Mehler A., Sharoff S., Santini M. (Eds.). Genres on the Web: Computational Models and Empirical Studies

Springer, 2010. — 378 p. — (Text, Speech and Language Technology). — ISBN-10 9048191777, ISBN-13 978-9048191772.As a reader, I’m looking for two things from a new book on genre. First, does it offer some new tools for analysing genres; and second, does it explore genres that haven’t been much studied before? Genres on the Web delivers brilliantly on both accounts, introducing as it does a host of computational perspectives on genre classification and focussing as it does on a range of newly emerging electronic genres. Lacking expertise in the computational modelling thematised throughout the book I can’t do much more here than express my fascination with the questions tackled and methods deployed. Having expertise in functional linguistics and its deployment in genrebased literacy programs I can perhaps offer a few observations that might help push this and comparable endeavours along.
First some comments as a functional linguist. Characterising almost all the papers is a two-level approach nicely summarised by Stein et al. in their Table
8.1. On the one hand we have a web genre palette, with many alternative classifications of genres; on the other hand we have document representation, with the many alternative sets of features used to explore web data in relation to genre. The most striking thing about this perspective to me is its relatively flat approach as far as social context and its realisation in language and attendant modalities of communication is concerned.
In systemic functional linguistics for example, it is standard practice to explore variation across texts from the perspectives of field, tenor and mode as well as genre. Field is concerned with institutional practice – domestic activity, sport and recreation, administration and technology, science, social science and humanities and so on. Tenor is concerned with social relations negotiated – in relation to power (equal/unequal) and solidarity (intimate, collegial, professional etc.). Mode is concerned with the affordances of the channel of communication – how does the technology affect interactivity (both type and immediacy), degree of abstraction (e.g. texts accompanying physical behaviour, recounting it, reflecting on it, theorising it) and intermodality (the contribution of language, image, sound, gesture etc. to the text at hand). In my own work genre is then deployed to describe how a culture combines field, tenor and mode variables into recurrent configurations of meaning and phases these into the unfolding stages typifying that social process.
When I referred to a flat model of social context above what I meant was that in this book these four contextual variables tend to be conflated into a single taxonomy of text types, without there being any apparent theoretically informed set of principles for the flattening. It may well be of course that for one reason or another we do want a simple model of social context and may wish to foreground one field or mode or tenor variable over another. But it might prove more useful to begin with a richer theory of context than we need for any one task, and flatten it in principle, than to try and build a parsimonious model from the start, and complicate it over time.
Turning to document representation, once again from the perspective of systemic functional linguistics, it is standard practice to explore representation in language (and other modalities of communication) from the perspective of various hierarchies and complementarities. The chief hierarchies used are rank (how large are the units considered – e.g. word, phrase, clause, phase, stage, text) and strata (which level of abstraction from materiality is being considered – phonology/graphology, lexicogrammar or discourse semantics). The chief complementarity used is metafunction (are we considering the ideational meanings used to naturalise a picture of reality, the interpersonal meanings used to negotiate social relationships or the textual meanings used to weave these together as waves of information in interpretable discourse).
The meanings dispersed across these ranks, strata and metafunctions are regularly collapsed into a list of descriptive features in this volume, when for different purposes one might want to be selective or value some features over others. Exacerbating this is an apparent need to foreground relatively low-level formal features which are easily computable, since manual analysis is too slow and costly, and in any case so much of the research here is focussed on the automatic retrieval of genres. Beyond this, as Kim and Ross point out, texts are regularly treated as bags of features, as if the timing of their realisation plays no significant part in the recognition of a genre. What saddens me here is the gulf between computational and linguistically informed modelling of genres, for which I know my colleagues in linguistics are responsible – since for the most part they work on form not meaning, and focus on the form of clauses and syllables, not discourse (they still think a language is a set of sentences rather than a communication system instantiated through an indefinitely large lattice of texts).
Next some comments as a functional linguist working in language and education programs over three decades. From the start we of course faced the problem of classifying texts – in our case the genres that students needed to read and write in primary, secondary and tertiary sectors of education, and their relation to workplace discourse and professional development therein. One thing we learned from this work was to be wary of the folk-classifications of genres used by educators. Our primary school teachers for example called everything their students wrote a story, when in fact, from a linguistic perspective, the students engaged in a range of genres. Complicating this was their tendency to evaluate everything the students wrote as a story, in spite of suggesting to students that they choose their own topics or even that they write in any form they choose. As an issue of social justice, we felt we had to replace the folk-categorisation with a linguistically informed one, and take the further step of insisting that this uncommon sense classification be shared between teachers and students. The moral of this experience I feel is that we need to treat folksonomies with great caution when classifying genres, and not expect users to be able to easily bring to consciousness or even demonstrate in practice a genre classification that will best suit the purposes of our own research.
Throughout this literacy focussed action research we have lacked the funding and computational tools to undertake the systematic quantitative analysis thematised in this volume. Instead we had to rely on manual analysis of texts our teacher linguists selected as representative (depending as they did on their own experience, advice from teachers, assessment processes and textbook exemplars). This meant we could build up a picture of genres based on thick descriptions of all the levels of analysis I worried about being flattened above; the great weakness of this approach of course is replicability – were our few texts in fact representative and would quantitative analysis support our findings over time? In practice, the only confirmation we received that we were on the right track lay in the literacy progress of our students, since we were interested in genre because we wanted to redistribute the meaning potential of our culture more evenly than schools have been able to do in the past.
At this point I suspect that most of the authors in this volume would throw up their hands in despair of finding anything useful in our work. So let me just end on a note of caution. What if genres cannot be robustly characterised on the basis of just a few easily computable formal features? What if a flat approach to contextual variables and representational features simplifies research to the point where it is hard to see how the texts considered could have evolved as realisations of the genres members of our culture use to live? Would we be wise to complement flat computationally based quantitative analysis with thick manual qualitative description and see where the two trajectories lead us? And do we need to balance commercially driven research with ideologically committed initiatives (who for example will benefit from the genre informed search engines inspiring so many of the papers herein)? I’ll stop here, concerned that this preface is turning into a post-script, or even a chapter in a book where prefacing is where I barely belong! My thanks to the editors for opening up this work, which will prove indispensable for readers with many converging concerns. I’ll do what I can to point my students and colleagues in the direction of the transdisciplinary dialogue which I’m sure will be inspired by the genre analysts dialoguing here.Riding the Rough Waves of Genre on the Web.
Identifying the Sources of Web Genres
Conventions and Mutual Expectations.
Identification of Web Genres by UserWarrant.
Problems in the Use-Centered Development of a Taxonomy of Web Genres.
Automatic Web Genre Identification
Cross-Testing a Genre Classification Model for theWeb.
Formulating Representative Features with Respect to Genre Classification.
In the Garden and in the Jungle.
Web Genre Analysis: Use Cases, Retrieval Models, and Implementation Issues.
Marrying Relevance and Genre Rankings: An Exploratory Study.
Structure-Oriented Models of Web Genres
Classification of Web Sites at Super-Genre Level.
Mining Graph Patterns inWeb-Based Systems: A Conceptual View.
Genre Connectivity and Genre Drift in aWeb of Genres.
Case Studies of Web Genres
Genre Emergence in Amateur Flash.
Variation Among Blogs: A Multi-Dimensional Analysis.
Evolving Genres in Online Domains: The Hybrid Genre of the Participatory News Article.
Prospect
Any Land in Sight?