanisms involved in language recovery of an attrited language.

Methodology
We conducted a multiple-baseline case study, assessing English and Hebrew language skills at baseline, providing treatment in Hebrew, and then re-assessing English and Hebrew post-Hebrew-treatment. This study was approved by the City University of New York ethics committee (ID 2017-0702).

Participant
EH02 is a right-handed, English-Hebrew bilingual man, age 65, with 19 years of formal education. Based on our language background questionnaire (see, Lerman, 2020), EH02 reported that he was born in the United States and acquired English as his native language. English remained his dominant language across the lifespan for speaking, listening, reading, and writing. He started acquiring Hebrew in elementary school (a Jewish Day School that prioritised spoken as well as written Hebrew-language skills), reaching overall moderate proficiency by age 18. For the 10 years that EH02 was married (from his early-40s into his 50s), he used his moderately proficient Hebrew with his wife (a native Hebrew-speaker) and his stepchildren, together with English at home. For all other aspects of his life (work, language of the environment, friends, other family members), EH02 used only English. He reported adequate hearing and vision, no neurological diagnoses other than the single stroke, no learning disorders and no history of addiction.
Fourteen years before participating in our study, aged 51, EH02 had a left thalamic haemorrhage resulting in an original diagnosis of non-fluent aphasia without apraxia and with right hemiparesis. Around the time of the stroke he divorced, and rarely spoke Hebrew since, although he reported minimal ongoing Hebrew exposure through prayer, Jewish text study, and history books. EH02 rated his post-stroke English proficiency as higher than his Hebrew proficiency in all modalities (see ,  Table 1). He resides in a nursing home and reported not receiving aphasia treatment from a speech-language therapist (SLT) since being discharged from hospital 14 years prior to our testing. t baseline EH02ʹs languages were differentially impaired (Hebrew more than English), as measured by the Western Aphasia Battery Revised (WAB-R) in English (Kertesz, 2006), and Hebrew. 2 In English, the most salient impairment was word retrieval, although language difficulties were mild overall, based on the Aphasia Quotient (AQ). In Hebrew, severe language difficulties were observed, based on the AQ, with low scores observed in all four subtests, although auditory comprehension was relatively better than other language skills (see, Table 1). The non-linguistic subtests of the Cognitive Linguistic Quick Test (CLQT; Helm-Estabrooks, 2001) showed non-linguistic cognitive abilities ranging from within-normal-limits to a mild impairment, remaining stable throughout the study.

