Paul Bayley
A corpus account of “accountability”
This research was funded by a grant from the INTUNE project (Integrated and United: A quest for Citizenship in an ever closer Europe) financed by the Sixth Framework Programme of the European Union, Priority 7, Citizens and Governance in a Knowledge Based Society (CIT3-CT-2005-513421)

1. Introduction

This brief paper intends to provide an example of how a lexeme can be analysed through the tools of corpus linguistics. Its purpose is to illustrate the potential of these tools in analysing the ways in which key social and political concepts are construed in naturally occurring texts. Its addressee is primarily composed of the political scientists in the IntUne project[1] and it is one of a series of papers produced by the Media Working Group aimed at highlighting some aspects of lexis that may be of interest to the Issue Areas in the project; this paper addresses in particular the “Scope of Governance” Issue Area.

2. Some comments on corpus linguistics

Partington’s IntUne working paper (2006) has already furnished a outline of the aims and purposes of corpus linguistics, the construction and editing of a corpora, the analytical instruments offered by the software, the kind of research question that can be addressed through the use of general and specific corpora, and finally the ways in which corpus linguistics can be combined with discourse analysis in order to produce both quantitative and qualitative results. I shall take the issues raised in the paper as “given”, although I would like to highlight two aspects concerning the notion of representativity and the question of corpus design.

Firstly, a corpus has been defined as “finite-sized body of machine-readable texts, sampled in order to be maximally representative of the language variety under consideration” (McEnery and Wilson, 1996: 24); the emphasis in this citation is mine and its purpose is to make the point that no strong claim to represent the whole of a language can feasibly be made. The “population” of a language is so vast, dynamic, and constantly expanding as to make “representation” an extremely problematic question. At best, corpora can “represent” a number of temporally and spatially situated varieties of language. Corpora can be “general” (heterogeneric), or domain-specific (monogeneric); in this paper I will draw on one of each – the British National Corpus (henceforth BNC), composed of 100 million running words, or “tokens” (90 million tokens of written English and 10 million spoken) which may be said to “maximally” represent contemporary standard (and mostly written) British English with particular reference to the 1990s, and a domain specific corpus, composed of the transcriptions of all the sittings of the House of Commons held in 2003 composed of approximately 10 million words, which may be said to “maximally” represent the language of a particular discourse community in a particular legislature. However, corpora have offered the linguist an access to a quantity of data previously unthinkable.

The corpus linguist is interested in the patterns that occur around particular lexical items – what other words they collocate with; what typical grammatical configurations they appear in; what kind of attitudinal value they take on in the light of their co-text; how they are distributed throughout the texts that make up the corpus, and so on. In short, the corpus linguist studies patterns of wording that emerge from large bodies of naturally occurring texts. Why should this be of interest to the political scientist? I would argue that patterns of wording are also and inevitably patterns of meaning, and when given patterns of meaning frequently recur over a wide number of texts, it may be presumed that this has some kind of social and/or political significance. In other words, studies of patterns found in a corpus allow us to make a “survey” of how meanings are construed across a particular body of texts. Working with multilingual corpora, moreover, may give us some insight into how different political cultures lexicalise political and social issues. For example, the word I shall be focusing on in this paper, accountability, is not readily translatable into Italian and other languages involved in the IntUne project, at least not as a single abstract noun.[4]

Corpus studies normally begin with the analysis of a particular lexeme or a particular combination of lexemes, an analysis that is subsequently extended on the basis of the findings that emerge from the empirical findings. The selection of the lexeme used as a point of departure depends on the nature of the task that is being tackled. As I said in my introductory paragraph, my task in this paper is to provide the “Scope of Governance” Issue Area with some kind of warrant for the use of corpus studies, to provide evidence that corpus data can give us insights into discourse patterns. Recurrent patterns of language provide “pre-fabricated means by which ideas can be easily conveyed and grasped” (Stubbs 1996: 158). Given that one of the principal aims of this group is to investigate perceptions regarding which policy areas should be dealt with at a local, national, or supranational level, my own intuition was that the notion of accountability, defined by Heywood as “an important feature of limited government, effective policy making and democracy” which “limits government power by establishing mechanisms of political control through which one institution oversees the working and the performance of another” (Heywood, 2000: 117), could provide some interesting data; it is one of the principles through which different levels of decision-making are decided and controlled. However, it should be said that experience in corpus work has shown that initial intuitions are not necessarily the right ones, firstly because corpora constantly reveal data that are not available through individual introspection, secondly because “notions” do not always emerge from a corpus from the name that is given to them. In general, corpus analysis often involves a great deal of trial and error. However, I believe that from an empirical level, the search for the abstract noun accountability and the adjective accountable has provided us with a good deal of information about how accountability is represented, who should be held accountable and to whom, and for what aspect of public life.

