Again, the positive feature of diversity comes with a drawback. What, from one perspective, is a fascinating ‘multiplicity of voices’ is likely to turn into a significant hurdle when it comes to identifying, categorizing and selecting sources. Making out who says what on the internet can be difficult, but if we engage in any kind of socially sensitive type of linguistics, we need to do just that, separating institutional from individual voices, public-service from commercial offerings, and locating contributions in the wider sociopolitical frameworks that help us assess communicative intent. It is possible, up to a point, to limit web searches. Many search engines, such as Google, and web concordancers, such as Webcorp (www.webcorp.org.uk), allow one to specify domain extensions, so that only URLs that end in, say, ac.uk are accessed. KWICFinder (http://www.kwicfinder.com/www.webcorp.org.uk), allow one to specify domain extensions, so that only URLs that end in, say, ac.uk are accessed. KWICFinder (http://www.kwicfinder.com/)
is an even more versatile web concordancer, offering a variety of exclusion and inclusion criteria that help focus searches.11 Alternatively, one could focus on particular website types, identified on the basis of genre (Wakeford, 2000), such as personal homepages (Chandler and Roberts-Young, 1998; Caldas-Coulthard, 2005). I should like to argue strongly, however, in favour of not making these decisions at the very outset of a project, but to first establish, through an extensive search on the web, as comprehensive a picture as possible of the range of voices involved. After all, using web-based data is ultimately about broadening one’s analytic field of vision, and narrowing that field down too soon can easily defeat the whole point of the exercise.