Procedure
Pre-and post-treatment assessments EH02 was assessed in both languages before and after treatment in Hebrew using our Revised English-Hebrew Aphasia Battery (REHAB) (Lerman & Goral, unpublished). 3 This battery includes subtests of production and comprehension, at word, sentence and discourse levels, and is comparable across English and Hebrew for psycholinguistic elements such as frequency, number of translation equivalents, etc. The subtests administered in this study were action naming, object naming, picture-based sentence construction, discourse (including picture descriptions, story sequences, procedural narratives, and personal narratives), noun comprehension, verb comprehension, and sentence comprehension.
The REHAB was split into three comparable parts in each language. At each time-point, assessment was conducted over several days, with one part of the REHAB administered each day in either English, Hebrew or both (in consecutive blocks). In addition, the English WAB-R, Hebrew WAB-R and CLQT were administered at each time-point. Frequent breaks were provided, and language order was counterbalanced throughout the study. Assessment was conducted in a quiet room in EH02ʹs nursing home by an experienced English-Hebrew bilingual SLT (the first author) or research assistant with experience administering the REHAB. All sessions were video-recorded, and later transcribed. Transcription accuracy was confirmed on 25% of the data by a second transcriber and high inter-rater accuracy was observed Table 1. Pre-and post-stroke subjective and objective language measures using self-ratings and the Western Aphasia Battery-Revised in English and Hebrew. Note. WAB-R = Western Aphasia Battery-Revised; AQ = Aphasia Quotient a On a scale of 1-10, with 10 being the highest. b Verbal report (> 99% across all tasks and both languages). Two raters scored the transcribed data blind to testing time. Inter-rater reliability was high (α = 0.99), as calculated using Krippendorff's Alpha coefficients (Hayes & Krippendorff, 2007) Treatment Verb Network Strengthening Treatment (VNeST) was provided in Hebrew, EH02ʹs L2.
VNeST is a therapy technique with a focus on verbs as the core of sentences. Participants practice producing and comprehending a variety of SVO sentences, using a highly structured approach. During VNeST, the semantic network is expected to be repeatedly activated and thus strengthened by the systematic retrieval of verbs together with their thematic roles . Thus, sentence production is expected to improve within both constrained tasks and discourse contexts (Edmonds et al., , 2009). The VNeST protocol has been reported in detail in previous studies (e.g. Edmonds, 2014;Edmonds et al., 2014Edmonds et al., , 2009) and we followed all its steps, with one modification. We added a written component to the original VNeST protocol, because EH02ʹs writing abilities in English were relatively spared, and his contact with Hebrew for the last 14 years had been predominantly in the written modality. See, Figure 1 for a visual representation of our modified VNeST protocol. Twenty 2-argument verbs were selected (excluding cognates) and rotated pseudo-randomly throughout the treatment block. Each completed protocol was counted as one verb cycle. Each verb was presented 3-4 times, resulting in 65 verb cycles completed overall. Treatment was provided by an experienced English-Hebrew bilingual SLT (the first author). EH02 received 29 hours of treatment (three sessions a week for six weeks). Sessions lasted for 1.25-2 hours, with 3-5 verbs targeted per session. A third of all treatment sessions were either video-recorded or observed live by a second SLT; treatment fidelity was calculated at over 95%, based on our pre-determined modified VNeST protocol. 4

Analyses
We analysed whether learning occurred during treatment to determine whether changes in language abilities could be attributed to VNeST. Throughout the treatment sessions, we counted every instance of lexical retrieval within each verb cycle and recorded whether retrieval was independent or followed a minimal or maximal cue (see, Figure 1). Retrieval abilities were averaged across verbs within each session, resulting in a score of 0-8 (eight being the maximum number of agents and/or patients that could be produced over four sentences per verb). We calculated correlations between treatment session and independent retrieval of lexical items, and between treatment session and retrieval of lexical items after being cued (minimal and maximal cues separately). We found that EH02ʹs ability to independently retrieve lexical items when presented with a verb during VNeST increased as sessions progressed, while the need for a cue to aid retrieval decreased (see, Figure 2), thus learning occurred during treatment. We therefore accepted that within-and cross-language treatment effects could potentially be attributed to VNeST.

Quantitative analysis: within-and cross-language generalisation
Treatment generalisation was measured by comparing post-treatment language scores with pre-treatment scores on the REHAB and WAB-R in each language. Analyses included the McNemar test of equal change (using the Benjamini-Hochberg approach to correct for multiple comparisons; Benjamini & Hochberg, 1995), and effect sizes as measured by Cohen's d and Non-overlap of Pairs (NAP). We accepted that improvement occurred when the McNemar score increased significantly and/or Cohen's d increased with a positive small change or greater, in conjunction with a positive medium or high NAP score. Furthermore, for type-token ratios, we considered an increase of 0.05 or more to indicate improvement. See Supplementary Table S1 for details of subtests and analyses.

Qualitative analysis
We transcribed and analysed EH02ʹs correct responses and errors during Hebrew pre-and post-treatment assessment and compared them to English. Here, we did not examine change following treatment, rather we identified the following response types as being potentially attributable to one, two or all three factors: the brain lesion and resulting aphasia, attrition, and/or moderate peak proficiency of Hebrew: (1) abandoned utterances; (2) phoneme substitutions; (3) declaring non-acquisition of words; (4) accuracy differences in comprehension vs. production across languages; (5) invented root-based lexical items; (6) invented compound words; (7) language mixing; (8) general-verb use; (9) frequency effects of naming accuracy.

