![]()
Chapter 1
What is corpus linguistics?
1.1 Features of corpus analysis
Corpus linguistics is concerned with understanding how people use language in various contexts and incorporates computational tools to identify recurring patterns in ânaturalâ or authentic language use. However, as McEnery and Hardie (2012: 1) observe, âcorpus linguistics is not a monolithic, consensually agreed set of methods and procedures for the exploration of languageâ and what they propose to call âcorpus methods in linguisticsâ offers a critical resource that can be implemented across and beyond the field of linguistics.
Corpus linguistics is founded on a frequency-based view of language that determines significance from the recurring patterns evidenced in observations of ârealâ language data. As Tognini Bonelli (2010: 19) remarks, âThe significant elements in a corpus become the patterns of repetition and patterns of co-selection. In other words, in corpus linguistics it is the frequency of occurrence that takes pride of placeâ. This measure of frequency can be in relation to the occurrence of particular features, or it might be the co-occurrence of one feature with another. Furthermore, this might be a measure of a distributional frequency, i.e. regular occurrence across a number of texts within a corpus. The underlying assumption is that the regularity of a formal pattern (i.e. the regular use of a particular form) reflects some functional difference, that language users are regularly opting for one form of expression over another and that this indicates something of its meaning.
We have looked at the role of the computer in our analysis, not least of all in being able to process much more data than a team of manual researchers would be able to â and considerably faster. It is the capacity to process large amounts of data that helps to identify patterns that may be beyond our intuition and which may escape our reading, even if we were given the time to analyse the same amount of data. As manual readers, we are susceptible to giving disproportionate value to some words over others. For example, politically loaded words like âanti-Semitismâ or âterrorismâ may stand out to us more than an unusually frequent use of functional words, such as âbutâ or âyouâ. Nevertheless, the frequent use of âyouâ might attest to a deliberate form of direct address in a piece of marketing discourse; the regular use of âbutâ might reflect the Âcounterclaims that typify a debate. The computer does not hold the same biases as we do and will count each word âequallyâ (so long as we program it to do so), which allows us to construct our reading on the basis of what is there in the text, though of course we can also conduct searchers for the features we are particularly interested in. Corpus analysis can also support us in validating observations that we have made based on smaller datasets; as Baker and McEnery (2015: 10) assert: âBeing able to draw conclusions based on extremely large samples of data adds validity to claims, even if they confirm what we suspected, while providing a quantitative summary gives substance to what may have been a suspicionâ. The corpus, then, is a resource for evidence-building, that can support our claims but equally, challenge assumptions made about language use that are not reflected in the data.
As much as corpus linguistics foregrounds frequency, corpus linguists emphasise the importance of recognising the context(s) in which language has been produced. Corpus retrieval tools allow the researcher to search for a term and view the co-text: the words or features surrounding a particular term that help to determine its meaning. In the most immediate sense, this is enabled by one of the most recognisable and fundamental forms of corpus analysis: Key Word in Context (KWIC). This presentation format allows the analyst to view each occurrence of a search term â the node word â as a series of concordance lines, providing a span of the original surrounding text. This process is known as concordancing and the presentation of KWIC allows us to read the data both âhorizontallyâ, to recover some of the context in which it is used; but also âverticallyâ, to elicit a sense of the patterns of how the term is used across texts.
Below is a series of concordance lines for the term âviralâ from the British National Corpus (BNC) 2014:
Each line represents the next occurrence of the node word (âviralâ) and these can be sorted in various ways: in the order in which they appear in the text (which is often the default), but also alphabetically by the word that appears, for example, to the immediate left or right of the node word. This helps to identify recurring patterns in how the word is used. For example, we can see in the above 20 concordance lines that the word âviralâ is often preceded by forms of the verb âto goâ (go/gone/went), referring to the process by which social media content becomes widely disseminated in a short space of time. This, of course, is a metaphorical extension of the meaning relating to viruses and how they are transmitted, which we can also see here in reference to âviral tonsillitisâ (line 15). While this is only a small sample from the larger corpus, we can also begin to observe that more frequently, âviralâ is used in relation to social media content than illness. Not only can we quantify this with respect to the entire corpus, but we can also compare this with occurrences from the original BNC, which was collected throughout the 1980s and early-1990s, to see how this has changed over time.
