In the Beginning

As we will make clear in what follows, chance (indicated repetitively in the text by the use of ‘fortunately’), played a big part in the early development of forensic linguistics, both as a recognised area of professional expertise and as an academic discipline.

Forensic linguistics, as a category of expert evidence and as an academic subject, began in the UK by chance – in a corridor of the English Department of the University of Birmingham. Malcolm Coulthard, who taught English linguistics and Discourse Analysis to undergraduate students of English, was approached by a colleague, who was already acting as an expert witness in forensic handwriting analysis, with a challenge – ‘as you teach a class showing students the difference between genuine interaction and the pseudo interaction playwrights put onto the stage, are you able to demonstrate that this supposedly verbatim record of a police interview with a suspect was falsified?’ In other words, could he show that a policeman was acting as dramatist? Coulthard wrote a report and was sitting in court ready to give evidence, but was never called, because his colleague’s evidence of documentary falsification was so compelling that the judge dismissed the case and severely criticised the police officers involved. As a consequence, an elite police unit, the West Midlands Serious Crime Squad, was immediately disbanded, fifty-one officers suspended, six charged, and four later convicted.

Not only forensic handwriting analysis, but also forensic authorship analysis acquired a sudden high profile, especially among those serving prison sentences. Coulthard became involved in a series of cases where men already convicted on the basis of evidence obtained by the West Midlands Serious Crime Squad claimed they had been ‘verballed’ – that is, convicted on the basis of confessions that had been (at least partly) created by the interviewing police officers. This included high-profile cases such as that of the Birmingham Six, who had been convicted of a pair of deadly pub bombings in Birmingham, UK, in 1974, whose convictions were overturned in 1991 following an appeal which drew on Coulthard’s expert evidence.

The University of Birmingham became known for forensic linguistics and again by pure chance a very gifted group of students, the majority of whom had graduated from other universities, enrolled to study for doctorates in the English Department. It was the diaspora of this group of students that later consolidated forensic linguistics as an academic discipline in the UK. Janet Cotterill, Chris Heffer, and Frances Rock established the world’s first Master’s degree in Forensic Linguistics at Cardiff; later, Tim Grant and Krzysztof Kredens (along with Coulthard) established Forensic Linguistics at Aston; Alison Johnson at Leeds and John Olsson at Bangor, while Sue Blackwell and Jess Shapero continued the tradition for a while at Birmingham.

With the initial building blocks in place, moves were made in the early 1990s to professionalise the field in the UK through the establishment of a journal and a professional association dedicated to forensic linguistic research.

The Association

Forensic linguistics as a label is strongly associated with England, having been used first in a legal context in 1967 by the Swedish grammarian Jan Svartvik, who at the time was working at University College London, to characterise his linguistic analysis of a disputed confession to murder by one Timothy Evans. Svartvik showed that parts of the confession had been falsified by the police. Malcolm Coulthard adopted the phrase in the late 80’s when he began to work as an expert witness. It was used in the title of a couple of international seminars at the University of Birmingham, which brought together academics interested in the relationship between language and the law. For these reasons, when the formation of an association was proposed in 1992, ‘Forensic Linguistics’ seemed to Coulthard to be the most appropriate label, indeed not least because it paralleled the name of the pre-existing association of experts in forensic speech science – the International Association for Forensic Phonetics. So, the International Association of Forensic Linguists was born.

Since its inception, UK academics have played a major role in the Association. Four of the fourteen Presidents of the Association have been from the UK – Malcolm Coulthard, Janet Cotterill, Tim Grant, and Isabel Picornell, with a fifth, Nicci MacLeod, to follow in 2027. Of the twenty-six International and Regional IAFLL Conferences, eleven were organised by academics with PhDs from the English universities of Birmingham and Aston. Aston hosted three times: Tim Grant, Nicci MacLeod, and Annina Heini; Birmingham once: Sue Blackwell; Cardiff once: Janet Cotterill. Outside of the UK, it has been hosted in Florianopolis, Brazil, twice: Carmen Caldas-Coulthard and Malcolm Coulthard; Lodz, Poland, once: Krzysztof Kredens; and Porto, three times (four as of 2027): Rui Sousa-Silva.

In the early days, Sue Blackwell set up and still curates the incredibly valuable online Discussion List for Language and the Law <FORENSIC-LINGUISTICS@JISCMAIL.AC.UK> which currently has some 500 subscribers and, as she notes in her entry in the Encyclopaedia of Applied Linguistics, has acquired a reputation for ‘lively and witty exchanges’.

