Dr Michel Wermelinger

In the last decade, the role of software engineering has changed rapidly and radically. Globalisation and mobility of people and services, pervasive computing, and ubiquitous connectivity through the Internet have disrupted traditional software engineering boundaries and practices. People and services are no longer bound by physical locations. Computational devices are no longer bound to the devices that host them. Communication, in its broadest sense, is no longer bounded in time or place. The Software Engineering & Design (SEAD) group at the Open University (OU) is leading software engineering research in this new reality that requires a paradigm shift in the way software is developed and used. This platform grant will grow and sustain strategic, multi-disciplinary, crosscutting research activities that underpin the advances in software engineering required to build the pervasive and ubiquitous computing systems that will be tightly woven into the fabric of a complex and changing socio-technical world. In addition to sustaining and growing the SEAD group at the OU and supporting its continued collaboration with the Social Psychology research group at the University of Exeter, the SAUSE platform will also enable the group to have lasting impact across several application domains such as healthcare, aviation, policing, and sustainability. The grant will allow the team to enhance the existing partner networks in these areas and to develop impact pathways for their research, going beyond the scope and lifetime of individual research projects.

The project will investigate resilience and automation in the socio-technical system that supports software development, which includes people , technical infrastructure, processes and artefacts . Recognising that users are developers and developers are users in different contexts, we’d like to investigate two communities that represent different points along that dimension: commercial software development community and professional end user development community.

In software engineering, developers must respond quickly to the change requests raised by product customers. For example, when a bug report arrives, developers need to locate relevant code to patch the bug. External to the product teams, however, customers can only describe the bug using their own (natural) language, with limited understanding of the software. Such a description may not reflect the developers’ rich experience and their deeper understanding of the software architecture, design, data structures, algorithms, etc. Hence substantial amount of time must be wasted on triaging, from a vast repository, where to start the investigation. It is like searching needles in a haystack [1]. To tackle this problem, existing work on bug localisation employs various heuristics in the natural language description of bugs and code to achieve state-of-the-art performance on information retrieval metrics, such as precision/recall and top-K [1]. Precision measures how many pieces of recommended code are indeed related to an average bug report; recall measures how many pieces of code known relevant to a bug report have been recommended; and top-K measures how many out of the K recommended pieces of code are relevant to a bug report. However, existing research methods developed since Zhou et al. [2] are competing against a benchmark of only 4~6 medium-sized open-source Java projects mostly taken from Eclipse development. Although various information retrieval approaches have been attempted, and compared on top of this benchmark, there is a lot of uncertainty as to how they apply to Huawei’s contexts. In this project, we will tackle these challenges with practicability and adoptability in mind.

In pair programming, two people work together on the same problem, collaboratively writing code (in a shared window or screen) whilst communicating with each other (via voice or text, face-to-face or remotely). Pair programming is used by professional software developers (Goth, 2016) and has also been shown to help beginning programmers improve their skills. For example, McDowell et al. (2002) found that pair programming increases retention and helps students become better programmers. Also, pair programming was found particularly beneficial for women, helping them persist in computer science (Werner, 2004). The practicalities involved in setting up pair programming sessions can be a barrier to adopting it, both in professional and educational settings (especially distance learning settings). The proposed research explores the use of chatbots to address this issue. Chatbots are dialogue systems that respond to users questions and requests with natural language responses. Task-oriented chatbots are used in business applications while open-domain chatbots are used for entertainment. Katz et al. (2014) explore use of chatbots in an education task. Our work will advance dialogue systems research by applying it in the context of pair programming. Research topic and approach Building on the research strengths in dialogue modelling, question generation, pair programming and multi-modality at the OU and Toshiba, we propose to develop and investigate an AI buddy for pair programming, which can collaborate with a software developer on solving a programming problem (either from scratch or by completing a skeleton program). The AI should be able to: - detect if the developer is stuck and requires help - make contextual suggestions and provide hints - question the developer about the rationale for their programming and problem solving decisions - thereby helping them self-evaluate their work - generate contributions appropriate for the developer's emotional state A distinctive aspect of the project will be the use of multimodal communication. Depending on the strengths of the PhD applicant one or more of the following modalities will be used by the AI to interact with the student: - Spoken voice: analysis (content of what the student says, but also emotion) & synthesis (of the AIs dialogue contributions) - Webcam video to monitor: emotion, focus, level of understanding, cognitive load, etc. - Text-based chat (as an alternative to spoken interaction) - Embodied computer-animated representation of the AI buddy (Andrist 2019)