A corpus-based deconstructive strategy for critically engaging with arguments

2.2.1.Cohesion and semantic domains

As I indicated in the introduction, cohesive structure is a key focus for electronic deconstructive analysis of a persuasion text, so I shall need to establish this for Hitchens’ text.

‘Cohesion’ is a linguistic term used to refer to how a text holds together semantically through lexical features as well as through grammatical features (Halliday and Hasan 1976).66 For example, in the following short text, each sentence is linked by the lexical words ‘decided’ and ‘decision’ as well as by the grammatical words ‘him’ and ‘he’ (referring back to ‘Jack’) and ‘she’ (referring back to ‘Jill’):

While identifying cohesive patterns across a text is, ultimately, a qualitative procedure, I will show how quantitative data about the text can significantly reduce selection bias in this identification. In relation to Hitchens’ text, my qualitative identification of global cohesive patterns will be assisted with quantitative data generated using the corpus linguistic software, WMatrix (Rayson 2009). There are a number of different forms of corpus linguistic software. What makes WMatrix different from other corpus linguistic software tools is that it can analyse a text's ‘semantic domains’ – sets of different words of related meaning. WMatrix does this by grouping semantically similar words together using ‘a semantic tagger’.77 This is a software program which brackets words into general semantic domains on the basis of a pre-configured lexicon. So, for example, with the semantic tagger that WMatrix uses, the words, ‘war’, ‘weapons’, ‘military’ are tagged with the general semantic domain ‘WARFARE, DEFENCE AND THE ARMY’. Semantically similar words provide cohesion across a text. That WMatrix groups semantically similar words in this way can significantly reduce labour and selection bias in identifying cohesive patterns across a text.

2.2.2.Semantic domains and statistical significance

WMatrix also has access to what is known as reference corpora.88 A reference corpus is a large electronic collection of texts from the same language which has been sampled across a variety of genres. Because of this, a reference corpus is seen as a representative snapshot of a language, a norm of language use. A reference corpus, just like a single text, can be tagged for its semantic domains. Using WMatrix to compare semantic domains in a text with the semantic domains of a reference corpus illuminates the statistical significance of the former. If a semantic domain in a persuasion text is statistically significant, this can help the analyst see the furthest reaching cohesive patterns in a relatively long text. This is useful since it is the software which is doing a good deal of the work here, not the fallible and subjective human being.

The largest written reference corpus in WMatrix has one million words.99 It has been tagged for semantic information. The statistical metric used in WMatrix is log likelihood (see Dunning 1993). In WMatrix, a log likelihood value of ≥7 is statistically significant (p<0.01)1010; so if a semantic domain in a text is found to have a log likelihood value of ≥7, then this semantic domain is statistically significant. Lastly, the size of the log likelihood value ≥7 is proportional to the statistical significance of the linguistic item.

For the purposes of trying to identify cohesive patterns in a relatively long text, not all statistically significant semantic domains which WMatrix generates are worth examining. This is because the number of words subsumed under even a statistically significant semantic domain may be only 1 and so this word is unlikely to be contributing significantly to global cohesive structure. Conversely, where log likelihood is both in excess of 7 and word frequency is more than 1, then this could be worth exploring and especially so if word frequency is far in excess of 1.

Table 1 contains all the semantic domains in Hitchens’ text which have statistical significance, i.e. over a log likelihood value of 7, and where the frequency for words subsumed under the semantic domains is greater than 1. Clearly, words subsumed, for example, under the semantic domain DECIDED (log likelihood, 36.5; frequency 12) will be worth investigating for their contribution to the cohesion of Hitchens’ text.

2.2.3.Care in use of quantitative data

Care needs to be exercised in basing judgements of cohesion on the basis of tagged data. First, the tagger that WMatrix uses has an accuracy of 92%. Second, being a software programme with a pre-configured lexicon, the semantic tagger has little scope for the kind of discerning semantic judgements that a human mind can make. Human judgements of cohesive patterning may be based on contextual information, something which a tagger is limited in replicating. For example, in the PARTICIPATING semantic domain (see Table 1), the words ‘intervention’ 1, ‘participation’ 9, ‘intervention’ 16 and ‘intervene’ [34] are part of a cohesive chain in Hitchens’ text which relates to US (‘our’) involvement in Iraq, while ‘collaboration’ 24 is not. Conversely, because a semantic tagger cannot judge meanings in a text contextually, it may choose a semantic domain for a word in a text which a human would not – and vice versa.