The Journal

As Forensic Phonetics and Forensic Linguistics were new research areas in the early 1990s, there was no dedicated journal for either, so it was decided to found a journal which catered for both associations and the prospective co-editors, Peter French and Malcolm Coulthard, approached Routledge which fortunately was looking to expand its journal portfolio and agreed to publish Forensic Linguistics: The International Journal of Speech Language and the Law, provided that the two associations guaranteed a minimum income by including the price of the journal in their annual subscriptions. To the editors’ surprise, indeed consternation, at the beginning of the third year, Routledge ‘advised’ that it would cease publication at the end of the year because the journal was not selling a sufficient number of copies to libraries and individuals, even though association numbers were ahead of predictions.

There was no possibility that any other publisher would take over the journal, but, fortunately, the Library of the University of Birmingham had just decided, for prestige purposes, to set up a Press which was therefore actively looking for manuscripts to publish and so agreed to become the publisher of the journal. In fact, it soon became clear that ‘Press’ was a misleading name, at least as far as the journal was concerned. The Press simply looked after the money; the whole publication process – receiving, reviewing and editing articles, contracting and paying copy-editors and printers, and employing a secretary and even posting out issues to subscribers – was devolved to the editors. Even so, the relief of having a respectable imprint was short-lived, because after seven years the University decided to close its Press and the journal became an orphan again. Fortunately, the subscriber numbers were now healthy and Equinox agreed to take over publication. Probably because the audience for the published research continued to grow beyond the UK, it was at this time that Forensic Linguistics was dropped from the title of the journal, thus situating it more squarely within the growing international body of Language and Law research. As we were editing this text, we received the following message: ‘We are pleased to announce that the International Journal of Speech, Language and the Law is transitioning from Equinox Publishing Ltd. to University of Toronto Press’. This transition has since been completed.

As in any research area, international connections have been invaluable, but in the context of this volume it is worth recording the UK’s ongoing influence on the journal, which over its whole life, at least up until the time of writing, has always had at least one UK forensic linguist as editor: Malcolm Coulthard, Janet Cotterill, and Frances Rock and now Alison May and Kirsty McDougall. The second journal Language and Law, Linguagem e Direito, followed in the same tradition with Rui Sousa Silva, ‘honorary Brit’ as Managing Editor, helped by Malcolm Coulthard.

The Second Beginning

Although Forensic Linguistics had been introduced as undergraduate and postgraduate modules and as an option for doctoral research at the University of Birmingham, an established academic discipline needs to train the next generation through dedicated postgraduate courses and Birmingham refused to contemplate setting up a Masters degree, raising the possibility that, as in the US at the time, the area would remain a personal interest for individual academics whose teaching priorities were elsewhere.

Fortunately, Janet Cotterill was appointed to a position in the English Department at Cardiff University at a time when the academic administrators were looking to increase the number of Master’s students. Cotterill’s proposal of an MA in Forensic Linguistics was accepted and, in more good fortune, Chris Heffer, Frances Rock, and Michelle Aldridge joined the department shortly afterwards. Suddenly, Cardiff had the largest concentration of academic forensic linguists in the world, and they still have the longest-running Master’s degree, founded in 2001 by Janet Cotterill.

Subsequently, Coulthard moved to Aston and fortunately, shortly afterwards, a new Vice Chancellor was appointed and, as new vice chancellors do, toured departments looking for areas with potential in which to invest. She thought forensic linguistics was one such area and provided funding for a post to which Tim Grant was appointed. Very fortunately, Aston was at the same time also looking to appoint a lecturer in translation studies and Krzysztof Kredens was the successful applicant – suddenly there were two UK universities with a critical mass in forensic linguistics and the discipline has not looked back. It is now at its historically most healthy in terms of permanent teaching and research staff, the number of courses and modules offered and the number of registered students.

Early Applications of Authorship Analysis in the UK Context: ‘Verballing’

Many of the earliest forensic linguistics cases in the UK involved what the accused called ‘verballing’, cases where it was claimed that police had falsified (parts of) supposedly verbatim verbal records of interviews and of dictated confessions. Over time, Coulthard developed a toolkit to evaluate such claims. In the 1960’s, Svartvik had only grammatical tools, but by the 1990’s, linguists could also draw on those of discourse analysis, particularly those developed for the analysis of supra-sentential structuring, that is, the way in which sentences combine into larger units to form coherent texts (Coulthard, Reference Coulthard1975: 73).