Quantitative analysis: Within and cross-language treatment effects
See Table 2 for measures of production at the single-word, sentence, and discourse tasks in Hebrew (within-language treatment effects), and in English (cross-language treatment effects). For Hebrew, raw scores were all higher post-treatment than pre-treatment. For the structured tasks in Hebrew, improvement was observed from pre-to post-treatment in object naming accuracy, action naming accuracy (both for treated and untreated verbs), and relevant subject-verb-object (SVO) sentence production.
For discourse tasks in Hebrew, improvement was observed for output, as measured by relevant SVO utterances, correct information units (CIUs) and rate of CIU production (CIUs/min). Improvement to the informativeness and quality of production was also observed, with increases in noun and verb type-token ratios, and the percentage of CIUs out of total verbal units (i.e. %CIUs). No change was observed for utterance grammaticality or correct verb conjugation.
For the structured tasks in English, improvement was observed for object naming from pre-to post-treatment. For the discourse tasks in English, improvement was observed for output, as measured by relevant SVO utterances. No other changes were observed, including no change for utterance grammaticality or correct verb conjugation, similar to Hebrew.
No significant change was observed in any of the comprehension tasks in either language, when we corrected for multiple comparisons.
In Hebrew, the WAB-R AQ score increased from 50.1 to 63.0 after treatment in Hebrew (an increase of 12.9 points), with increases in spontaneous speech, repetition, and naming and word finding. In English, the WAB-R AQ score remained stable at 87.5 pre-Hebrew treatment and 87.8 post-Hebrew treatment. Type-token noun 0.61 0.72 n/a n/a n/a 0.64 0.67 n/a n/a n/a Type-token verb 0.49 0.57 n/a n/a n/a 0.67 0.65 n/a n/a n/a relSVOs 24 38 n/a Note. Pre = pre-treatment scores; Post = post-treatment scores; NAP = Non-overlap of All Pairs; relSVO = relevant SVO utterances; %relSVO = relevant SVO utterances/total utterances; CIU = Correct Information Units; %CIU = Correct information units/total verbal units; %grammatical = no. of grammatical utterances out of total no. of utterances; % conjugation = no. of correctly conjugated verbs out of total no. of verbs. Significant change is marked in bold. a Significance: p < 0.05, FDR corrected for multiple comparisons, marked with an asterisk (*); p < 0.1 marked with a ǂ b Cohen's d: Small > 1.2, Medium > 1.7, and Large > 3.3 c NAP using zero chance level: 0.32-0.84 = medium, 0.85-1.0 = strong. A negative score indicates higher scores pre-treatment than post-treatment.
In summary, after treatment in EH02ʹs more-impaired Hebrew (L2), within-language generalisation was observed in Hebrew in both the REHAB and the WAB-R. Cross-language generalisation was observed in EH02ʹs less-impaired English (L1), for object naming and relevant SVO production in discourse.
We also note, anecdotally, that the observed improvements to EH02ʹs Hebrew production were not maintained after Hebrew treatment ceased. We observed widespread decline in his Hebrew production when attempting to reassess his language skills, both immediately after the second (English) treatment block, and after 4-weeks without treatment, although his comprehension seemed stable. This decline in production was more pronounced in Hebrew than in English after 4-weeks without any treatment.

Qualitative analysis of responses
See, Table 3 for a classification of EH02ʹs errors and responses in Hebrew attributed to aphasia, attrition, and pre-stroke moderate proficiency levels, with examples.

Response types attributed solely to aphasia
Abandoned utterances are common in people with aphasia when the frustration of trying to say something becomes overwhelming and they either give up or attempt to reframe their production. In EH02, utterances were abandoned in both English and Hebrew during the sentence and discourse tasks. We therefore attributed this type of production error to the brain lesion and resultant aphasia only.

