The British National Corpus (BNC) has been available to the research community for more than two decades. Over the course of its three editions to date, this 100-million-word database, containing samples of both transcribed speech and written texts representing British English of the 1990s and earlier, has established itself as a valuable resource used around the world in a wide range of language-related applications.

The importance and widespread popularity of the BNC can be accounted for in terms of an array of highly desirable properties it displays. Therefore, it comes as no surprise that the tool is often cited as a prime example of a prototypical corpus (Gilquin & Gries, Reference Gilquin and Gries2009: 6). First of all, being a national corpus and the first of its kind, it has fulfilled the much-needed role of a general reference corpus of modern British English, and its design, aiming to maximize representativeness and balance, has repeatedly served as a model for other national corpora (Xiao, Reference Xiao, Lüdeling and Kytö2008: 383–85). The size of the corpus constitutes another major advantage. It is worth noting that the database contains one of the largest and most widely used collections of spoken data (10 million words) available to the public at no cost (Love et al., Reference Love, Dembry, Hardie, Brezina and McEnery2017a: 321–23). The XML edition, which is the latest version of the corpus, can currently be downloaded free of charge from the University of Oxford Text Archive or accessed via several user-friendly interfaces, including the BNCWeb provided by Lancaster University and BYU–BNC maintained by Brigham Young University.

Notwithstanding its many useful features, the BNC is by design a static dataset, which means that the moment when it would no longer represent modern British English sensu stricto was bound to come sooner or later. Fortunately, the growing demand for a new, similarly large and functional but more contemporary database was aptly addressed by a team of experts from Lancaster University when they undertook, several years ago, the challenging task of designing and compiling a successor to one of the most important corpora in the world. While the written component of the new corpus is still under construction and will be released at a later time, the spoken part, prepared in collaboration with Cambridge University Press and containing 11.5 million words of orthographically-transcribed conversations from the years 2012–2016, was made accessible to the public in September 2017. Before that, in June 2017, the much-awaited release of, what is now known as, the Spoken British National Corpus 2014 (Spoken BNC2014) was celebrated with a symposium hosted by Lancaster University. The event was a valuable source of first-hand information about the new corpus, and therefore it merits a wider account. In this report I demonstrate that the event provided the opportunity to learn about the design and compilation of the corpus. I also discuss ways in which the event showcased research conducted with the use of the new dataset.

The Spoken BNC2014 Symposium took place on 26 June 2017. This one-day event was organized by the ESRC Centre for Corpus Approaches to Social Science (CASS) at Lancaster University and was held in one of the lecture theatres of the Faraday Lecture Theatre building. The symposium brought together an engaged audience of more than 50 researchers and students, many of whom were also participants of the Lancaster Summer Schools in Corpus Linguistics and other Digital Methods beginning on the following day.

The programme consisted of five main presentations and could informally be divided into two parts – an introduction to the design and content of the Spoken BNC2014 given by the corpus compilers, and a series of papers exploring specific research topics with the use of the data from the new corpus. The talks in the second part were in most cases delivered by the recipients of the early access data grants, who had obtained access to a five-million-word subset of the Spoken BNC2014 in January 2016.

After a short introduction by Robbie Love, the main organizer of the symposium from Lancaster University, the meeting was officially opened by the CASS Director, Andrew Hardie. Having greeted the presenters and the attendees, he provided a concise picture of the impressive corpus linguistic traditions of Lancaster University, ranging from corpus construction (LOB, BNC, BE06, as well as, more recently, a number of specialized corpora), through development of powerful tools for corpus analysis (Wmatrix, CQPweb) and annotation (USAS, CLAWS), to conducting corpus-based research that cuts across a range of disciplines in the humanities and social sciences. With this background of expertise, it seems only natural that CASS has taken up the task of building the successor to the original British National Corpus, now referred to as ‘the old BNC’ or, somewhat more formally, ‘the BNC1994’. As was pointed out by Andrew Hardie, the need for a more modern general corpus of British English hardly requires any justification. Since the BNC 1994 was created before the mass popularization of the Internet, it no longer represents current British English and yet, for lack of a comparably large resource, it continues to be used for research purposes as a representative sample of contemporary spoken and written language. Thus, it is all the more opportune that the Spoken BNC2014, compiled in collaboration with Cambridge University Press, can already be introduced to a wider audience. Additionally, the written component of the corpus, although it was not the focus of the symposium, is also nearing completion and was promised to be released in September 2018.