The case that first raised the profile of forensic linguistics was the 1998 Appeal of Derek Bentley, a nineteen-year-old accused, convicted and executed for the shooting and killing of a policeman in November 1952. At Bentley’s trial, three police officers swore on oath that Bentley’s statement was the product of unaided monologue dictation, whereas Bentley asserted that it was, in part at least, the product of dialogue, in other words, that police questions and his replies to them had been conflated and recorded as monologue. Supra-sentential evidence in the form of narrative incoherence was discovered to support Bentley’s claimFootnote ¹.

Narratives, and particularly narratives of murder, are essentially accounts of what happened and, to a lesser extent, of what was known or perceived by the narrator and thus reports of what did not happen or what was not known are rare and special. There is, after all, an infinite number of things that did not happen and thus the teller needs to have some special justification for reporting any of them to the listener, for them being newsworthy. A typical example of ‘coherent’ usage in the Bentley confession is:

• A little later Norman Parsley and Frank Fasey called.

• I did not answer the door or speak to them.

When Bentley reported that his friends had called, the listener would reasonably expect him to have at least talked to them and so this is a quite natural denial of a reasonable expectation. (It could still of course have been elicited by the police and probably was.)

However, there are some negatives in Bentley’s statement which have no such narrative justification:

• Chris then climbed up the drainpipe to the roof and I followed.

• Up to then Chris had not said anything.

• We both got out on to the flat roof at the top.

Chris is not reported as beginning to talk once they have got out onto the roof, nor is his silence contrasted with anyone else’s talking, nor is it made significant in any other way later in the narrative. A similarly incoherent example is:

• He caught hold of me and as we walked away Chris fired.

• There was nobody else there at the time.

• The policeman and I then went round a corner by a door.

Similarly, none of the possible inferences from this denial makes narrative sense. So, the most reasonable explanation for the textually incoherent negatives in these two examples is that, at this point in the statement-taking process, a policeman asked a clarificatory question to which the answer was negative and the whole sequence was then recorded as a negative statement.

The fact that some of Bentley’s statement was elicited in this way was crucially important because the denial in this incoherent sequence:

• We were there waiting for about ten minutes.

• I did not know he was going to use the gun.

• A plain clothes man climbed up the drainpipe and on to the roof.

was later singled out by the judge as incriminating. In his reading the statement was a monologue narrative account of the events leading up to the shooting and therefore it showed that Bentley knew his friend had a gun. By this reading, Bentley was lying when he later said ‘I did not know Chris had one [a gun] until he shot’ and that he could be guilty of murder because he had gone to commit a crime knowing they were armed.

However, once it is accepted that there were question/answer sequences underlying Bentley’s statement, it follows that the sequencing of some of the information was not under his direct control. Thus, the placing of the reporting of at least some of the crucial events depended on a decision by the police questioner to ask his questions at those points, rather than on Bentley’s reconstruction of the narrative sequence. And so, crucially, the inference drawn by the judge in his summing up was not justified. If ‘I did not know he was going to use the gun’ was a response to a police question, with its placing determined by the interrogating officer, there is no longer any conflict with Bentley’s later denial: ‘I did not know Chris had one [a gun] until he shot’. Neither is there any significance to be attached to Bentley saying ‘the gun’. Sadly, this analysis, made some forty years after Bentley had been executed, could only support the appeal for a retrospective quashing of the conviction.

A second tool, probably used in a British case for the first time, was corpus analysis. Like Svartvik before him, Coulthard noted several instances of an idiosyncratic grammatical feature – the inversion of subject and adverb in phrases involving ‘then’, for example, ‘I then’, ‘Chris then’, which, he hypothesised, was a feature of police report writing register. Unlike Svartvik, he was able to draw on a corpus of spoken British English, The COBUILD Corpus, at the time in the process of being created at the University of Birmingham. The corpus showed that ‘I then’ occurred only once every 165,000 words in spoken English. An analysis of a specially created corpus of police reports showed that, by contrast, the phrase occurred once every 194 words, a frequency almost 1,000 times greater, indeed there were nine occurrences in one single 980-word police statement, as many as in the entire 1.5 million-word spoken corpus. This was extra evidence to support the claim that the confession was, as Bentley had claimed, a multi-author document.