In a nutshell, the tagging facility of WMatrix is very useful because it helps the analyst comprehensively to track cohesive patterns across relatively large texts and in so doing significantly reduce labour and selection bias. But, it is important to make sure that the software analysis of semantic meaning is corroborated by human judgements of semantic meaning. In Section 3, I bear all this in mind when I identify arguments in Hitchens’ text in relation to their cohesive structure.

3.1.1.Not ‘extracting’ propositions, but leaving cohesion intact

I identify Hitchens’ persuasion text as consisting of three sub-arguments, two of which (and possibly all three) are related via a Supra-argument; I order my identification of these arguments as they appear in Hitchens’ text.

In my reconstruction, I do not ‘extract’ premises from the sentences as is often the case in argumentation studies and critical thinking studies (e.g. Walton 2006; Bowell and Kemp 2010). This is because my focus, in Section 5, will be on exploring whether or not the cohesive structure of arguments in Hitchens’ text is unstable. For this reason, in my identification, I keep the language of the text and thus its cohesive devices almost completely intact, relating it to words from the semantic domains in Table 1 as appropriate. (Where I make alterations to Hitchens’ text in premise identification is, largely, in transforming a clause into subject-predicate form.) In taking account of the language of the text in my reconstruction of the argument, I accord with Scott Jacobs’ normative pragmatics (e.g. Jacobs 2000, 2009).

3.1.2.Sub-argument 1: the 2003 Anglo-American intervention in Iraq was not an exception from US involvement in Iraq and so not a discrete war

The repetition of words subsumed under the statistically significant semantic domain DECIDED is a crucial part of the cohesion of Sub-argument 1 (see Table 1 and bold above). In other words, US involvement in Iraq since 1968 is constructed as a series of decisions. There are also terms from the statistically significant semantic domain WARFARE, DEFENCE AND THE ARMY (see Table 1 and underlining above) such as in ‘never\ldots “lied into war”’ 10) which, not surprisingly, are in cohesion with one another given that Hitchens’ argument involves negating that the US-led coalition intervention of Iraq was a discrete war. (Moreover, there are terms in Sub-argument 1 from the statistically significant semantic domain NEGATIVE (see Table 1 and italics above) though these do not provide textual cohesion).

3.1.3.Sub-argument 2: although incompetent, the Iraq intervention has positives

3.1.4.Sub-argument 3: implication of moral superiority of Iraq intervention

There are two instances of words from the semantic domain DECIDED (bolded above) which facilitate cohesion in Sub-argument 3. There is one instance of a term from WARFARE, DEFENCE AND THE ARMY (italicised above). This instance – ‘war’ – is in the expression ‘war of choice’, which Hitchens places in inverted commas. Hitchens’ qualification of ‘war’ here links back to the qualified use of ‘war’ in Sub-argument 1. (In Section 3.1.5, I discuss use of ‘intervene’ – see last premise above – a word from the PARTICIPATING semantic domain.)

I found Sub-argument 3 harder to reconstruct than the previous two sub-arguments, hence why I use question marks around the conclusion and my recovery of implicit premises. I shall discuss my reconstruction frustrations in Section 3.2.

3.1.5.Supra-argument: links to the sub-arguments

The last substantive sentence in the argument is [34]:1111

In the Supra-argument, Hitchens constructs the 2003 ‘decision to intervene’ [34] as another decision in a series of them taking place in US involvement in Iraq since (no later than) 1968. In Section 3.2, I discuss some difficulties I experienced in identifying the Supra-argument – flagged by the use of ‘?’ above and below.