3. Frequency data

A search for a word or a lexeme in a corpus immediately produces data on the frequency of the item. Although frequency is not necessarily of great interest in and of itself, it does often provide a useful analytical point of departure. However, frequency data may hide some tricky problems of interpretation, not least because of the polysemic nature of a great deal of words. For example, a search for unaccountable in the BNC yields 75 occurrences, over half of which are of the kind unaccountable reason, which realises a meaning quite different to that given by Heywood, which instead is found in strings like unaccountable central bankers. Accountability, on the other hand, exclusively occurs in the meaning of political, social or economic “answerability”. This is not merely the core meaning of the word, but its only meaning in the BNC.

Using the BNC, a search for accountab yielded 1,209 instances of the abstract noun accountability and 659 instances of the adjective accountable. As might have been expected, this frequency, which is normally expressed in terms of “relative frequency” (0.012 per 1,000 tokens for accountability) does not place it among the most frequent lexemes in the corpus. In fact accountability ranks as the 6,244th most frequent word in the BNC. Moreover, it is not distributed evenly across the corpus; it occurs in only 257 out of a total of 4,125 texts. It is thus a rather specialized lexical item. In contrast, and not unexpectedly, it has a higher frequency in HoC2003; accountability occurs 601 times (relative frequency 0.060 per 1,000 tokens, frequency ranking 1,690th), while there are 426 instances of accountable. Moreover, the occurrences of the lexeme are distributed quite evenly across the corpus, with occurrences in 137 of the 152 files (each file representing a daily sitting).

However, some caution over numerical data is required. Corpora are sensitive to the period of time over which they were collected and compiled, and by the criteria with which the texts were selected; the smaller the corpus, the more acute this problem becomes. Let us take the example of an abstract noun which may be seen as another word in the semantic area of “governance”, and which might intuitively be supposed to have a similar frequency in general English to accountability – sovereignty. In the BNC, its absolute frequency is 1,203, only six instances fewer than accountability and so the frequency of the two words matches very well. However, its frequency in HoC2003 is 0.0018 per 1,000 tokens, much lower than the 0.060 for accountability. In a relatively small corpus, frequency data may depend on the specific social (or political) contingencies that were in play at the time of collection, particularly if the corpus is focused on current affairs; if one issue dominates during a data collection period, corpus data may be skewed. It seems that the notion of “sovereignty” was not an issue in the House of Commons in 2003, despite the very serious questions of international relations that it was called upon to discuss. To make a further telling contrast, in a corpus of House of Commons debates on European integration held between 1992-1999, the relative frequency of sovereignty was 0.191 (for the semantic patterns around sovereignty in this corpus, see Miller, 1999).

4. Collocation data

Having summarily looked at the question of counting, what really concerns the corpus linguist is not the frequency of a term but the linguistic environment in which it occurs. A first approach to this can be made by looking at the collocates of the search word, typically those words which are co-selected with the search word, or node, within a window of five words to its left and right (although this parameter may be changed according to specific requirements). In this paper, I shall focus on the adjective form accountable. In the BNC, its most frequent collocates (the numbers in brackets refer to absolute frequencies), excluding the highly frequent functional words such as to, for, and be, are as follows:

A similar list from HoC2003 (with the exclusion of most functional words) is as follows:

These lists of collocates are indeed very similar, and I shall make some comments on them before illustrating how they actually appear in the form of concordance lists. Firstly, in HoC2003 there are two collocates which do not figure in the BNC, namely house and state. This can be explained by the register-specific characteristics of British parliamentary discourse. House is a very frequent lexical item in the discourse of the House of Commons and is the term that MPs refer to their own chamber (while the House of Lords is traditionally referred to as “the other place”). It can thus be subsumed under (or considered as a hyperonym of) parliament, which is a significant collocate of accountable in both corpora. An examination of the word state in the concordance lines reveals that all instances occur in the cluster Secretary of State, and can thus be subsumed under government, present in both corpora. The last items in these two lists, based on a line which I drew rather arbitrarily, is occupied by two rather different but at the same time similar words – police in HoC2003 and schools in the BNC. Their similarity derives from the fact that closer examination of the data reveal that there are three major public institutions, beyond local authorities, parliament and government, that regularly collocate with accountable – the police force, education authorities and health authorities. Within the corpora, education and health, in particular, need to be traced through a wider range of lexemes which, as a result of an apparent preference for concrete nouns over abstract ones, represent parts of the whole; for example schools, teachers, pupils for education and hospitals, doctors, nurses and patients for health. This illustrates the fact that when looking at collocates, or co-occurrences, it is not always to sufficient to look for single lexemes via simple computational procedures, but rather for a lexico-semantic sets. This list of collocates initially produced by the software is useful but may be considered as a point of departure rather than arrival.

The data concerning collocations are displayed in concordance lines, as can be seen in the tables below. I have organised the concordances to display co-occurrences of at least two lexemes. I shall begin with the co-selection of accountable with two words: local and government. Table 1 provides us with important but fairly straightforward and transparent information about what is being held accountable – largely services such as education and health – and the persons to whom local authorities are accountable – citizens and voters. What might be of more interest is that the 42 instances of the co-occurrence of local and accountable produce only 30 concordance lines. This is because the word local tends to cluster around accountable and it occurs twice in the same concordance line in eight instances, and three times in a further three instances. Lexical repetition of this kind tends to place an emphasis and an attitudinal orientation on the repeated item; in other words, “localness” is highlighted and in general, at least apparently, expresses positive evaluation.

However, in order to say something about the positive or negative stance that a word may instantiate, the information given in a short concordance line is generally inadequate. It is necessary to have information on the context in which the utterance occurs, such as who is speaking to whom and through what channel. It is also necessary to have access to larger stretches of text. Let’s take the first line in Table 1: local councillors, drawn from and accountable to the local electorate, to determine local priorities. Looking at the insistent repetition of local, one might imagine that this was an extract from a political speech lauding the virtues of local action. However, it is in fact drawn from a textbook on public administration (Greenwood and Wilson 1989), and, as can be seen from the extended version below, the authors certainly give a positive spin to “flexibility” (…government needs to be flexible) but they do not fully endorse the proposition that follows because they uses the “distancing” adverbial allegedly. Any evaluation of the adjective accountable that can be found in the example is not attributable to the authors but to an un-named source.

Because different communities have different needs, the system of government needs to be flexible; local government allegedly provides this by enabling local councillors, drawn from and accountable to the local electorate, to determine local priorities.

Consequently, it is necessary to be circumspect in drawing conclusions about the meanings realised in decontextualised 10 word strings, and this is why the Media working group is particularly interested in integrating corpus analysis with discourse analysis.

This having been said, however, there is enough corpus evidence to suggest that “local accountability” is construed positively. This can be seen, for example, through chains of positively-oriented adjectives collocating with accountable; words that are apparently neutral or “institutional” may acquire positive or negative attitudinal value from the words that typically collocate with them. For example, in a study of British parliamentary discourse in the 1990s, it was shown that the word federalism was consistently construed negatively as a result of its collocational profile (Bayley, 1999). In Table 1, examples of this can be found in the nominal group – a more effective, efficient or accountable system of local government or the verb expressing “desirability” in the clause: …we want hospitals that are accountable to local people.

Table 1. accountable - local

Table 2 shows co-occurrences of accountable and government (instances of local government have been deleted). It can be observed that most examples posit the government as accountable to either parliament or to the electors but unlike Table 1, there is very little information about which areas of public policy governments should be accountable for. Discourse about “government accountability” seems to be embedded in the context of more abstract discourse. There are traces of positive evaluation, which emerge largely from the concatenation of other adjectives (e.g. more open and accountable government, accountable and genuinely representative).