The Bridgwater Four caseFootnote ², again one of claimed verballing, required the development of another tool. In this case, four men were accused, and subsequently convicted, solely on the basis of the confession of one of them, Patrick Molloy, of the murder of a thirteen-year-old newspaper delivery boy, Carl Bridgewater. There was no corroborating forensic evidence and Molloy later retracted his confession, but to no avail. To reinforce the credibility of Molloy’s confession, the police produced what they claimed was a contemporaneous verbatim record of an interview which contained substantially the same information expressed in words very similar to those of the confession. Molloy denied that this interview had ever taken place and counter-claimed that the interview record had been made up later, on the basis of the, by then pre-existing, confession. There was again textual incoherence, but the concept of uniqueness of encoding was developed to provide extra evidence. Below is an extract from the questioned interview record with Molloy’s responses numbered for ease of reference; (the significance of bold will be explained later).

1. Extract 1

   (1)
   P	How long were you in there, Pat?
   M	I had been drinking and cannot remember the exact time that I was there, but whilst I was upstairs I heard someone downstairs say ‘be careful someone is coming’.
   (2)
   P	Did you hide?
   M	Yes I hid for a while and then I heard the bang I have told you about.
   (3)
   P	Carry on Pat?.
   M	I ran out.
   (4)
   P	What were the others doing?
   M	The three of them were still in the room.
   (5)
   P	What were they doing?
   M	They all looked shocked and were shouting at each other.
   (6)
   P	Who said what?
   M	I heard Jimmy say ‘it went off by accident’.
   (7)
   P	Pat, I know this is upsetting but you appreciate that we must get to the bottom of this. Did you see the boy’s body?
   	(Molloy hesitated, looked at me intently, and after a pause said,)
   M	Yes sir, he was on the settee.
   (8)
   P	Did you see any injury to him? (Molloy stared at me again and said)
   M	Yes sir, he had been shot in the head.
   (9)
   P	What happened then?
   M	I was appalled and felt sick.

The first thing the reader notices is the lack of fit between some of the police questions and Molloy’s apparent replies, particularly in exchanges four, five, seven, and nine. For instance, questions four and five ask for ‘doing’ but receive a ‘being’ response. As noted earlier, Molloy claimed that the interview record must have been fabricated on the basis of the content of the confession statement. This was truer than he could have imagined because, in fact, all the sequences highlighted in italics in the above extract occur identically and in the same sequence in Molloy’s statement, so the incoherence in the sequence is a result of badly formulated questions being inserted to apparently elicit the pre-existing answers. In other words, this is an unusual example of plagiarism. The police defence that in two separate tellings of the same events Molloy would use identical linguistic formulations was dismissed through experimental results both from people being asked to retell the same narrative and from evidence derived from the Google search engine which demonstrated clearly that linguistic encodings by speaker/writers quickly become unique – the moment a sequence reaches ten running words the chances of it having ever been produced before, even by the same author is close to zero (Coulthard, Reference Coulthard2004) and the possibility that Molloy said identically on two separate occasions ‘I had been drinking and cannot remember the exact time that I was there, but whilst I was upstairs I heard someone downstairs say “be careful someone is coming”’ is zero.

This concept of linguistic uniqueness proved useful in many subsequent cases as lay audiences can grasp it easily. The example below was used in a report written by Coulthard for the 2003 Appeal of Robert Brown, who had been convicted of murder and sentenced to life in prison on the strength of a disputed statement and interview transcript. Brown claimed that the confession statement was a combination of actual interview content and incriminating details added by the police, and that the ‘interview’ record had been created from the statement (Coulthard, Reference Coulthard2007). Supporting this version of events, certain phrases appeared verbatim in both documents. While the police argued that one would expect this level of consistency in two descriptions of the same event, Coulthard demonstrated (using one of the questioned phrases ‘I asked her if I could carry her bags she said “Yes”’) that even in a Google-sized collection of texts, we do not see linguistic formulations repeated across documents in this way (See Table 1):

This example has been quoted many times since, most recently in 2022, in an article in the Independent newspaper What are the mathematical chances two Tory MPs would tweet the exact same thing? (Independent 03.05.2022, accessed 21.05.24). While the phrase itself no longer returns zero search results on Google, the many more results that it does return all refer back to this original use.

Detecting Plagiarism

In the 1990’s student plagiarism was growingly seen as a problem for universities. Basically, unlike the police officers in the cases discussed above, the students were employing a strategy called ‘patchwriting’ whereby text was borrowed and then adapted by changing some of the words and the sequencing. Alison Johnson (now May) developed an innovative solution to the detection of this form of plagiarism. She moved away from a reliance on strings or sequences of words as diagnostic, disregarded grammatical words altogether, on the grounds that they are small in number and more likely to be shared anyway, and instead concentrated on the percentages of shared individual lexical items, counting both types and tokens.