The first full presentation of the symposium was an extensive tour of the complex process of creating the corpus, and it was delivered by the corpus compilers – Robbie Love and Andrew Hardie again, as well as Claire Dembry from Cambridge University Press. The title of the talk was ‘The Spoken BNC2014: designing and building a corpus of everyday conversations’. By way of setting the stage, the motivations for creating a new corpus of spoken British English were recapitulated, but apart from the arguments already mentioned above, two other points were also strongly emphasized. On the one hand, the Spoken BNC2014 is hoped to open new avenues for the study of short-term language change that has taken place over the last 20 years, as this is the interval between the sampling periods of the new corpus and the conversational part of its predecessor, the BNC1994. On the other hand, from the perspective of English language teaching, the corpus is expected to become a resource that will facilitate reducing the unfortunate gap between the language that is actually produced in natural communicative settings and the much less authentic counterpart that can often be encountered in teaching materials.

In the next step, the presenters focused on the division of labour in their collaborative enterprise, which was as follows. Both parties, that is, Lancaster University and Cambridge University Press, designed the corpus, recruited participants through media campaigns, and bore the costs of the project. The collection and transcription of data was the responsibility of Cambridge University Press, whereas the Lancaster team were in charge of the methodological considerations, documentation, and annotation, as well as storing the corpus on the CQPweb server and, ultimately, preparing it for public release.

The outcome of this collaboration, which is the Spoken BNC2014, is a large 11.5-million-word corpus containing transcripts of 1000 hours of conversations produced by nearly 700 speakers in the years 2012–2016. Unlike the Spoken BNC1994, which has both a demographic and a context-governed component, the BNC2014 is a fully demographically-sampled corpus. It is richly annotated with metadata such as speakers’ age, gender, education, occupation, dialect, linguistic origin, nationality, as well as several other features, including some further contextual information pertaining specifically to the recordings, for example, title, date, length, location of the recording, relationship between speakers. The metadata provided in the BNC2014 may be characterized as somewhat richer than in the BNC1994, and the presenters often emphasized that the tension between introducing the necessary methodological improvements and ensuring a sufficient degree of comparability between the two corpora had been present throughout the whole process of building the database. For instance, one major innovation that made its way into the metadata encoded in the corpus concerned the classification of regional dialects. As discussed by Robbie Love, in the hope of obtaining more accurate information than could be derived from the standard metadata collected from each speaker, the speakers were asked to provide their own dialect categorizations, which were subsequently mapped onto the names of the regions used by the Office for National Statistics.

The remainder of the talk focused on the transcription process. Claire Dembry explained that the original scheme used in the BNC1994 was applied in BNC2014, only it was made considerably more explicit. Additionally, the transcription included information on the final stages of processing the corpus files, such as conversion to XML and metadata encoding, lemmatization and part-of-speech tagging (using both the CLAWS C6 tagset and another, simplified system of POS tags). This was briefly discussed by Andrew Hardie, who also gave a short presentation of some of the functionalities of CQPweb for the early access data.

The second part of the symposium consisted of four equally interesting and informative papers. These reported on specific analyses and case studies based on the early access data from the Spoken BNC2014 typically combined with the data from the BNC1994 for the purposes of comparison or otherwise. That session began with a talk entitled ‘That's fucking brilliant: The use of fucking in the Spoken BNC2014’ by Karin Aijmer from the University of Gothenburg. The presentation concerned the syntactic, semantic, pragmatic and sociolinguistic patterning of the intensifier fucking, investigated from the perspective of short-term language change in the context of other intensifiers denoting a high degree of a property (such as really, very, awfully, etc.). One of the findings was that the overall frequency of the intensifier had increased over the 20-year period under analysis. From the analysis of its collocational range one learns that fucking typically pre-modifies negative adjectives (e.g., horrible, weird, etc.), although some uses with positive or neutral adjectives (e.g., funny, amazing, or long) are also attested. While this could be a sign of an ongoing process of delexicalization, or weakening of the intensifier's original meaning, Karin Aijmer stressed that fucking is still predominantly used with negative prosody, and no major increase in frequency can be observed for the remaining patterns. The word generally fulfills expressive and emotional functions and is also currently used as an identity marker, as can be concluded from its frequent co-occurrence with fashionable adjectives, such as awesome, massive, or cheesy. When it comes to the distribution by gender, fucking is nowadays used more frequently by women than by men, and it is also characterized by age-grading, being primarily attested in the 19–29 age group. This also suggests that the intensifier is currently the most popular among younger speakers, although any effective comparison between the two corpora would have to take into account the different age ranges they provide. Summing up her talk, Karin Ajmer concluded that the data spanning two decades may help observe some interesting changes in frequencies and collocational preferences relative to a number of (socio)linguistic variables, but it may not provide enough evidence for some major processes of language change.

