Making Sense of the EU (Withdrawal) Bill: Computer-Assisted Analysis of Amendments in Parliament

In the first blog from our Making Sense of Parliaments conference, Matt Williams, Jonathan Jones and Ruth Dixon draw on computer-assisted methods of analysis adapted from biological sciences and natural language processing to show how the text of the EU (Withdrawal) Bill was altered through its parliamentary passage; how government and non-government amendments to the Bill differentially changed the text, and how the Bill’s language compares to other contemporary legislation. 

Our aim is to analyse and explain how Parliament amended the Withdrawal Bill during its legislative passage. This will improve our understanding of how Parliament adapts its standard operating procedures to extraordinary circumstances. This analysis builds on a thriving debate in legislative scholarship which asks – Why is legislation ever amended? Why, more precisely, would a government Bill be amended when it had been subject to public consultation (in this case no less than a nationwide referendum), had been drafted by legal experts, and where the government commanded confidence of a legislative majority? To answer this, we have used cutting-edge bioinformatic and natural language processing computational tools to describe and explain the amendments made. The Act’s more than 2,000 lines were analysed using techniques analogous to those used to study genetic mutations. In addition to which, language in the Act’s more than 10,000 words was analysed by machine-reading technologies, allowing us to place it relative to the 4 million words enacted in all 300 primary laws enacted in the decade 2008-18.

These tools have been used to test hypotheses derived from ‘bounded rationality’ theory. The central argument is that amendments to legislation are conceded by governments when they lack information on optimal policy, and are bounded by limited party or popular support. Amendments are expected to impose constraints on government discretion to execute policy. This can be observed by the assertion of normal constructions of legislative language that can be used to make sense of abnormal contexts. In short, under exceptional circumstances Parliament can be expected to ‘take back control’ of legislative language by imposing norms of legal discourse.

Let us first set out examples of how these tools have been utilised in this research. What follows are two amendments to the Withdrawal Bill introduced during the House of Lords Report Stage. Together, they removed discretionary power and replaced it with more determinate legal language. First is the language removed by amendment:

A Minister of the Crown may by regulations make such provision as the Minister considers appropriate to prevent or remedy any breach, arising from the withdrawal of the United Kingdom from the EU, of the international obligations of the United Kingdom.

All adjectives, adverbs, determiners, conjunctions, indeterminate auxiliary modal verbs (‘may’) and enabling verbs (‘make’) have been italicised. These parts of speech tend to expand the exposure of a text’s subject-verb nexus to unseen contextual data. There are, in other words, probabilistic yardsticks in the unmodified text (‘may…make such provision…as appropriate’). The text could even be described as analogic rather than logical, in that decisions have to be inferred from similar decisions that were also deemed ‘appropriate’. This betrays the bounded rationality of the government and their desire to determine textual meaning from context, rather than allowing text to shape context. In short, the government wants to be pragmatic, rather than rule-bound.

The language inserted by the Lords includes the following:

Without prejudice to any other statutory provision relating to the withdrawal agreement, Her Majesty’s Government may conclude such an agreement only if a draft has been— approved by a resolution of the House of Commons, and subject to the consideration of a motion in the House of Lords.

The amended language is far clearer and therefore more readily able to guide the execution of law, even in uncertain circumstances. In fact, this text was removed altogether by government amendment at ‘ping pong’. But a further expectation is that government amendments also clarified legislative language, in order to secure the acceptability of the Bill to a majority in Parliament. Indeed, given the government’s constraints of time and information, it is expected that they dominated the amendments process, thereby clarifying their own law as it moved through Parliament. It is also expected that indeterminacies of language will not be eliminated through the amendments process, but especially genre-deviant language will be pulled back closer to average usage.


Our specific expectations are therefore as follows. Firstly, Bill complexity will have increased with more content added than taken out. The language of the Bill will have become clearer through the legislative process. Government amendments will outnumber non-government amendments as they have the advantage of procedural and partisan dominance, and will have bargained with MPs on a passable final text.

Hypothesis 1a: Amendments will lengthen the legislation [in words and lines].

Hypothesis 1b: Government amendments will outnumber non-government amendments.

Hypothesis 2a: A majority of amendments will remove textual indeterminacies.

Hypothesis 2b: Non-government amendments will remove more textual indeterminacy than government amendments.