Johnson’s initial research (Johnson, Reference Johnson1997) was based on three student essays, all answering the same question, which she felt were too similar to have been composed independently. In order to create a baseline for comparison, she took the first 500 words of three other essays from the same batch, which she had no reason to suspect and calculated the shared lexical vocabulary. This group was found to share only thirteen lexical types, of which seven – language, languages, school, children, multilingual, bilingual, and policy – were almost predictable given that the topic was ‘the advantages and disadvantages of bilingual and multilingual education’. As these words were central to the topic of the essay, they were not only shared but also repeated frequently and thus this set of thirteen shared types made up some 20 per cent of the total lexical tokens used. With these figures as a baseline for comparison, Johnson returned to the three suspect essays and found that they shared seventy-two lexical types, five times as many, which accounted for some 60 per cent of all the lexical tokens. So here was one more tool for the toolkit – degree of lexical overlap.

It is also possible to approach such texts from a different perspective and to examine how much lexical uniqueness there is. Whereas for the control group, the vocabulary unique to each essay constituted between 54 and 61 per cent of the lexical tokens, two of the suspect essays had only 16 per cent of unique lexis and the third, which had appeared to the examiner to be less derivative, had 39 per cent. Intensive testing has shown that such measures of lexical overlap successfully separate texts which share common vocabulary simply because they are on the same topic, from those which share much more vocabulary because one or more of them is derivative (Woolls & Coulthard, Reference Woolls and Coulthard1998; Woolls, Reference Woolls2003).

The problem for the human reader in trying to detect such collusion becomes evident when Johnson points out that none of the three essays identified as colluding was individually unusual when compared with any of the control essays – in other words had the three ‘guilty’ essays each been marked by different professors none of them would have been suspect … . So, what is needed is an automated checking procedure to enable rapid and reliable comparison of every essay with every other one and this quickly becomes a very large task, even with a comparatively small group of essays – comparing thirty essays each with every other one necessitates 435 comparisons. However, it turned out that such a task could be done very rapidly using Copycatch Gold (Woolls, Reference Woolls2002), a program which computerised Johnson’s original insights – it calculates the quantity of shared lexis and allows the individual user to decide, depending on subject areas and text length, what percentage of overlap will trigger further investigation. Copycatch was later adopted by UCAS, the UK Universities Admissions Service, to detect plagiarism in the 500-word personal statements applicants submit with their applications, a task which requires comparing thousands of essays literally.

Developments in Corpus Methods for Comparative Authorship Analysis

Many of the advances that have taken place in the development of forensic linguistics in the UK run in parallel with the advances made in computer processing power, and perhaps most importantly from a linguistic perspective, the availability of large collections of machine-readable text (or ‘corpora’), and corpus tools with which to process them. In the UK context, pioneering work in the field of corpus linguistic analysis emerged, just as work in forensic linguistics did, at the University of Birmingham, where John Sinclair, assisted by a group of lexicographers whose work runs through this Element oversaw the creation of the Collins Corpus. At the time, it was the largest corpus of English texts in the world, and was created in order to inform the first corpus-driven dictionary, the Collins COBUILD English Language Dictionary (Reference Sinclair and Hanks1987).

Much of the early work in forensic linguistics in the UK involved methods that would be familiar to corpus linguists (e.g., collecting representative language samples and comparing word frequency and patterns of grammar use across those samples). Prior to the availability of computational methods, linguistic analysis was necessarily manual. As corpus data and tools started to become available at the University of Birmingham, Forensic Linguistic researchers soon spotted their potential. Writing on the move from manual to corpus methods in the mid 90s, Eugene Winter noted that this approach not only improved but arguably resuscitated quantitative approaches to linguistic analysis “[manual] word counts died out because of the time-consuming nature of counting words. In any case, nobody seemed to know what to do with the results. Now that we have computers to do such work, the matter can be investigated further” (Reference Johnstone1996: 142)

In 1994, Coulthard published ‘On the use of corpora in the analysis of forensic texts’ in the first issue of the Forensic Linguistics journal. In it he noted that corpus linguistics ‘provides not only a new kind of data, but also new ways of analysing those data on which forensic linguists have traditionally concentrated’ (Reference Coulthard1994: 27). The great advance offered by corpus analysis was that it started to provide information about ‘norms’ of the language. Armed with this information, it was possible to start answering questions of typicality and rarity when examining texts of disputed authorship. By using corpora, linguists have been able to process and analyse vastly more text and do so more reliably and consistently.