Response types attributed solely to attrition
EH02 produced phonological errors on three relatively frequent words in Hebrew, indicating that he vaguely remembered the words but could not fully retrieve them. This type of production error did not occur in English at all throughout the study, thus is likely not related to aphasia. Since this phenomenon occurred with relatively frequent words, it is also unlikely to be related to his moderate pre-stroke proficiency level.
Response types attributed solely to moderate peak proficiency EH02 often declared (in his better-spared English) that he had never acquired certain words in Hebrew, when presented with specific stimuli.

Response types attributed to either moderate peak proficiency or attrition
Several response types and errors could be attributed to either EH02ʹs moderate pre-stroke proficiency or his attrition, and we were unable to distinguish between the two putative causes because EH02 had never obtained full proficiency in Hebrew. First, the difference between his better-spared comprehension and his more impaired production was more pronounced in Hebrew than in English. Second, in Hebrew, EH02 invented nouns and verbs using the correct root with the wrong verbal pattern (which is possible in Semitic languages). Third, EH02 was observed to sometimes combine two higher frequency words into a compound instead of producing the correct, more specific, noun. These last two errors are relatively complex and not commonly observed in aphasia (see , Table 3 for examples).

Response types attributed to aphasia, attrition, and/or moderate peak proficiency
Several response types were potentially attributable to any of the three factors we were examining. First, EH02ʹs language mixing was not parallel across the two languages; he produced many words and utterances in English when assessed in Hebrew but almost none in the opposite direction (e.g. Goral et al., 2019). Second, EH02 frequently used general verbs rather than specific verbs (e.g. Hallowell, 2017;Higby et al., 2019;Lerman et al., 2019). Third, he produced more frequent lexical items more accurately than less frequent ones (e.g. Köpke, 2004;Köpke & Schmid, 2004;Paradis, 2007).

Rehabilitating an attrited language after a stroke
In answer to our first question, we found strong support that an attrited language can be rehabilitated after a stroke. After an intense 29 hours of VNeST over six weeks, significant and widespread generalisation to other stimuli and other contexts was observed for production in the treated language across single-word, sentence, and discourse level tasks. Similarly, the 12.9 point-improvement observed in the WAB-R AQ is well above the 8.26-point mark suggested as the minimal detectable change that can be credited to treatment (albeit on the English WAB-R; Menahemi-Falkov et al., 2021). We attributed part of the observed improvement in Hebrew to the specific treatment provided and consequently to rehabilitation of aphasia after a brain lesion. For example, our data showed specific improvement of lexical retrieval with an increase in noun and verb retrieval for single-words and within sentences in different contexts, as expected based on previous research with VNeST (Edmonds, 2016). Crucially, in the REHAB discourse tasks, improvement was observed for a number of measures of relevant and informative output that were expected to change due to a strengthened lexical-semantic network after VNeST (e.g. Edmonds, 2016;Edmonds et al., 2009), but no improvement was observed for grammatical measures that were not specifically targeted during VNeST. Additionally, treatment effects were observed to partially generalise from the treated Hebrew to the untreated English. This cross-language generalisation occurred specifically in two tasks repeatedly targeted during VNeST (retrieving nouns and producing relevant SVO utterances); performance on these tasks also improved significantly in Hebrew. If the Hebrew treatment had only reversed the effects of attrition, rather than targeted language skills specifically affected by the aphasia, we would not have expected any change to EH02ʹs English language skills. Indeed, we might even have expected inhibition of English during Hebrew treatment that may have lingered during the post-treatment English assessments (e.g. Goral et al., 2013;Lerman, et al., 2022). However, this was not the case. Rather, even though English was only mildly impaired following the stroke, improvement was observed, suggesting that the semantic network, assumed to be shared across languages (Paradis, 1993), was strengthened as part of the rehabilitation of the language system damaged by the stroke. Our results are consistent with a study by Goral and colleagues who reported improvement after aphasia treatment in two languages that may have attrited (Goral, 2012;Goral et al., 2013). While several other studies on aphasia treatment in multilingual people describe one language that appears to have attrited, they do not provide treatment in the attrited language (e.g. Filiputti et al., 2002;Knoph et al., 2015;Meinzer et al., 2007).