Figure 1 displays the results from quantitative analysis of the Bill. Our analysis revealed that the Bill was significantly more amended than the average. Over 55% of the Bill’s lines were altered, with its length increased by 63% (supporting Hypothesis 1a). Previous research into legislative amendments from 2007-2015 found that a typical bill sees a third of its lines altered, culminating in a 40 per cent increase in length (Dixon and Jones 2018).

In Figure 1, amendments are shown by bars whose density indicate the number of amended lines and colour reflects government or non-government origin. Most amendments (both government and non-government) were made at Lords’ Report Stage. The figure shows that the majority of amendments were ‘government’ amendments – that is, proposed in the name of a minister (supporting H1b). Some non-government amendments (i.e. those appearing in the Marshalled Lists under non-ministerial names), particularly in the Commons Committee stage, were accepted by the government and agreed without a vote. Others were opposed by the government and only agreed after a government defeat. Amendments during ping-pong are shown in grey-blue reflecting a combination of government and non-government demands. Many non-government amendments were reversed during ping-pong, though in several cases they resulted in further changes to the bill as ‘amendments in lieu’.

W,J,D fig 1

Figure 1

Turning to the textual analysis, Figures 2-6 describe how key textual elements changed from the Bill as first introduced to the final Act. These textual elements are adjectives and adverbs, conjunctions, indeterminate modal verbs (‘may’, ‘might’), determinate modal verbs (‘shall’, ‘must’) and enabling verbs (‘make’, ‘amend’). The figures show transformations in the Bill compared to linguistic norms of all 300 primary Acts of Parliament enacted in the preceding decade. Standard deviations from the mean are displayed with z-scores (z=±1 covers ~68% of observations, z=±2 covers ~95%, and z=±3 covers ~99.7%). Figures 2 and 3 therefore show that with regards to its use of adjectives, adverbs and conjunctions, the Bill was highly unusual, having a greater proportion of these ‘indeterminate’ parts of speech than 95% of contemporary laws. But the Bill was far less unusual as regards its use of verbs (Figures 4-6). And the final Act contained language that was closer to the mean in all five parts of speech (supporting H2a), with the exception of indeterminate modal verbs (such as ‘may’) which were marginally less utilised in the Act than the average law from the past decade.

W,J,D fig 2

Fig 2

W,J,D legend

W,J,D fig 3

Fig 3

W,J,D fig 4

Fig 4

W,J,D fig 5

Fig 5

W,J,D fig 6

Fig 6

Figure 7 shows how the four indeterminate parts of speech (adjectives/adverbs, conjunctions, indeterminate modal verbs and enabling verbs) changed in aggregate at each stage of the legislative process in which significant numbers of government and non-government amendments were made. The patterns demonstrate that non-government amendments added a reduced quantity of indeterminate parts of speech as compared to language removed by those amendments and as compared to government amendments. These non-government amendments thereby worked towards textual clarification of the Bill (supporting H2b). Both government and non-government amendments, moreover, were more determinate than the language of the Bill as introduced. Overall, therefore, both government and non-government amendments increased the textual determinacy of the Act.

W,J,D fig 7

Fig 7


Taking back control in a democracy is at root at discursive exercise. It is control of rhetorical and legal discourses that matters in securing policy victory. If, that is, physical violence is to be avoided. In short, shared language is the basis of legitimate social power in a democratic state. But the revelation of policy in plain language can weaken the bargaining position of a government, especially one confronted by decision-making boundaries. Sustained informational asymmetry can be an effective executive strategy, and this has implications for the very concept of legislation in post-Brexit Britain.

Nonetheless, in the case of the EU Withdrawal Bill, the government’s bounded rationality encouraged MPs to push for the reassertion of linguistic norms and thereby significantly clarified and enlarged the final Act. When it comes to estimating this new law’s operability, we can at least say that in procedural terms it is now easier to predict how the law will be implemented. There are still significant question marks over the substance of future policy, but Parliament has secured a far greater role for itself by diminishing executive discretion. Parliament has taken back some control.


Dr Matt Williams is Access and Career Development Fellow at Jesus College, University of Oxford. Find out more about Matt’s research on his website or follow him on Twitter: @magl_williams 

Professor Jonathan Jones is a Fellow of Brasenose College and Associate Professor in Physics. Follow him on Twitter: @nmrqip

Dr Ruth Dixon is a Research Fellow at the Blavatnik School of Government, University of Oxford. Follow her on Twitter: @ruth_dixon 

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s