With some notable exceptions (e.g., Kredens, Reference Kredens and Stalmaszczyk2002), it wasn’t until the early 2010s that significant amounts of corpus linguistic-based authorship analysis research started to emerge in the UK, most notably work that made use of n-grams (which is to say, strings of consecutive words or characters) in authorship analysis. Authorship analysis has always focused on the extraction and comparison of features from a text. N-grams were an appealing proposition as an authorship marker as they had the potential to provide a lot of comparison data per feature, and to some extent, were immune to accusations that features has been ‘cherry picked’ to suit the intuitions of the analyst. Corpus analysis made massively easier the task of extracting n-grams from texts.

Pioneering work by Jack Grieve (Reference Grieve2007), before his move to the UK, detailed the usefulness of a range of authorship markers that had been used in stylometric authorship analysis. He found that the traditional measures of author style such as word/sentence length and vocabulary richness (in addition to positional stylometry – i.e., where a word appears in a sentence), were not good features for determining authorship. However, n-grams (sequences of words or characters) showed promise. Two and three-character n-grams were found to be particularly good at distinguishing between authors. As the number of characters increased, the accuracy decreased as the long combinations of characters started to become more indicative of full words and therefore topic, rather than style (Reference Grieve2007: 261). A slow but steady increase in interest in n-grams as the focus of stylistic authorship analysis around the turn of the decade started to materialise into publications. Barlow (Reference Barlow2013) carried out an analysis of bi-grams, tri-grams, and part-of-speech n-grams in the speeches of White House Press secretaries, comparing frequencies using a correspondence analysis and finding them to be highly effective at distinguishing between authors.

While advances were being made in the identification of more reliable authorship markers, advances were also being made in the more ‘methodologically rigorous’ (Grant, Reference Grant2013: 472) approaches to analysing these markers. What is required for authorship analysis in a forensic context is a method in which the measurement of similarity or difference between authors’ styles or texts can be represented statistically, but is also both explainable in linguistic terms and ideally with a known error rate. While authorship analysis has a proven record in the UK context, it was still seen by some as lacking in some of the scientific rigour and quantitative power of other forensic techniques. An approach developed by Tim Grant (Reference Grant, Coulthard and Johnson2010, Reference Grant2013) was one of the first to address some of these concerns, complementing the established stylistic approach employed by Coulthard.

The approach was developed to examine text message data in a murder case. In June 2005, teenager Jenny Nicholl disappeared, and the main suspect was her estranged lover, David Hodgson. After her disappearance, four text messages were sent from her mobile phone. Coulthard was asked to look for evidence to support the hypothesis that the four messages had not been written by her, and whether, if not, they could have been written by Hodgson.

The police provided Coulthard with two sets of pre-crime text messages: one known to have been written by Nicholl and the other known to have been written by Hodgson, against which the four questioned texts could be compared. Coulthard identified a set of nine items for which the pair made different linguistic choices and which served to distinguish between the texting styles of Nicholl and Hodgson (Table 2).

Each of the four messages contained items where the Hodgson choices were used and none of them where the Nicholl choices were selected – for example

Coulthard gave evidence at the trial of David Hodgson where, rather than attributing authorship, he gave the measured opinion that ‘Linguistic features identified in Mr Hodgson’s and the suspect texts are compatible with their having been produced by the same person’ (Grant, Reference Grant, Coulthard and Johnson2010: 517). Crucially for the status of forensic linguistics in the UK, when Hodgson’s conviction was appealed on the basis that the linguistic evidence was unsound, the appeal failed. This effectively confirmed the acceptability of this kind of linguistic evidence in UK courts.

Despite the success of descriptive methods such as those used by Coulthard, the stylistic approach could be criticised for relying heavily on the intuition of the expert witness. In a Reference Grant, Coulthard and Johnson2010 chapter, Grant detailed how Coulthard’s analysis in this case could be given quantitative support by borrowing measures of statistical similarity that had previously been used in behavioural crime linkage (e.g., Woodhams et al., Reference Woodhams, Grant and Price2007). Rather than simply identifying notable features in the texts of known authorship and highlighting their use in the texts of disputed authorship, Grant gave these features a binary value based on their presence/absence, and calculated similarity scores, using the Jaccard co-efficient (a statistic which expresses the similarity between two sets of binary figures as a number between zero and one), between texts of known and unknown authorship. In this way, he was able to demonstrate that the texts of known authorship were internally consistent and distinctive from one another. He was also able to show that there were distinct differences between the questioned texts and Nicholl’s texts, but no distinctive difference between the questioned texts and Hodgson’s texts.