I attempt an interpretation of the Supra-argument as follows: on my account, it consists of two conclusions – major and minor. Here is my reconstruction of its structure:

Having identified the arguments in relation to their cohesive structure, I could just jump to Sections 4 and 5 and explore whether or not there is instability in the cohesive structure of these arguments; if the cohesive structure deconstructs, the arguments in the persuasion text thus reduce in credibility. In Section 3.2, however, I discuss some problems I experienced with my identification of the logical structure of Hitchens’ arguments which, in turn, would hinder critical assessment of their logical soundness. The reason for this discussion is so the reader will appreciate how, in Section 5, electronic deconstructive analysis of the text circumvents such problems by still potentially facilitating a form of critical engagement with a persuasion text.

3.2.1.Sub-argument 3 and implication of moral superiority of Iraq intervention

3.2.2.Supra-argument

Hitchens says that former President Bush's speech was:

In [32], Hitchens says explicitly that ‘the right thing’ to do in relation to Burma, Darfur and Rwanda ‘cannot be known in advance’. Is, then, Hitchens analogously implying that the ‘right reasons’ [34] for the US intervening in Iraq must include the post hoc ones of Sub-argument 2? It is difficult to tell and thus hard to know whether or not Sub-argument 2 is part of the Supra-argument. Again, comprehensive understanding of the text is impeded.

3.2.3.Discussion

I should emphasise that it is possible to isolate into arguments most of Hitchens’ persuasion text. Sub-argument 1 is identifiable (although one could argue that there is a potential processing stutter created by inclusion of the book title1717); all of Sub-argument 2 is identifiable. Whether or not Sub-argument 3 is deliberately obfuscatory, it is still in my opinion unhelpfully allusive in parts and as a result I am neither sure of my recovery of implicit premises nor my isolation of its conclusion (hence my use of question marks in the text annotation of Section 3.1.4).

As far as the Supra-argument goes, I am fairly confident of most of my reconstruction; this is because Hitchens has provided clear structural links between [34] and Sub-argument 1 and Sub-argument 3. (One could still argue that it would have been more helpful if these links had come much sooner in the text since the reader cannot fully appreciate the overall (Supra-) argument until almost the last sentence.) However, it is ambiguous whether or not i) Sub-argument 2 is part of the Supra-argument; ii) the implied other reasons for the intervention in Iraq (as laid out in President Bush's speech) 13 are also part of the Supra-argument. This is why I employ question marks in the text annotation of Section 3.1.5.

In sum: I cannot know whether or not Hitchens has a strategy of obfuscation in parts of his persuasion text.1818 I can, however, make the reasonable claim that there are areas in his text's arguments which are not so straightforward to identify. In turn, this creates impediments for critical assessment of the logical soundness of Hitchens’ argument. Despite these impediments, in Section 5, I show how the corpus-based strategy of Electronic Deconstruction can still potentially afford critical engagement with Hitchens’ persuasion text. This is because its evaluative basis is the cohesive structure of a persuasion text which, as a set of surface features, is usually much easier to identify than all premises of an argument relevant to its conclusion, particularly where these are implicit premises or it is difficult to ascertain relevance.

In Section 2, I laid out the first part of Electronic Deconstruction – the identification of the cohesive structure of arguments in a persuasion text and how a corpus linguistic software tool can assist this. In Section 3, the second part of Electronic Deconstruction, I identified arguments in relation to their cohesive structure. Section 4, the third part of my outlining of Electronic Deconstruction, lays out the theory and method for deconstructing a persuasion text's cohesive structure. Since Electronic Deconstruction employs the concept of ‘collocation’, an important one in corpus linguistics, this is where I begin.

4.3.1.Absence from normal collocation

It may turn out that there is an absence from normal collocation in use of a word in a persuasion text in relation to its topic (or sub-topic). This absence from normal collocation may just be a case of ellipsis with relevant contextual information supplied elsewhere in the text. Even if it not a case of ellipsis, the absence of normal collocation for the topic (or sub-topic) may not be deliberate. Alternatively, the author may have intentionally avoided normal collocation for a particular topic because otherwise its inclusion would undo the argumentative line they have constructed.2323

4.3.2.Exploring possible tensions in cohesive structure

Electronic Deconstruction explores, via a very large corpus, potential tensions between normal collocation in discussion of the persuasion text's topic and the way in which discussion of the topic in an argument is collocated. If the analyst finds that there is an absence from normal collocation for the topic (or sub-topics), the next part of the procedure is the following: bestowing normal collocation on the text and examining whether or not this affects the consistency of its cohesive structure. A persuasion text's credibility is dependent, amongst other things, on its cohesion. If the cohesive structure is found to be unstable on bestowing normal collocation for that topic (or sub-topic), the credibility of the persuasion text diminishes.

