Generalisation: exploring change across language levels

In aphasia intervention, it is equally critical to understand both direct therapy effects and those effects that extend beyond the immediate focus of treatment, that is, that generalise to untrained behaviours, to different contexts and/or to the broader domains of communication. This understanding underpins the development of our therapies, the evaluation of outcomes and ensuring therapy has the maximum benefit. Over recent years, there has been a burgeoning of interest in generalisation that has extended beyond looking for change in untreated items/structures to considering change in everyday communication. Alongside this have been attempts to understand the mechanisms for generalisation and why or why not generalisation takes place. We are still, however, some way from having a theory-driven and evidenced-based platform to systematically inform our planning for and evaluation of generalisation. In this Special Issue of Aphasiology, we focus on linguistic change and generalisation across the linguistic levels of words, sentences and discourse, and aim to progress a step closer to building theories around generalisation. We have focused primarily on poststroke aphasia but have included one study, by Croot et al. (2015), involving participants with primary progressive aphasia (PPA). Further, the majority of studies involve intervention in spoken production, reflective of the relative focus within the literature, although Kiran, Des Roches, Villard, and Tripodis (2015) consider the gains seen in discourse comprehension following written sentence comprehension treatment. The issues discussed here are, however, of relevance to the study of linguistic-based treatments across language modalities and across a broad range of client groups. This Special Issue presents studies where generalisation is seen and where it is not, recognising that negative evidence is just as important for understanding therapy outcomes and the mechanisms of therapy.

We begin this volume with Webster, Whitworth, and Morris’ (2015) review paper in which definitions of generalisation are considered and a framework offered for describing generalisation both within and across the levels of language. The importance of understanding generalisation in this way is demonstrated through a review of studies of spoken production considering therapies targeting single word, sentence, and discourse production. Key issues are highlighted from the literature, namely using theory and evidence to predict change across levels and how to measure change in a reliable and sensitive way. Webster et al. propose that only by approaching measurement systematically will we grow our confidence in selecting the right measures and the right contexts to best seek evidence of generalisation at the predicted level of language change. The wider impact of our interventions, such as those related to increased participation or improved self-esteem, are viewed here outside the realm of generalisation although are considered no less important to evaluating aphasia interventions. The framework proposed within this paper distinguishes between therapies that target a single level (as in the papers of Croot et al. and Kiran et al.), and those that target multiple linguistic levels (as in Edmonds, Obermeyer, & Kernan, 2015; Whitworth et al., 2015; and Carragher, Sage, & Conroy, 2015). The literature is then reviewed in terms of whether the focus of therapy is on single words (as in Croot et al.), sentences (Kiran et al.), single words and sentences (Edmonds et al.) or single words, sentences and discourse (Whitworth et al. and Carragher et al.). All of the studies seek to measure and understand change across levels.

Our examination of current evidence begins with Croot et al. who consider the effects of naming therapy on word retrieval in connected speech. Unlike the other papers in this volume who present findings from people with aphasia, the two participants had a diagnosis of PPA. The intervention protocol involved repetition and/or reading aloud in the presence of pictures of words and phrases about personally relevant topics, and resulted in item-specific gains in treated phrases but no change in overall production of either treated or untreated items in structured interviews about topics. Croot and colleagues suggest that this lack of generalisation across levels was consistent with the hypothesis that generalisation in connected speech would only be seen if generalisation was seen to untreated words within a naming context. One participant, however, showed a trend for change in some aspects of connected speech that may be indicative of improved lexical retrieval, that is, increased words that included both content and function words. This study highlights the challenges involved in choosing functionally useful words and measuring changes in lexical retrieval in everyday communication.

Also targeting a single language level, Kiran et al.’s study, involving 40 participants, explored generalisation effects arising from a syntactic intervention for written sentence comprehension on a novel assessment of discourse comprehension predicted to capture generalisation. The discourse contexts varied in their syntactic complexity and were chosen for their overlap with sentence types targeted in therapy. Therapy resulted in improved comprehension of treated structures but, in contrast to their predictions, there was no improvement in the accuracy of discourse comprehension. Kiran and colleagues conclude that the lack of generalisation was due to the gains in sentence comprehension being insufficient to facilitate improvement on the written comprehension task, primarily due to task differences and the cognitive demands of discourse comprehension, for example, working memory; this highlights the work to be carried out in both understanding these constructs theoretically and then developing suitable assessments to measure change. These initial studies also show the importance of reporting negative evidence of generalisation to assist in building our theories of language processing and recovery.

In contrast to the above papers, Edmonds et al. report extensive generalisation following Verb Network Strengthening Treatment therapy, a protocol aimed at improving lexical retrieval around verbs by strengthening the relationship between verbs and their varied thematic roles, thereby activating semantic networks shared between verbs. Therapy resulted in overall gains in untrained lexical retrieval of nouns and verbs, untrained sentence production, and changes in discourse. This study focuses on exploring individual variation in patterns of generalisation based on an analysis of pretreatment sentence production abilities/difficulties with the aim of exploring the mechanism of therapy and understanding generalisation processes. Edmonds et al. conclude that improved lexical retrieval underpins gains in all participants, but the extent of generalisation seen within sentence and discourse production is dependent on the presence of co-occurring sentence level difficulties. This theme reinforces discussion by Webster et al. on the need to understand the comorbidity in terms of both the presence and nature of co-occurring impairments.