Table 2. accountable - government

The first two tables have focused on the co-selection of nouns and adjectives. More complex grammatical and semantic relations can be found by examining the co-selection of the adjective with verbs. The verbs which collocate most frequently with accountable in both corpora are make and hold and taking into consideration the all forms of these verbs, they are co-selected with accountable in 149 of the 659 occurrences of the node in the BNC. This marked tendency to co-selection could be explained by the fact that both verbs are very frequent in general English. However they are also polysemic verbs, and when they collocate with accountable they take on special meanings. In the collocation patterns found in the corpora, with very few exceptions, make is a causative verb, x makes y happen, while hold may be classified as a verb expressing, or entailing, “authority”,[5] x holds y to do p. From a semantic point of view, these collocation patterns are interesting. “Accountability” is construed as something that does not come about naturally; institutions must be made accountable, and some authority (most examples are realized in the passive voice and so the authority is unspecified or implicit) must hold them accountable. They are also interesting because they may, although not always, have a similar grammatical configuration with two complements, one representing the person or thing which is held or made accountable, and the other the eventual beneficiary: for example, from the BNC:

…and thus to make local authorities more accountable to their voters for their spending.

In a liberal democracy government is held accountable to citizens by means of regular free elections.

Concordance results are displayed in Table 3, which includes 30 of the 39 instances of the co-selection of make and accountable. It is possible to identify a number of bodies which seem to be typically “made accountable”, including governments, schools, health care providers, the EU commission, private enterprises, as well as individual citizens.

Table 3. make - accountable

Similar features can be observed with concordances of hold/held and accountable. But the following table illustrates another semantic trait. These items are frequently co-selected with modal operators expressing the meaning of necessity, and therefore, in most instances, of desirability. In fact must, should and ought to collocate with accountable in 73 of the 659 instances. The following table sets out examples of the co-occurrences of expressions of modality together with held and accountable:

Table 4. should - held - accountable

That “accountability” is construed as something that is desirable is also illustrated by the frequent collocation of more and accountable. Of course it would be nonsense to suggest that the combination of more and an adjective necessarily means that the adjective’s quality is desirable.[6] One needs to look at the value orientation of the adjective itself as well as the co-text in which it occurs. However, strings such as less powerful and more accountable, more open and accountable, more democratic or at least more accountable, more efficient and accountable, or the police should be more accountable to the public, show, by association, that accountable is to be interpreted in a positive fashion. The following is a selection of 30 of the 82 occurrences of more and accountable.

Table 5. should/ought to/must - held - accountable

Finally, another feature of the word patterns that emerge from this study of accountable, and one that can be seen in the tables above by looking at the word immediately to the left of the node, is that the adjective is frequently (in 100 instances out of 658) qualified by an adverb. The qualifications are in terms of grading:

totally, fully, partly, vaguely, highly;

of efficacy or veracity:

sufficiently, genuinely, properly, finally, ultimately, truly, clearly, directly, effectively, readily;

and of typology:

ministerially, democratically, publicly, managerially, financially, locally, politically, professionally, socially, environmentally, legally.

This would, then suggest that accountability is not a unitary notion; it can be full or partial, it can be genuine and effective, and it can divided into different types.

Conclusions

In conclusion, this account of accountability has highlighted a number of lexical semantic and grammatical features that have emerged from a corpus analysis of the word accountable based on the BNC, and, in less detail, from a corpus of parliamentary debates. It can be observed that:

• the company that the word keeps is very similar in both the general and the specialised corpora.
• the corpus data is able to identify a number of institutions that seem to be frequently called upon to be accountable.
• accountability seems to generally carry positive attitudinal value (more of it is better), which can be discerned through verbs or modal operators that express desirability.

References

Bayley P. 1999. “Lexis in British Parliamentary Debate: Collocation Patterns”. In J. Verschueren (ed.) Language and Ideology: Selected Papers from the 6th International Pragmatics Conference. Antwerp: International Pragmatics Association. 43-55.

Greenwood J. and Wilson D. 1989. Public administration in Britain today. London: Unwin.

Heywood A. 2000. Key Concepts in Politics. Basingstoke: Palgrave.

Kirkpatrick B. (ed.) 1998. Roget’s Thesaurus of English Words and Phrases. Harmondsworth: Penguin.

McEnery T. and Wilson A. 1996. Corpus Linguistics. Edinburgh: Edinburgh University Press.

Miller D. R. 1999. “Meaning up for grabs: Value-orientation patterns in British Parliamentary debate on Europe”. In J. Verschueren (ed.) Language and Ideology: Selected Papers from the 6th International Pragmatics Conference. Antwerp: International Pragmatics Association. 386-404.

Partington A. 2006. “Aims, Tools and Practices of Corpus Linguistics”. Intune Papers, No. ME-06-05.

Stubbs M. 1996. Text and Corpus: Computer-assisted Studies of Language and Culture. Oxford: Blackwell.