Presentation on theme: "Paul Baker July 2011. Structure 1 Background to a corpus approach to (C)DA 2 A 9 stage model description illustration evaluation 3 An experiment in analyst."— Presentation transcript:
Structure 1 Background to a corpus approach to (C)DA 2 A 9 stage model description illustration evaluation 3 An experiment in analyst consistency
Discourse “a set of meanings, metaphors, representations, images, stories, statements and so on that in some way together produce a particular version of events… Surrounding any one object, event, person etc., there may be a variety of different discourses, each with a different story to tell about the world, a different way of representing it to the world.” (Burr 1995: 48)
Critical discourse analysis Identifies discourses in texts A politically driven form of analysis Several levels of analysis e.g. Text Production and reception Intertextuality Social context (society’s politics, history) Usually qualitative and with small datasets
Criticisms of CDA “Your analysis will be the record of whatever partial interpretation suits your own agenda” (Widdowson 1998: 148) “what is distinctive about Critical Discourse Analysis is that it is resolutely uncritical of its own discursive practices” (Widdowson 1998: 151)
The benefits of using corpora Interpretation “grounded in systematic language description” Need to account for much larger amounts of text Accurate and fast calculations Corpus-driven techniques reduce researcher political and cognitive bias (primacy effect, clustering illusion). Potential to find exceptional cases
Contributors to the approach Michael Stubbs (1994, 1996) Gerlinder Mautner (1995, 2004, 2007, 2009) Carmen Caldas Coulthard (1995) Ramesh Krishnamurthy (1996) Alan Partington (Corpus Assisted Discourse Studies) (2004, 2008, 2010) Susan Hunston (2002, 2003) Kieran O’Halloran and Caroline Coffin (2004) Paul Baker (2005, 2006, 2008)
My own research Gay Men (2004a,b, 2005) Refugees and asylum seekers (2005, 2008a,b) Fox-hunting (2006) Islam and Muslims (2010) Gender (2006, 2008, 2010) Foreign doctors (2011)
Nine stage Corpus-assisted CDA (Baker et al 2008) 1Context-based analysis of topic via history/politics/culture/etymology Identify existing topoi/discourses/strategies via wider reading, reference to other CDA studies. 2Establish research questions/corpus building procedures. 3Corpus analysis of frequencies, clusters, keywords, dispersion etc – identify potential sites of interest in the corpus along with possible discourses, strategies, relate to those existing in the literature 4Qualitative or CDA analysis of a smaller, representative set of data (e.g. concordances of certain lexical items or of a particular text or set of texts within the corpus) – identify discourses, strategies etc. 5Formulation of new hypotheses or research questions 6Further corpus analysis based on new hypotheses, identify further discourses, strategies etc. 7Analysis of intertextuality or interdiscursivity based on findings from corpus analysis 8New hypotheses 9Further corpus analysis, identify additional discourses, strategies etc
Stage 1 – wider reading Van Dijk’s (1987: 58) four topic classes for racist discourses They are different, they do not adapt, they are involved in negative acts and they threaten our socio-economic interests Karim’s (2006: 119-20) four primary stereotypes of Muslims: ‘having fabulous but undeserved wealth (they have not earned it), being barbaric and regressive, indulging in sexual excess, and… the “violent Muslim”’
Stage 2 – corpus building/research questions Do representations of Muslims match with van Dijk’s or Karim’s categories? What differences occur over time or between different types of newspapers? 143 million word corpus (200,037 articles)
Stage 3 – corpus-driven analysis Keyword comparison of broadsheet and tabloids Omar Bakri and Abu Hamza were strong tabloid keywords
Stage 4 – Concordance analysis EVIL hook-handed Muslim cleric Abu Hamza is using a legal trick to delay getting the boot from Britain for THREE years and rake in thousands more in hand-outs. The People, March 21 st, 2004 RANTING Muslim cleric Omar Bakri Mohammed pulled off another handouts coup by claiming disability benefit to get a £28,000 car, complete with satellite navigation system. Yet he walked into the showroom with barely a limp. The Sun May, 16 th, 2005
Stage 5 – new hypotheses Stories about ‘undeserving Muslims on benefits’ originate in the tabloids and influence the discourse of right- leaning broadsheets.