Echoing what I wrote earlier, just because an analyst discovers that there is an absence from normal collocation, this does not mean that this is deliberate. That said, Electronic Deconstruction highlights that not employing normal collocation for discussion of a topic could be a hostage to fortune. If an author has not provided good reasons for avoiding how a topic (or sub-topic) is normally collocated, their persuasion text could be potentially vulnerable to electronic deconstructive analysis once normal collocation is bestowed.

Deconstruction with the most impact will be that which relates to the whole of the persuasion text or at least to significant portions of it. For this reason, the procedure seeks out possible deconstruction which affects the cohesive structure of any Sub-arguments or a Supra-argument in a persuasion text. An analyst may begin on the basis of an intuition that there exists an absence from normal collocation in the persuasion text, or on a more trial and error basis.

4.3.3.Reducing subjectivity and electronic deconstruction

If we do not know much about a persuasion text's topic (or sub-topic), we may not intuit absence from normal collocation and just assume the cohesive structure is stable. Electronic Deconstruction, as I will show, is a useful way of revealing instability in a persuasion text's cohesive structure for the unsuspecting. Even if absence from normal collocation can be intuited because the analyst has knowledge of the topic (or sub-topic), what is to prevent the charge that this is a reflex of the subjectivity of the analyst? The intuition of absence from normal collocation could, in theory, be batted back to the analyst by the author as ‘well, just that's your opinion’. Since Electronic Deconstruction draws on corpus linguistic method and statistical metrics, subjectivity is significantly reduced in judgements of absent meaning and, in turn, in evaluations of instability in cohesive structure.

Drawing on the two-billion word OEC, Section 5 demonstrates Electronic Deconstruction by performing the following: evaluative analysis of the stability of the cohesive structure of Hitchens’ persuasion text which WMatrix helped to illuminate (see Section 2.2). The OEC is made available through Sketchengine (http://www.sketchengine.co.uk/), the software I use to search through this corpus. Since the OEC consists of website material from 2000 onwards, this makes it very useful for investigating collocational norms for discussion of the US-led coalition intervention of Iraq in 2003. Through comparison with these norms, I will show that the cohesive structure of arguments in Hitchens’ text deconstructs. For all OEC investigations, I use the same word span n±5.

5.1.1.Normal collocation for ‘intervention in Iraq’

In Sub-argument 1, Hitchens uses the expression ‘Anglo-American intervention in Iraq’ 1. This cohesively links with ‘intervene’ [34] in the Supra-argument. In the OEC, ‘Anglo-American’ occurs only once as a collocate of ‘intervention in Iraq’ and unsurprisingly with no statistical significance. In contrast, ‘military’ collocates with ‘intervention in Iraq’ 126 times (t-score 11.2). In fact, ‘military’ is the highest lexical collocate of ‘intervention in Iraq’. These collocation results tell us that the intervention in Iraq has been discussed much more in ‘military’ than in ‘Anglo-American’ terms. (Moreover, it is worth flagging that Hitchens refers to ‘intervention’ in 16 and ‘intervene’ in [34] without modification from ‘militar(il)y’ either.)

I should stress that finding out what normal collocation is for ‘intervention in Iraq’ is not the same as finding out an opinion, i.e. how that normal collocation is used in an evaluation. To do that, we would need to go beyond collocation and explore the clauses and sentences in which ‘military’ collocates with ‘intervention in Iraq’. Studying the wider context of instances of this collocation in the OEC reveals that sometimes ‘military intervention in Iraq’ is assented to, sometimes it is disagreed with or sometimes it is described neutrally, e.g. ‘the military intervention in Iraq was justified/a disaster/took place in 2003’. Corpus evidence below for other normal collocation, likewise, does not tell us about the evaluative nature of opinions expressed.