The final two studies focus intervention at multiple language levels but seek to measure the impact on these different levels in different ways. Whitworth and colleagues compare the outcome of a multilevel treatment approach (NARNIA: a Novel Approach to Real-life communication: Narrative Intervention in Aphasia) combining work on lexical retrieval, argument structure, and discourse macrostructure, with usual speech-language pathology that, importantly, did not systematically involve multiple components nor focus on macrostructure. Macrostructure was targeted across a range of discourse genres, scaffolding the word and sentence components of the therapy. While there were few significant differences between the NARNIA and usual care groups following intervention, there were significant within group differences pretherapy and posttherapy. The group who received the multilevel therapy demonstrated improvements in verb usage, sentence production, and discourse organisation in the context of everyday discourse. In contrast, minimal improvements were evident for the group where multiple levels were not targeted, with the authors suggesting that, for change at the discourse level to be seen, discourse needs to be targeted in therapy.

Carragher et al. also present a multilevel intervention that focuses on word, sentence, and connected speech in the context of targeting social communication; generalisation effects were measured in the information exchange of novel untrained storytelling tasks. The multilevel treatment approach shares some similarities with Whitworth et al., with a focus on the production of appropriate words and argument structure, and drawing on principles of story grammar and thinking for speaking to help the person with aphasia plan and develop strategies to convey information in a clear way. Additionally, communication partners received conversation coaching to enable them to develop strategies to check and clarify information, with the therapy attending to both the transactional nature and coconstruction of information exchange. Therapy resulted in generalisation to untrained novel stories for three of the four participants, showing improvement in the communication of main ideas and changes in the structure of simple narratives. Interestingly, the authors report no straightforward relationship between changes in the ability of the participants with aphasia to communicate their ideas and the extent to which they were understood by their conversation partners.

While some similarities are present in the final two studies, with Whitworth et al. and Carragher et al. focusing on some of the same language targets, their evaluation of effectiveness and approach to measuring generalisation was different. What is measured is intimately bound up with the ultimate goals of intervention, attention to interaction between and across levels, and whether the change is predicted as a direct outcome of therapy. The relationship between generalisation and impact is also paramount here, a concept introduced in Webster et al.’s review. As our evidence and understanding grows, the direct links between language change and communication change will surely bring generalisation and impact closer together. Carragher et al.’s paper also raises another important issue in this area in that, when multiple therapeutic approaches are used (e.g., multimodal approaches that draw on a variety of verbal and nonverbal strategies, conversation coaching), we can be at risk of diluting our understanding as to what aspects of intervention are influential in the change process. We may use multicomponent and multilevel therapies to maximise the prospect of targeted and generalised change but we need to ensure we do not package everything up into one therapy protocol without understanding what contributes to the improvement and whether all aspects are important and necessary.

Some of the themes introduced within the initial review paper are reinforced by the findings and discussion from the other papers. These issues will need to be addressed if we are to continue to develop our understanding of generalisation and demonstrate the effectiveness of our intervention. Webster et al. highlight that assessment of treatment outcome (both direct therapy effects and generalisation) should be motivated by theory and predictions based on previous evidence. In some cases, generalisation is not predicted, for example, Croot et al. suggested that change in conversation would not be seen if there were item-specific effects at word level (although a trend for more generalised lexical improvement in one client does raise some issues for further exploration). In other cases, generalisation is predicted and not seen (as in Kiran et al.), and thus, there is a need to refine the theories and/or consider the sensitivity of our assessment measures.

All of the studies highlight the challenges associated with measuring change in less constrained and more everyday contexts. Croot et al. draw attention to the challenges involved in selecting discourse contexts that will elicit the targeted vocabulary and the fact that the communicative function of the discourse may be lost on multiple repetition. The variability of conversation is also highlighted (Croot et al. and Carragher et al.), with variations due to topic, interest, and motivation influencing the linguistic content of the conversation. This may lead to a “middle ground between experimentally controlled tasks for eliciting monologic aphasic data and more ecological but unconstrained sampling of conversation data” (Carragher et al. p. XX), with the use of structured interviews (as in Croot et al.), information exchanges (as in Carragher et al.) and monologic discourse (as in Whitworth et al.) being employed when an intervention warrants these. Of key importance, however, in measuring linguistic change is an understanding of the relationship between levels and the tasks used to assess them. Osgood (as cited in Croot et al.) states the degree of generalisation is predicted by the degree of similarity of task and stimuli. With across-level generalisation, there are changes of task and additional linguistic and cognitive demands that should be considered. These additional demands may reduce the extent to which generalisation is seen. More importantly perhaps, these additional demands may require us to explicitly introduce specific steps within therapy to aid generalisation across levels or to use multilevel interventions that incorporate discourse as one of the therapy contexts.

So, as we move forward, we need to continue to consider the theoretical relationships between language levels and the tasks we use to assess each of these; only then will we truly understand the mechanism of our therapies and how to promote change in everyday discourse. As always, demonstrating these gains in a reliable and sensitive way needs to remain paramount. Finally, we need to stay cognisant of considering linguistic treatment effects and generalisation within the broader context of the impact on the person with aphasia (and their communicative partners), how they feel about their communication, and the extent to which they can participate in everyday activities. We trust that the studies which make up this Special Issue will help to progress our understanding of this critical issue in aphasia intervention.

Disclosure statement

No potential conflict of interest was reported by the authors.