The next presentation was given by Jacqueline Laws from the University of Reading. The title of the paper, co-authored by Chris Ryder and Sylvia Jaworska, was ‘A diachronic corpus-based study into the effects of age and gender on the usage patterns of verb-forming suffixation in spoken British English’. The purpose of the study was to investigate the changes that can be observed in four verb-forming suffixes –ate, –en, –ify, and –ise over the period of two decades, i.e. again on the basis of the data from the two BNC corpora. Three variables were considered for that purpose: diversity, density (operationalized respectively as type and token frequencies), and creativity (estimated on the basis of the frequency bands adopted in the Oxford English Dictionary). Among the findings reported by Jacqueline Laws was the increased density and diversity of all suffixes except for –en, for which a significant decline in frequency was observed. As far as the effects of age and gender are concerned, the greatest diversity was attested for the 35–59 age range in the BNC2014 data, and women were shown to be nowadays more creative than men in their use of the suffix –ify, but no similar difference between genders was found for the remaining forms.

Another paper, entitled ‘The British National Corpus as a sociolinguistic dataset: Exploring individual and social variation’, was presented by Susan Reichelt from Lancaster University. She discussed the efforts to create two comparable subsets from the BNC1994 and the BNC2014 in a way that could enable the study of language variation. She explained that a balanced and representative sociolinguistic dataset should include similar and sufficiently large numbers of speakers representing each sociolinguistic category, such as age, gender, and social class. Given that it is also desirable to have similar amounts of speech from every speaker, it is a challenge in itself to adapt one corpus to the demands of sociolinguistic analysis. Also, it becomes even more difficult when, for the purposes of the study of both language variation and change, one additionally needs to achieve sufficient similarity not only within a single corpus, but also between two corpora under investigation. The conversational component of the BNC1994 and the BNC2014 differ substantially in terms of the total number of words (4 million vs. 11.5 million), the number of speakers (1,233 vs. 668), and finally, the highest as well as the lowest word count per speaker (70,000 vs. 351,000 and 1 vs. 18, respectively). Thus, in order to create two comparable datasets, Susan Reichelt first adjusted the metadata for the differences between the two corpora in terms of socio-demographic variables and then attempted to create two subsets with maximally equal distributions of speakers across the categories of age, gender, social class and regional background. At present the subsets have over 250 speakers each, and they will be made available on the CQPweb server at a later time. The presentation concluded with a demonstration of an apparent-time analysis of selected intensifiers performed on the BNC1994 subset.

The last presentation in the symposium was delivered by Jonathan Culpeper and Mathew Gillings from Lancaster University. The talk was entitled ‘Politeness variation in England: A north-south divide?’ On the basis of the combined data from the BNC1994 and the BNC2014, they tested a hypothesis, emerging from a widespread popular view, that speakers from the north of England are generally characterized by greater politeness than those from the south. The authors were mainly interested in three types of politeness, i.e. tentativeness, deference and solidarity politeness, as operationalized by the frequency of their corresponding politeness expressions. Some examples of politeness markers include constructions such as could you and please (expressing tentativeness), sir, madam and thank you (signalling deference), and finally, love, mate and cheers (conveying solidarity). While it was hypothesized that the northern data would be characterized by lower numbers of tentative politeness expressions, and higher numbers of deference and solidarity politeness markers, the results actually revealed the opposite tendencies across all three types of politeness. The presenters concluded that what is often perceived as politeness variation between the north and the south of England may be an effect of formality, since the data do in fact show variation along the formality cline in that more formal politeness expressions tend to be preferred in the south and dispreferred in the north.