It is interesting to see what happens in Sub-argument 1 if we replace the relatively rare collocate of ‘intervention in Iraq’, ‘Anglo-American’, with the most common lexical collocate, ‘military’:

The collocation absence of ‘military’ may not, of course, be deliberate. However, if it were a deliberate omission and, if Hitchens had responded to this electronic deconstruction, he might have argued that the reason he did not modify ‘intervention in Iraq’ with ‘military’ is he did not want to create the false impression that the intervention was a war. However, the corpus evidence for ‘intervention in Iraq’ creates difficulty for this kind of retort because Hitchens would also have to demonstrate that common discussion of the intervention in Iraq in military terms is misguided whether people agree or not with the intervention. I shall show later that the fact ‘military’ is the most common collocate of ‘intervention in Iraq’ creates further difficulties for cohesion in Hitchens’ persuasion text.

5.1.2.Normal collocation for ‘hostilities’

Hitchens uses ‘hostilities’ 4 in:

There are 5914 instances of ‘hostilities’ in the OEC. The first, second and third most common lexical collocates are ‘end’ (615; t-score 24.7), ‘cessation’ (392; t-score 19.8) and ‘outbreak’ (230; t-score 15.2). ‘Began’ (126; t-score 11.1) is the sixth most common lexical collocate and ‘ended’ (123; t-score 11.1) the seventh. ‘War’ (134; t-score 11.4) is the fifth most common lexical collocate of ‘hostilities’. Conventionally when ‘hostilities’ is used, the beginning or end of a particular war is communicated, such as in:

In sum, by exploring collocation of ‘hostilities’, I am able to create a picture of how one aspect of the topic of ‘war’ is normally discussed: when the word ‘hostilities’ is used, it is normal for places and/or points of time and/or the particular war to be indicated.

Let me now bestow normal collocation for ‘hostilities’ where the term is used in Sub-argument 1 and explore possible deconstruction in cohesive structure. To assist exploration, I reproduce Sub-argument 1 and annotate common collocation around ‘hostilities’ together with making other relevant annotations; I also reproduce [34], a key sentence for the structure of the Supra-argument:

From qualitative exploration of the persuasion text, assisted by the WMatrix analysis of semantic domains in Section 2, we know that:

In sum, while Hitchens constructs US involvement in Iraq from 1968 to 2003 as a continuity of decisions, normal collocation for ‘hostilities’ conflicts with this construction. Unwittingly, for Hitchens, ‘hostilities’ is a ‘stray signifier’ which leads to further instability in the cohesion of Sub-argument 1.

5.2.1.Normal collocation for ‘intervention in Iraq’

In Section 3.2.1, I highlighted how Hitchens uses, as part of Sub-argument 3, an extended medical metaphor such that the US is a quasi-doctor who ‘intervenes’ [34] in a ‘traumatised’ [29] Iraq, the US having had ‘responsibility’ [29] and ‘involvement’ [29,34] for this quasi-patient's ‘life and death struggle’ [34] for many years.

Corpus evidence supports an interpretation that an audience could well infer this medical allusion: the most common lexical collocate of ‘intervention’ is ‘health’ (1395; t-score 36.9); other collocates of ‘intervention’ that can be regarded as falling under a ‘health’ semantic field, and carrying statistical significance, are ‘medical’ (837; t-score 28.7), ‘care’ (748; t-score 26.8), ‘treatment’ (607; t-score 24.3), ‘surgical’ (566; t-score 23.8), ‘patients’ (519; t-score 22.4) and ‘trauma’ (25; t-score 5) itself. I have flagged cohesion for this extended medical metaphor below in bold:

In [34], Hitchens uses the expression ‘we decided to intervene’ where ‘we’ links back to ‘Anglo-American’ 1. But, we know from Section 5.1.1 that ‘military’ is the highest lexical collocate of ‘intervention in Iraq’, that is, this is normal collocation. Modifying ‘intervene’ [34] with ‘militarily’, as I have done above, conflicts with the medical-related cohesive chain. Cohesion in Sub-argument 3 is destabilised with knock-on effect on its credibility.