Task 1.1
Go to the BNC search interface available at https://corpus.byu.edu/bnc/ and type âviralâ into the search box. This will show you how many occurrences there are in the BNC and by clicking on VIRAL you will be able to view the concordance lines.
⢠What types of words tend to follow the use of âviralâ? Are they mostly in reference to illness?
⢠Can you find any examples of a figurative use of âviralâ, in the way that social media content has been described as âgoing viralâ in the examples above?
The search interface offered by BYU also indicates what category of text the concordance line has been taken from (e.g. newsp_tabloid for a tabloid newspaper, ac_medicine for an academic paper in the field of medicine, fict_prose for a text of fictional prose).
⢠Do you notice any patterns for how âviralâ is used that relate to the type of text in which it appears?
You will find a commentary on this task at the end of the book.
Corpus linguists will also often document other important aspects of the circumstances in which the text has been produced to inform their reading. For example, they might document the identity of the speaker/author (e.g. in terms of age, gender, class); the mode of communication (spoken conversation, academic Âjournal, social media post); or, for example, the purpose and topic of the communication. This type of metadata can help us to explore questions in relation to who is using the term and in what contexts. In the concordance lines presented above each line is labelled with an alphanumeric code, which will refer to the text from which the example has been taken. This provides some initial indication of the extent to which the node word occurs across a range of texts. Concordance lines present us with a lot of condensed information and are therefore a useful way of identifying patterns of language use that we can examine in closer detail, drawing more from the context of the original text.
In some of the instances from the concordance lines provided above it may still be unclear what is being communicated. Most concordance tools will allow the user to extend the span of the co-text and even directly link to the full text. Even then, the analyst may still have questions about the circumstances in which the text has been produced, for what purposes and for whom. One of the concerns that has been expressed in relation to corpus linguistics is that in favouring an exploration of broader patterns, some of the nuance of less pervasive uses could be overlooked. As Tognini Bonelli (2010: 19â20) observes, âThe type of information gathered from a corpus is evaluated as meaningful in so far as it can be generalised to the language as a whole, but with no direct connection with a specific instanceâ. Egbert and Schnur (2018) use the analogy of âmissing the trees for the forestâ to convey one of the potential drawbacks of relying solely on corpus linguistic methods, which is based on an understanding of the difference between the âcorpusâ and the âtext(s)â. They explain: âIn order for discourse analysts to learn about language use within a discourse community, their focus must be on the actual texts created by individual members of that discourse community, rather than on a contrived data setâ (Egbert and Schnur, 2018: 159).
It is certainly worth remembering that the producer of the text will have an awareness for the internal structure of what they have generated but little or no awareness of the other texts that researchers include in the sampling design of their corpus. Texts are self-contained and function as a cohesive unit, which is significant in terms of understanding how certain information is foregrounded; we would likely want to distinguish an occurrence of a language feature that appears in a headline from another singular instance that appears in a paragraph further down. Partington (2008) reminds us to be cautious about the ârecontextualisationâ of corpus data and encourages researchers to familiarise themselves as best they can with the texts that make up a corpus in order to get some sense of their original context. This is not always possible of course, particularly if the researcher is using a corpus they have not themselves collected. We may be relying on minimal (or no) metadata, which is also a reminder of the significance of the value of the appropriate markup and annotation of corpora. These terms, which refer to how information can be added to a corpus, are explored in the next chapter.
1.2 Corpus linguistics in combination
In the research, we find plenty of examples of studies where corpus linguistics is combined with other types of language analysis, with a view to capitalising on the strengths of the approach(es) and mitigating some of the potential drawbacks. In what follows, I review some of the more common and effective combinations that have emerged in studies in applie...
