Concordancing, or the arranging of passages of a textual corpus in alphabetical order according to user-defined keywords, is one of the oldest and still most widely used forms of text analysis. It finds applications in areas such as lexicography, computational linguistics, translation studies and computer-assisted machine translation. Yet, the basic form of visualisation employed in the analysis of textual concordances has remained essentially the same since the keywordin-context technique was introduced, over fifty years ago. This paper presents a generalisation of this technique as an analytical abstraction of concordances represented as undirected graphs, and then characterises keywords in terms of graph eccentricity properties. We illustrate this proposal with two distinct visual renderings: a mosaic (space-filling) display and a bi-directional hierarchical display. These displays can be used in isolation or in conjunction with traditional keyword-in-context components in an overview-plusdetail pattern, or as synchronised views. We discuss scenarios of use for these arrangements in lexicographical corpus analysis, in translation studies and in text comparison tasks.

An approach to establishing requirements and developing visualization tools for scholarly work is presented which involves, iteratively: reviewing published methodology, in situ observation of scholars at work, software prototyping, analysis of scholarly output produced with the support of text visualization software, and interviews with users. This approach is embodied by the software co-designed by researchers working on the Genealogies of Knowledge project. This paper describes our co-design methodology and the resulting software, presenting case studies demonstrating its use in test analyses, and discussing methodological implications in the context of the Genealogies of Knowledge corpusbased approach to the study of medical, scientific, and political concepts.

BACKGROUND: It is important to use outcome measures for novel interventions in Alzheimer’s disease (AD) that capture the research participants’ views of effectiveness. The electronic Person-Specific Outcome Measure (ePSOM) development programme is underpinned by the need to identify and detect change in early disease manifestations and the possibilities of incorporating artificial intelligence in outcome measures. Objectives: The aim of the ePSOM programme is to better understand what outcomes matter to patients in the AD population with a focus on those at the pre-dementia stages of disease. Ultimately, we aim to develop an app with robust psychometric properties to be used as a patient reported outcome measure in AD clinical trials. Design: We designed and ran a nationwide study (Aug 2019 – Nov 2019, UK), collecting primarily free text responses in five pre-defined domains. We collected self-reported clinical details and sociodemographic data to analyse responses by key variables whilst keeping the survey short (around 15 minutes). We used clustering and Natural Language Processing techniques to identify themes which matter most to individuals when developing new treatments for AD. Results: The study was completed by 5,808 respondents, yielding over 80,000 free text answers. The analysis resulted in 184 themes of importance. An analysis focusing on key demographics to explore how priorities differed by age, gender and education revealed that there are significant differences in what groups consider important about their brain health. Discussion: The ePSOM data has generated evidence on what matters to people when developing new treatments for AD that target secondary prevention and therein maintenance of brain health. These results will form the basis for an electronic outcome measure to be used in AD clinical research and clinical practice.

Neural language models are becoming the prevailing methodology for the tasks of query answering, text classification, disambiguation, completion and translation. Commonly comprised of hundreds of millions of parameters, these neural network models offer state-of-the-art performance at the cost of interpretability; humans are no longer capable of tracing and understanding how decisions are being made. The attention mechanism, introduced initially for the task of translation, has been successfully adopted for other language-related tasks. We propose AttViz, an online toolkit for exploration of self-attention-real values associated with individual text tokens. We show how existing deep learning pipelines can produce outputs suitable for AttViz, offering novel visualizations of the attention heads and their aggregations with minimal effort, online. We show on examples of news segments how the proposed system can be used to inspect and potentially better understand what a model has learned (or emphasized).

We present a tool for visualization of transcripts of multi-party dialogues, with application to the analysis of communication in medical teamwork. The visualization is based on a "temporal mosaic" metaphor, which provides a temporal overview of dialogues and supports the tasks of transcript browsing and information access, by segmenting the dialogue and laying out the keywords of the different segments on interactive visual "tiles". The tool has been tested on a corpus of transcribed dialogues among the members of a (simulated) critical care team. An analytical evaluation is presented which demonstrates the potential uses of the tool in an educational setting and highlights areas for improvements.