Beth Learns Corpus Linguistics

I'm taking a Massive Open Online Course (MOOC) from Lancaster University in Corpus Linguistics. I'll update this blog with the things I've learned. Also check out bethlearns.wordpress.com

[bethlearnscorpuslinguistics-blog]

Using Corpus Linguistics

I considered using corpus linguistic methodology to analyse newspapers from N. Ireland but I'm not sure if this would be the best use of this course. I'm also not sure how to go about doing this yet: where to find the newspapers, how to transfer the corpus to machine-readable files etc. The practicalities of a project like this seem like a lot of work right now. 

I may instead use an existing corpus like the BNC or the Bank of English to continue my master's dissertation research on the use of 'like' as a discourse marker and quotative. This way I can take advantage of the transcribed speech files that are held within these corpora. (I wish I had known more about corpus linguistics when I doing this research this last year!)

Having begun to use AntConc as part of the Week 1 activities, I am more aware of the potential to use software to uncover larger overall trends within large amount of spoken data.

[bethlearnscorpuslinguistics-blog]

A real-life photograph of a real-life Neo-Firthian*.

*not actually a real-life photograph of a real-life Neo-Firthian

[bethlearnscorpuslinguistics-blog]

Corpus Linguistics: Method, Theory and Practice, McEnery, T. and Hardie, A (2012)

Chapter 1 

This chapter discusses the practical constraints of corpus linguistics, many of which I had not considered. I won't cover everything from this chapter, just the points which interest me the most!

Parliamentary Transcripts

The problems with using certain types of spoken corpora can be problematic due to the transcriptions of these. For example, the Hansard reports - transcripts of UK parliamentary debates - were not transcribed by linguists and therefore present the speech in a much more orderly fashion than was likely to have taken place. The natural turn-taking would have been omitted for the purposes of the report and so cannot represent true-to-life spoken language.This is a pity, although I'm sure that even in their current format, these parliamentary debates hold interesting political metaphor which could show the underlying views of the speakers. I guess it's important to differentiate the Hansard reports by classifying them as a distinct style, separate to natural speech. I wouldn't rule them out completely, but would analyse them differently to the COLT corpus of London teenage speech, for example. 

Corpus-driven linguistics v. Corpus-based Studies

So what I gather from this distinction is that some linguists ('Neo-Firthians', I believe) say that corpus linguistics is a theory of language: 

It is thus claimed that the corpus itself embodies its own theory of language (Tognini-Bonelli 2001: 84–5)

However, this course is entirely based on corpus-based studies. Prof. McEnery mentioned at the start that corpus linguistics is a methodology and not a theory. I don't fully understand the argument that the Neo-Firthians make (that's all later in the book!) but so far I can only see the field as an approach to analysing and gathering data (like Labov's field methods in Sociolinguistics), rather than a theory in its own right.

The Web as a Corpus

This section points out the potential of using the internet as a corpus. Unlike, for example the BNC, the web does not contain pre-defined genres of language, making an analysis more difficult. The web could be useful for a wider analysis of orthographical or grammatical changes over time, however. 

Annotation

I'm still unsure about the whole concept of corpus annotation. How it is carried out (by hand or by a computer)? How much detail is necessary? How much knowledge of syntax is necessary? (Drawing syntax trees was not exactly my strongest subject at uni...) Apparently those pesky Neo-Firthians don't like annotation because it takes away from the purity of the text.

I'm still not sure who these Neo-Firthians are but they sound like trouble-makers to me.

[bethlearnscorpuslinguistics-blog]

Snapshot Corpora

Two examples of snapshot corpora are the Brown and LOB corpora. Snapshot corpora allow the user to see a 'snapshot' of a particular place and time.

The Brown Corpus was developed at Brown University and gives a snapshot of US English in the early 1960s. Similarly, the LOB Corpus (Lancaster-Oslo-Bergen) only includes language from the UK during the early 1960s. These can be used to compare UK and US English at this same point in time and can also be used to compare similar genres of text. 

In each corpus there were 15 genres and totalled around 1,000,000 words in size each. 

I'm sure this data would be very useful in semantic research by finding out how the meanings of words have changed over time. It's good to know that these are available for future research.

[bethlearnscorpuslinguistics-blog]

Corpora and Language Teaching

Dictionary Creation

Tony began this lecture by talking about the importance of corpora in dictionary-creation. Apparently today, it is almost impossible to find a dictionary that has not been created using a corpus. I checked this out in a Collins Gem dictionary that I was handy to me and, sure enough, it thanks the Bank of English in its foreword. 