Differentiating between aphasia and attrition
In answer to our second question, we found that the effects of aphasia and attrition could only be partially distinguished in EH02. His left thalamic haemorrhage may have differentially affected his languages due to the connections between the thalamus, the basal ganglia and the frontal lobe, thus potentially disrupting the language control network and consequently impacting each language differently (e.g. Abutalebi & Green, 2007;Nadeau & Crosson, 1997;Verreyt et al., 2013). However, the extreme differences between his languages at baseline indicated that his language difficulties in Hebrew were likely not based on the effects of the brain lesion alone, rather, other factors also contributed. Our results support this hypothesis.
Attrition is expected to affect lexical retrieval before other aspects of language, such as comprehension and grammar (e.g. Bardovi-Harlig & Stringer, 2010;Keijzer & De Bot, 2019;Köpke, 2019;Obler, 1982) and thus the patterns of data that we observed are also consistent with re-exposure and are not limited to specific treatment effects. Additionally, the results from the WAB-R indicated that not only were improvements observed for spontaneous speech (particularly for information content) and naming and word finding, which might be expected based on the goals of VNeST (Edmonds, 2016), but improvement was also observed for repetition (of words, phrases, and sentences), which was not specifically expected to change. This may indicate a general improvement in Hebrew through re-exposure, although it is also possible that the improvement to EH02ʹs strengthened lexical-semantic network resulted in improved reliance on that network when repeating full or partial sentences (e.g. Nozari et al., 2010).
Importantly, EH02 was mostly isolated in the nursing home, communicating briefly with nurses, aides, or visiting family throughout the week, and always in English. Thus, after treatment in Hebrew, he returned to a reality of minimal contact with Hebrew, and it appears that any improvements to Hebrew were not maintained when Hebrew treatment ceased. These patterns of change suggest that re-exposure to Hebrew during treatment contributed to the improvements observed, and that a return to minimal exposure resulted in a rapid return to language production difficulties. Change in levels of activation is not uncommon with immersion in an attrited language (Köpke & Schmid, 2004) but may be particularly sizeable following treatment for aphasia.
Similarly, our qualitative analysis of EH02ʹs responses and error types further indicated that aphasia was not the only factor contributing to his language difficulties in Hebrew, even though it was challenging to tease apart the different factors and their effects on language (particularly to distinguish between attrition and moderate pre-stroke proficiency of Hebrew). For example, while EH02ʹs comprehension was better-spared than production, which was not unexpected based on the subcortical location of the lesion, the difference between these abilities was more pronounced in Hebrew than in English. This difference could reflect his pre-stroke high proficiency in English versus his moderate proficiency in Hebrew, or it could be the result of Hebrew attrition after 14 years of minimal use, because language decline due to attrition is more salient for production than for comprehension (e.g. Bardovi-Harlig & Stringer, 2010;Higby et al., 2019;Köpke, 2004Köpke, , 2019. Additionally, EH02ʹs uni-directional language mixing may be due to the stroke, based on the subcortical location of the lesion, and its connections with the basal ganglia. However, it is likely that the imbalance of pre-stroke proficiencies and attrition also contributed to the differential language-mixing patterns (see, Lerman, Pazuelo et al., 2018). In summary, we suggest that EH02ʹs widespread albeit temporary improvements in Hebrew were likely the result of specific strengthening of the lexical-semantic network through VNeST, potentially boosted by a general activation from re-exposure to the attrited language.