Stage 6 – corpus analysis INVESTIGATORS discovered £180,000 in a London bank account held by a radical Muslim cleric accused of fomenting and financing terrorism. Sheikh Abu Qatada, who lives on benefits in Acton, west London, had his assets frozen at the weekend after appearing on a Treasury list of people suspected of "committing or providing material support for acts of terrorism". Telegraph, October 18 th, 2001 The taxpayer will also fund at least £12,000 per year in benefits for Qatada, his wife and five children, even though Qatada was once found to be carrying £170,000 in cash when he was stopped by police.” Telegraph, June 18 th, 2008
Stage 7 - intertextuality DAVID Blunkett has ordered a benefits blitz on Islamic hate clerics who sponge off the state. The Daily Express, August 17 th, 2005 So, David Blunkett is to have a blitz on Muslim clerics who sponge off the state ("Benefits blitz on the hate preachers", August 17). Daily Express, letters, August 18 th, 2005
Stage 8 – new hypotheses Some newspapers use readers’ letters as a legitimation strategy to print more Islamophobic representations.
Stage 9 – analysis of corpus PIGGYBANKS are facing the axe - because some Muslims could take offence. Britain's top High Street banks have ruled the money-boxes are politically incorrect. But last night the move sparked snoutrage. And one of Britain's four Muslim MPs, Khalid Mahmoud, said: "A piggybank is just an ornament. Muslims would never be seriously offended." The Star, October 24 th, 2005
Stage 9 – analysis of corpus TEXT MANIACS (The Star, October 25 th, 2005) muslims r offended by our piggy banks! ? Then the £56 me n ma wife n ma 4 girls have got in our piggy bank 2 help the ppl in pakistan wil b spent on a fry up. Y shud we change r way of life just 2 stop offending muslims. they aint neva gonna change theirs. Maybe they shud try eating pork. a nice bacon sarnie cud change any1's mind.
Evaluation of method Fruitful in identifying numerous features of the corpus which could not have been considered in advance Researchers need to be trained in conducting different forms of analysis (or utilise a team of researchers with different skills). Each stage can open up multiple pathways and/or hypotheses, not all of which can be followed due to time and money constraints. At times, the number of ‘directions’ that the research could go in felt overwhelming and endless. Did researcher bias impact on paths followed, outcomes?
An inter-analyst consistency experiment Subjects – 5 analysts all with prior experience of combining corpus linguistics and discourse analysis or CDA. The corpus search term: “foreign doctors” + similar terms All British national newspapers (about half a million words) Research Question “How are foreign doctors represented in the British press 2000-10?” Any form of corpus methods or software allowed
Corpus Tool Software used 12345 WordSmith5 ✓✓✓✓ Antconc ✓ Wmatrix ✓ Other reference corpus ✓✓✓
Findings Finding12345 Poor language ✓✓✓✓✓ incompetence ✓✓✓✓ Need to regulate/test ✓✓✓✓ killer/killed ✓✓✓✓ NHS shortages ✓✓✓✓ Flood metaphor ✓✓✓ Invasion (by Germans) ✓✓ Taking British jobs ✓✓ Generalising ✓✓ Terrorist threat ✓ Health risk ✓ FDs ignoring vacancies ✓ Needed in own country ✓ Very expensive ✓ NHS Cost-cutting measure ✓ Evil/villainous ✓ “out of hours” ✓ “alive” ✓ “NHS red tape” ✓ NHS failing ✓ Tory vs Labour ✓ Profession vs government ✓ FDs under stress ✓ Junior doctors at risk ✓ “Racist” if we complain ✓ Consultants as warners ✓
Shared findings 12345 Total findings 1-25%29%33% 58% 2--29%15%36%38% 3---13%31%46% 4----25%19% 5----- e.g. calculated by: # of shared findings made by analyst 1 and 2 # of findings made by either analyst 1 or 2 Analysts 2 and 5 – most similar findings (36%) Analysts 3 and 4 – hardly any similar findings (13%)
Did the analysts uncover different things? Only one finding (4%) discovered by every single analyst About a quarter of findings discovered by the majority (3+) of analysts. But 65% of findings only discovered by 1 analyst Distinction between “major” and “minor” findings. Analysts agreed on the overall ‘feel’ of the data, but the specifics differed. Two “productive” strategies: spend a long time on one technique (1, 3), use lots of different techniques (2). Time and ability/experience are important factors.
Conclusion Corpus tools give a reasonably high degree of consistency for identifying larger patterns Caution in concluding the techniques remove all bias Procedures direct attention to unforeseen aspects of the data Resulting in more interesting questions and hypotheses