5.2.2.Normal collocation for hypothetical and actual humanitarian interventions

In Section 3.2.1, I argued that Sub-argument 3 can be seen as implying that the Iraq intervention was humanitarian and thus morally superior to the lack of interventions in Burma, Darfur and Rwanda. Using Sketchengine software, I investigate the extent to which ‘military intervention in’ and ‘humanitarian intervention in’ collocate with any of these places in the OEC. I use these multi-word units for the sake of consistency with Section 5.1.1. There were no results for Burma or Rwanda but there were for Darfur/Sudan: ‘humanitarian intervention in’ + ‘Darfur’ (4; t-score 4); ‘military intervention in’ + ‘Sudan’ (14; t-score 3.7). Moreover, the OEC provides evidence of collocation with actual humanitarian interventions in the Balkans and Somalia. For ‘humanitarian intervention in’, we have: Bosnia (7; t-score 2.6) and Somalia (7; t-score 2.6); and for ‘military intervention in’ we have: Balkans (15; t-score 3.9), Kosovo (21; t-score 4.6) and Somalia (17; t-score 4.1).

The corpus analysis suggests that actual humanitarian interventions, and hypothetical (at the time of writing) humanitarian interventions for Burma, Darfur and Rwanda, are also discussed in military terms. This makes perfect sense since humanitarian interventions, particularly where genocide is involved, will usually require military intervention. (This is supported by the information in the UN weblink in footnote 12.) What this means is that for Hitchens to create credibly, in Sub-argument 3, the seeming implication that the Iraq invasion was some kind of humanitarian intervention, it would be unusual if it were not also associated with a military intervention. However, as already mentioned, collocating ‘intervene’ [34] with ‘militarily’ would conflict with the medical-related cohesive chain. If it really is the case that Hitchens deliberately avoids modification of ‘intervene’ [34] in order to help to create a smokescreen via implication that the US-led coalition invasion of Iraq was a humanitarian intervention, then a deliberate omission would be hoist by its own petard.

5.2.3.Normal collocation for ‘a country collapses’

In this section, I highlight how normal collocation for ‘a country collapses’ also creates tensions in the cohesion of the extended medical metaphor. At first sight, it might appear that ‘a country collapses’ sits well with the extended metaphor of Iraq as patient and US as doctor. Exploring normal collocation in relation to ‘a country collapses’, I show that this is not the case.

There are 87,462 instances of the broadly conceived lemma, COLLAPSE.2424 Using Sketchengine software, I filtered all these instances on the lemma COUNTRY (for an n±5 span) given that Hitchens uses the expression ‘a country collapses’ [29]. This left me with 676 instances of COLLAPSE collocating with COUNTRY in texts in OEC. I then calculated collocation for n±5 span. The most common lexical collocates are ‘economic’ (63; t-score 7.9) and ‘economy’ (28; t-score 5.3); other related collocates are ‘banking’, ‘banks’ and ‘financial’ (see Table 2).

Here is an example from OEC:

Even for other common collocates, such as ‘Soviet Union’, it is common for there to be economic/financial contexts such as in:

Prototypically, if a country is said to collapse, the OEC informs us that this is related to economic reasons. Put another way, the OEC tells us that default understanding of ‘a country collapses’ is likely to treat ‘collapse’ as a dead metaphor, a convenient shorthand description for deterioration in a country's economy. However, this conventional understanding of ‘a country collapses’ is in tension with Hitchens’ use of the expression in constructing US involvement in Iraq as a live medical metaphor.2626 That conventional usage of ‘a country collapses’ does not, in fact, sit well with an extended medical metaphor inadvertently leads to further deconstruction of cohesion in Sub-argument 3, and the conclusion of the Supra-argument. For Hitchens to prevent this deconstruction, he could just be clearer about what he means by ‘a country collapses and succumbs to trauma’ [29]. If the vagueness of this larger expression is, indeed, a deliberate obfuscatory strategy then, on an electronic deconstructive reading, it would be hoist by its own petard.