Mechanisms accounting for aphasia and attrition
We posit that by strengthening the lexical-semantic network with VNeST, increased activation spread throughout that network and the activation threshold was lowered, as described by Paradis in the ATH (e.g. Paradis, 1985Paradis, , 2007. This strengthened lexical-semantic network positively affected both languages, due to a predominantly shared semantic network across all languages of multilingual people (Paradis, 1993). Consequently, not only were improvements to word retrieval observed in EH02ʹs treated Hebrew, but some improvements were also observed in his untreated, less-impaired English. The overall re-exposure to Hebrew during the treatment block would have provided an extra level of spreading activation, further lowering the activation threshold and resulting in more widespread improvement in Hebrew, especially to more frequent words relative to less frequent words (Köpke, 2004;Köpke & Schmid, 2004), a pattern observed in his responses.
While the ATH can explain the processes involved in improved language skills due to both attrition and the brain lesion, the TDH can only partly explain our results, because improvement to the untreated language would be unlikely if connections to phonological information were strengthened, as the TDH posits (Burke & College, 1997;Burke et al., 1991). Rather we would expect improvement to be more language specific. However, the TDH could explain at least part of EH02ʹs improvement in Hebrew, because with re-exposure access to phonological information of Hebrew should improve, resulting in better production of Hebrew more generally (Goral, 2004).
Furthermore, the competition of resources theory (Green, 1998) complements the ATH and the TDH because as the activation thresholds are lowered by strengthening the semantic system, and better access to phonological information is obtained, it should require fewer resources to control interference of the non-target language. This is especially likely when the non-target language is highly proficient and less-impaired post-stroke, as is the case here. Certainly the location of the haemorrhage (thalamus) in EH02, and the connections between the thalamus to the basal ganglia and the frontal lobe, which make up part of the control network (Abutalebi & Green, 2007) support the possibility of interference control involvement and subsequent improvement, potentially contributing to the overall improvement of his Hebrew during treatment.

Limitations and future directions
One of the main limitations of our study was the presence of confounding factors other than the brain lesion and EH02ʹs minimal exposure to Hebrew for 14 years, such as the fact that he had only reached moderate peak proficiency in Hebrew. Furthermore, his motivation visibly dropped as the study progressed into the second treatment block, a factor that was likely exacerbated by his environment, in which he had limited interactions (in English only) with others in the nursing home where he resided. Thus, it was challenging to isolate the factors that we were studying. However, it is rare to find people with aphasia who have no secondary issues, and indeed in multilingual people with aphasia the picture can get quite complex. It is therefore particularly important in this population to try to tease apart the different factors involved and to understand the interaction among them in order to facilitate assessment and treatment.
A further limitation is the single-participant design, restricting generalisation of the results to other multilingual people with aphasia. However, the single-participant design allows for detailed qualitative analysis as well as quantitative analysis, which is essential to better understand the influence of attrition on post-stroke language abilities. There are few published papers that describe multilingual participants with aphasia who potentially have one or more attrited language(s) (e.g. Filiputti et al., 2002;Goral et al., 2013;Knoph et al., 2015;Meinzer et al., 2007), but the focus of those studies is not on attrition. By focusing future research on dissociating the effects of exposure and of aphasia treatment on attrited languages, together with comparing participants with different language proficiencies, aphasia types, and attrition patterns, we should better understand the underlying mechanisms of language processing within the framework of both aphasia and attrition.

Conclusion
In our study of a bilingual person with aphasia whose L2 had attrited in the 14 years since his stroke, we found that intense aphasia treatment resulted in improvements to both the treated Hebrew (L2) and untreated English (L1), as measured by within-and cross-language generalisation. Change in Hebrew was likely boosted by general re-exposure of the treated attrited language. Thus, even after minimal exposure to one language for many years, with poor post-stroke language skills relative to other languages, aphasia treatment may be effective without requiring a period of prior re-exposure, although without continued contact with the attrited language, treatment-induced change appears to be temporary. We conclude that EH02ʹs language difficulties were due to an interaction of aphasia, prestroke incomplete mastery of Hebrew, and language attrition. We suggest that attrition is not an exclusionary factor in providing aphasia treatment and that if a specific language is important to a patient, but has not been used for many years, there is merit in treating that language.