Amidst financial crises, an increasingly threatened environment, and countless worker rights conflicts more visible than ever due to growing globalisation, it can come as no surprise that ‘sustainability’ has grown from an attractive buzzword to a core component of almost all areas of public life. Governments and business are making changes – at times even reinventing their identities – to maximise not just what they have in the present, but what will be left of it in the future. Environmental, social and organisational sustainability have joined profit or loss as core elements of a company’s scorecard – or, if not that, then at least of their rhetoric. In many respects, a company that seeks only to compete financially is no longer competitive.

Along with this change has come a shift in (what many perceive to be) companies’ fiduciary duties. Companies can no longer prioritise their shareholders above all others – they are also accountable to their stakeholders: employees, consumers, communities local to their operations, etc. The primary vector for that accountability – the annual report – has evolved along with that change. A considerable majority of companies now publish a sustainability report in addition to or together with their (financial) annual report to disclose the key aspects of their non-financial performance, just as the annual report does for the financial. While the genre originally invited greenwashing – i.e. a focus on favourable presentation rather than substantive action – it continues to evolve towards greater (self-)regulation and mandatory disclosure.

A key difference between the genres, however, lies in its audience. While the financial report is a specialised genre chiefly aimed at experts (analysts and investors), the sustainability report, as a primary vector for non-financial accountability, addresses a potentially much wider group of heterogeneous stakeholders. That makes the sustainability report a specialised genre addressing a far less specialised group than its financially-oriented sibling. The financial report’s reputation for difficult language and impression management may not deter its expert audience, but that same language translating to sustainability reporting is likely to impede many of the readers not part of the core shareholder audience. This study investigates to what extent the sustainability report adapts to a wider audience, how that reflects in its language. and builds on those outcomes to address how sustainability reports can better accommodate the linguistic requirements of their wider audience. After a theoretical exploration of the concepts of sustainability (reporting) and readability, we pursue four primary avenues of research into the linguistic genre traits of sustainability reporting.

A first, broad-scope exploration of a 470-document, approximately 2.75-million token corpus of year-2012 sustainability reports and their accompanying letters – as well as letters from annual reports – finds that sustainability content is, if anything, less readable than financial content. It makes this observation not just based on the conventional formula-based approach to readability measurement, but also integrates Natural Language Processing as a more nuanced estimate of a text’s characteristics. Notably, this enquiry finds little, if any evidence of obfuscation – companies concealing unfavourable outcomes in complex language – in spite of considerable evidence of the phenomenon in financial reports according to previous research. It does, however, find a notable association between a report’s language variety and its readability and syntax. For instance, UK reports are significantly more passive than US ones, which might impact the efficacy of cross-varietal reporting.

Second, we investigated to what extent the readability of a report’s accompanying letter influenced readers’ perceptions of the company by creating a lightly and heavily simplified version of such a letter and submitting them to a panel. Those amongst the panellists unfamiliar with the genre reacted more positively to the original, most complex version of the letter than those with previous experience did, while neither of the simplified versions showed a difference between the groups. These results are likely indicative of the efficacy of sustainability reports’ language, in spite of its difficulty: it may well have a positive influence on how laypersons – which many potential readers amongst its stakeholders may be – perceive the company. However, neither group’s opinion of the text or company declined as readability went up, signalling that it may be safer for companies to attempt to simplify their disclosures than they might expect.

A third inquiry investigated the extent to which the ‘Pollyanna Effect’ – a bias towards positivity in (reporting) language - occurs in sustainability reporting. This inquiry relied on annotators following an annotation scheme designed to account for the genre’s potential tension between various areas of performance, e.g. environmental stewardship coming at the cost of short-term profit. Analogous with financial reports’ reputation of excessive positivity, we found that reports likely contained more positive information than an aim of balance would suggest. More significantly, this enquiry, contrary to the first, did show an association between the positivity of the outcomes reported on and use of company-oriented agency framing; in other words, companies appear to attribute positive outcomes to themselves more than they do negative ones.

A final component attempted to discern what influences perception of readability for those unfamiliar with the genre. In an experiment based on human perception of readability, we attempted to distil the chief (perceived) contributors to readability or a lack thereof from scorers’ comments. This informed the training of a genre-adapted machine learning system meant to improve upon readability prediction for sustainability reporting. Traditional formulae tend to consistently rate corporate reporting as very difficult, but machine learning allows much greater nuance than ‘generic’ readability measures do, both in the features that inform it and the resolution of scores it can assign. This proof of concept demonstrates the merit of Natural Language Processing in helping authors write more accessible reports that better meet their widening audience’s needs.

In that respect, this dissertation attempts to contribute to corpus linguistics research through genre adaptation of a diverse range of linguistic techniques, including NLP techniques, with machine learning as the most notable. It contributes to business communication research by analysing the linguistically underexamined genre of sustainability reporting.

In summary, we explore a number of ways in which the linguistic properties of sustainability reports keep the genre from achieving its potential as a vector for accountability and engagement towards a broad, heterogeneous group of stakeholders. The final part of the study attempts to formulate avenues for improvement. Based on the aforementioned results, we see the greatest potential for improvement in companies using an active, narrative style that creates engagement through personal pronouns, and trust that simpler language will not damage their perceived credibility or professionalism. On a process level, we find that although readability measures can be a valuable aid or ‘second opinion’, an author’s own judgment of readability is inevitably more complete than any heuristic might be. However, a genre-adapted learner can still vastly improve on traditional readability formulae. Readability initiatives that set formula-based targets are likely to only hamper the genre by encouraging authors to ‘write to formulae’. The author and editor retain access to the best yardstick for readability – the human ability to process language holistically. Nevertheless, easy reading remains ‘damned hard writing’.