In his report in the case of the suspected murder of Amanda Birks in January 2009, Grant worked with a descriptive, corpus-based, stylistic methodology, like the one successfully applied by Coulthard. As in the Nicholl case discussed above, the evidence was a series of questioned text messages sent from Amanda’s phone from midday on the day she died. Grant was provided with three sets of messages, one sent by her husband, a second by Amanda and the questioned set. Again, the task was to discover whether the writing style of the disputed messages was closer to that of Amanda’s known writings or to those of her husband. Grant identified eighteen features which had ‘discriminative power’, ten of which were characteristic of Amanda’s style, and eight of her husband’s. These features included alternate spelling variants such as ‘bak’ for ‘back’, ‘wud’ for ‘would’, ‘wen’ for ‘when’, and the use of the digits two and four with no space before the following word. Grant was able to show that Amanda and her husband had consistent and distinctive text messaging styles.

While Grant’s approach of using similarity measures did not form part of the evidence presented to the court in this case, he continued to demonstrate its potential in academic papers, publishing a case study of how this method could have been applied in the Birks case in 2013. In the paper, the questioned messages were coded for the presence or absence of each of the eighteen style markers. The binary string of presences and absences of features were compared using the Jaccard coefficient. Based on these statistical results and a further more qualitative examination of the data, Grant concluded that the stylistic choices in the questioned messages were significantly different from those of Amanda’s known messages, but not from those of her husband. A full description of this method can be found in Grant (Reference Grant2013).

Building on Grant’s experimental use of the Jaccard similarity coefficient in the Nicholls and Birks cases, Johnson and Wright (Reference Johnson and Wright2014) developed a methodology for identifying authors using data from the anonymised Enron email database. They set out to capture idiolectal collocations first creating sets of n-gram textbites, that is, lexical strings between two and six words long and then identifying the set of the personal co-selections or ‘collocates’ of these textbites, by using a piece of lexical analysis software called Jangle (Woolls, Reference Woolls2013). They discovered, for example, that although please occurred in the emails of 165 of 176 authors in the corpus and a total of over 11,000 times, it was still possible to identify individual and idiolectal variation within please-initial word strings. For example, ‘please print the message’, ‘please format and print’ and ‘please proceed with your’ occur exclusively in the emails sent by one employee. Such strings, it was argued, can have population-level distinctiveness, can characterise an author’s writing style, and so serve to identify authors.

Johnson and Wright found that five-word strings (n-grams) correctly identified the author of a set of emails with 100 per cent accuracy, when the sample size was over 750 words. Of course, in real-life cases samples are frequently much smaller and when the size of the disputed samples was reduced to 55–145 words, the success rate for five-word n-grams fell to 30 per cent, with an error rate of 70 per cent. But at the same time it must be emphasised that this was an extremely rigorous test with 175 possible authors, while in a forensic comparative context there are regularly only two candidate authors and very, very rarely more than four. This was confirmation of the claim made in Coulthard (Reference Coulthard2004) that idiolects were distinguished and therefore distinguishable in terms of preferred co-selections of words and phrases, although successful identification, in any given case, would depend on whether there was a sufficiently large corpus of known and questioned text(s).

With the use of these corpus approaches gaining momentum, much of the work that has been done in the past decade has involved the development and refinement of some version of this approach. Nini (Reference Nini2018) used n-grams as both a point of comparison and a jumping-off point for further qualitative analyses of stylistic difference. His study examined letters of disputed authorship in the ‘Jack the Ripper’ case. Like Wright, he chose to focus on word n-grams noting,

Like Grant (Reference Grant, Coulthard and Johnson2010, Reference Grant2013) and Wright (Reference Wright2017), Nini used the Jaccard coefficient measure to compare the presence/absence of these n-grams in letters of relatively undisputed authorship with those that were received when the initial letters had gone public. To establish the population-level distinctiveness of the n-grams, Nini also examined their frequency in contemporary sections of the Corpus of Historical American English (COHA), the Corpus of Late Modern English Text (CMLET3), and the Old Bailey corpus.

Nini found that two famous letters in the series ‘Dear Boss’ and ‘Saucy Jack’ were more similar to each other in their use of two-word n-grams than 95 per cent of all other possible text pairings in the Jack the Ripper corpus. Furthermore, the two texts share a distinctive four-word n-gram ‘letter back till I’ which not only individuating within the Jack the Ripper corpus, but also is extremely rare in the general comparison corpora. On this basis, Nini argued, there is compelling evidence that these letters were written by the same author, and that other famous letters in the dataset were likely not written by the same author.