"This dictionary has been compiled by referring to the Bank of English, a unique database with examples of over 450 million words enabling Collins lexicographers to analyze how English is actually used today and how it is changing. This is the evidence on which the material in this dictionary is based". (Collins Gem English Dictionary, 2003) 

I had been aware of the use of corpus linguistics by lexicographers but had never paid enough attention to a dictionary to notice the mention of a particular corpus.

Limitations of Corpus Linguistics

Corpus linguistics can only provide the raw data, it cannot provide explanations for why something happens. Many more factors need to be considered when discussing a finding in a corpus.

For example, if women use the word "wash" more then men, what societal factors may have led to this higher usage? If spoken language is being analyzed, we cannot look at the context in which these conversations were set (although multi-modal corpora are now becoming available which include pictures and additional visual information). 

Course Readings

McEnery, Tony & Wilson, Andrew (2001) Corpus Linguistics. Edinburgh: Edinburgh University Press. Chapter 1.

Hunston, S. (2002) Corpora in Applied Linguistics. Cambridge: Cambridge University Press. Chapter 1.

[bethlearnscorpuslinguistics-blog]

Parallel Corpora

I received a reply from Tony McEnery about the use of parallel corpora. I wasn't sure why people would spend a long time creating a corpus that is the exact translation of another in a different language. This is what Tony said: 

Hi Beth, well parallel corpora can be used for both of the things you mention. However, they are also really important for machine translation systems. If you have used something like Google Translate, behind the scenes somewhere corpora like this are being used. Parallel corpora tell machines a lot about the process of translation - and sometimes they give them pre-translated examples that the computer can simply pick up and pass on to a user.

This helped me realise the importance of corpus linguistics. As Tony stated in one of his first lectures, corpus linguistics is not a theory but a methodology. This can be seen in its use throughout different areas of linguistics.  As mentioned in a previous post, the Bank of English is updated with new texts on a daily basis and this helps sociolinguists keep track of language trends and emerging patterns. Likewise, the Helsinki Corpus aids historical linguists with diachronic language analysis of a particular time period. In a similar sense, machine translations are benefiting from corpora like the CRATER corpus where algorithms can choose from pre-existing translations to translate at a higher standard.

[bethlearnscorpuslinguistics-blog]

What is Corpus Linguistics?

Corpus Linguistics is the analysis of collections of language, like that found in books, newspapers, or other 'real world' contexts. It also includes spoken language which has been meticulously recorded and transcribed by a number of people. 

There are a number of corpora (the plural of corpus) available to use by linguists and other researchers. 

Examples of corpora are:

British National Corpus (BNC):

This contains an astonishing 87,284,364 written words and 10,341,729 spoken words, spanning categories such as "informative writing" e.g. world affairs, arts, belief and thought, social/natural sciences etc., "imaginative writing" (fiction), "spoken context governed"  (speech recorded at public talks), and "spoken demographic" (informal spoken speech). 

Multilingual Corpus

These may be collections of English and Spanish words, or American English and Indian English. I imagine these may possibly be good for metaphor analysis. I'll have to find out more though.

Parallel Corpus

These are corpora where exactly the same texts have been translated, for example, the CRATER corpus. I think these must be for translations and language-learning, but I've asked if there are additional usages of this type of corpus in the course forum.

Learner Corpus

This is language use created by people learning a particular language. An example of these is the International Corpus of Learner English. I guess this could be interesting if you want to find patterns in the way people learn a language. 

Historical or Diachronic Corpus 

This type is probably the most interesting to me. These corpora contain texts from a particular time period. The Helsinki Corpus, for example, contains 1.5 million words of texts from 700AD to 1700AD. 

Monitor Corpus

This is a very interesting type of corpus that is updated on a very regular basis. By doing so, we can see how language changes on a daily basis and can find the origins of new words in a language. An example of this is the Bank of English which is maintained by Birmingham University. Text is added to this as regularly as on a daily basis.

[bethlearnscorpuslinguistics-blog]

Introduction

I've studied linguistics for 6 years (undergraduate and master's degree) but I've never taken a course in Corpus Linguistics.

Last year I used AntConc - a corpus linguistics tool developed by Laurence Anthony, a facilitator on this course - to find uses of 'like' in my conversational data for my master's dissertation. The topic of the dissertation was 'the use of 'like' in Modern British English as a marker of epistemic modality'. AntConc was really useful when I wanted to find instances of the word across a large number of conversations but, even then, I knew I wasn't using the software to its full potential. 

I'm currently studying Software Development and I know that by learning Corpus Linguistics I should be able to develop my interests in this area of linguistics.

By the end of this course, I'm hoping to have gained the skills to research text from Northern Irish newspapers over the last 50 years to see how the descriptions of the UK and Ireland have changed in this time.