It is also worth mentioning that an interesting comment on the results was made by Andrew Hardie during the following discussion. He remarked that politeness is primarily evaluated on the basis of interaction between strangers, and as the two datasets comprise mainly intimate conversations, the data may not lend itself easily to analyses in terms of politeness parameters. Shortly after the final discussion, the symposium was closed by Robbie Love, who thanked the presenters and the attendees for their participation.

The Lancaster symposium was one of three complementary public events, each with its own specific scope and function, organized with the aim of promoting the Spoken BNC2014 in the relatively final stages of its compilation and, more generally, undertaken as part of the process of preparing the corpus for the upcoming release. The remaining events comprised the already-mentioned release of the early access data grants that resulted in a number of research studies, including the ones summarized above, and the workshop which took place in Birmingham during the 9th Corpus Linguistics Conference in July 2017, i.e. one month after the symposium.

The two perspectives on the Spoken BNC2014 presented during the symposium, that of the compilers and that of the first users of the corpus, undoubtedly enabled a fuller view of various theoretical and methodological aspects of the new database than either one of them alone would have done. The first part of the event focused on the creation of the corpus, providing details of its design and content, and pointed to a range of intended future applications of the corpus tool. The papers given in the second part concretized those applications and demonstrated that the Spoken BNC2014, used in conjunction with its predecessor, can be fruitfully employed to investigate patterns of short-term language change, as it was especially shown in the talks by Karin Aijmer and Jacqueline Laws. They also proved that the rich set of metadata encoded in the corpus makes it a useful resource for addressing sociolinguistic research questions. Although the specific objectives of the individual case studies were naturally very different, all of them involved the use of at least some of the key sociolinguistic variables, such as age, gender, socio-economic status or geographical region. In addition, the issue was also explored from a theoretical and methodological angle by Susan Reichelt, who discussed the complexity of the process of creating a balanced sociolinguistic dataset based on the two BNC corpora.

It should be noted that the question of comparability of the Spoken BNC2014 to its predecessor actually resounded through all the contributions of the symposium and can be considered one of the most important take-home messages of the whole event. As it was emphasized by Robbie Love and his collaborators, while ensuring comparability between both corpora was one of the main objectives for the development of the Spoken BNC2014, a number of methodological improvements were implemented as necessary, and therefore the two datasets ultimately differ in several respects, i.e. their size, sampling criteria, metadata categorizations and tagsets applied for POS tagging. The differences should by no means be considered a disadvantage. On the contrary, all the new solutions adopted were informed by some important and well-documented considerations, but the issue will nevertheless need to borne in mind by prospective users of the corpus, many of whom will be interested in studying phenomena of recent language change. Some important resources that can be consulted for more details include the special issue of the International Journal of Corpus Linguistics, edited by Tony McEnery, Robbie Love and Vaclav Brezina (‘Compiling and analysing the Spoken British National Corpus 2014’), which features a comprehensive introduction to the Spoken BNC2014 (Love et al., Reference Love, Dembry, Hardie, Brezina and McEnery2017a), and several research articles drawing on the early access data, as well as the sociolinguistically-oriented collection due to appear shortly in the Routledge Advances in Corpus Collection series (Brezina, Love & Aijmer, Reference Brezina, Love and Aijmer2018). Finally, the most exhaustive account of the methodological decisions made at all stages of building the database is provided in the corpus documentation (Love, Hawtin & Hardie, Reference Love, Hawtin and Hardie2017b).

The Spoken BNC2014 was released on 25 September 2017. It is currently available via the CQPweb server, where it can be accessed free of charge upon acceptance of the terms of the end-user license agreement. The importance of the new research tool can hardly be overestimated. At present, it constitutes the largest and the most modern collection of conversational British English available to the public at no cost, and it is the successor to one of the most widely used spoken corpora in the world. These two qualities alone make it relatively straightforward to predict that the Spoken BNC2014 will become an invaluable resource for many diverse fields of application, from all branches of linguistics employing corpus-linguistic methodology, through lexicography and language teaching, to corpus compilation and beyond. Finally, as soon as it is joined by the written component, the research community worldwide will gain full access to the brand new general reference corpus of British English.