Most recently the power of n-grams to distinguish between authors has been developed by Grieve et al. (Reference Grieve, Clarke and Chiang2019) into a method called ‘n-gram tracing’ which involves: extracting all of the n-grams that appear in the dataset; considering the presence versus absence of features (using the Overlap Co-efficient, rather than Jaccard); measuring the percentage of n-grams in the questioned document that occur at least once in the possible author sample; attributing the document to the author that uses the highest percentage of n-grams. Grieve et al. used this method to investigate the long-disputed authorship of the Bixby letter, a 139-word text sent by Abraham Lincoln to Lydia Bixby, the mother of five US soldiers supposedly killed during the Civil War. Ultimately, the n-gram tracing method provided strong evidence that it was not Lincoln, but Lincoln’s secretary John Hay, who authored this most famous piece of correspondence. They found that both word and character level n-grams performed well in distinguishing between Lincoln and Hay, with character n-grams proving particularly powerful. Between five- and ten-character n-grams achieved an attribution accuracy of 95 per cent on average, with seven and eight character n-grams performing as well as 98 per cent. Grieve noted that the character-level n-gram results in particular were ‘especially remarkable given the brevity of many of the texts, a majority of which contain fewer than 200 words and 10% of which contain no more than 50 words’ (Reference Grieve, Clarke and Chiang2019: 24).

The distinction is sometimes made between ‘stylistic’ and ‘stylometric’ approaches to authorship analysis, the former involving the identification and comparison of style markers, and the latter a comparison of authors using the relative frequency of sets of linguistic features. While much of the research described above is stylistic, recent research in the area has also made use of stylometric approaches. A recent example of this is Clarke and Grant in their (Reference Clarke and Grant2025) work on the ‘specific imposters method’ (Reference Clarke and Grant2025). This is a variation on the ‘general imposters method’ (Seidman, Reference Seidman2013; Koppel and Winter, Reference Koppel and Winter2014; Potha and Stamatatos, Reference Potha, Stamatatos, Jones, Lawless and Gonzalo2017), in which rather than comparing known and questioned texts directly, they are compared to ‘imposter’ texts using a random subset of features. Where the known and questioned texts are more similar to each other than they are to any of the ‘imposters’, it is suggestive of shared authorship. Clarke and Grant used this method to compare the writings of eco-terrorist Nikolaos Karvounakis and a note claiming responsibility for planting an explosive device, to contributions of other authors to his online zines, finding Karvounakis to be the likely author of the note.

As the methods for authorship analysis have become more sophisticated, attempts have also been made to maintain a degree of accessibility, most notably in the form of Andrea Nini’s (Reference Nini2024) ‘Idiolect’ package in the programming language R. This computational tool suite was developed specifically for authorship analysis in the forensic context. It is designed to assist comparative authorship analysis through the functions listed below, which reflect the typical workflow involved in forensic authorship casework:

1. (1) Data input

2. (2) Optional masking of texts

3. (3) Running analyses including Delta, n-gram tracing Imposters Method and LambdaG

4. (4) Performance Testing

5. (5) Applying method(s) to questioned/disputed text(s) and calibrating Likelihood Ratios

Importantly, this package is designed to suit the needs of both academic researchers and those engaged in forensic authorship casework, making methods that have evolved a long way since the early corpus work in the field, accessible to academic and professional audiences.

Cross-Genre Authorship Analysis and the Idea of ‘Idiolect’

In addition to pioneering work developing and refining methods for authorship analysis, researchers in the UK have also interrogated some of the founding assumptions of this work. While the notion of an individual version of language (or ‘idiolect’) has been theoretically central to most if not all authorship analysis research, a tolerated limitation of this work has always been that we cannot hope to capture the entirety of an individual’s language use (i.e., their literal idiolect). Any language that an author produces will be context-dependent, and any sample we have of it will always be a partial picture. Success in authorship tasks relies on having a comparable language sample.

It was always good practice to ensure that comparable reference material was available, but the importance of taking genre into account was confirmed by Kredens et al. (Reference Kredens, Heini and Pezik2022), who found that even the most advanced computational methods for authorship analysis struggle when the variable of genre is introduced. The data used in this research came from the 100 Idiolects Project, in which samples of seven discourse types were collected from 112 authors (Heini and Kredens, Reference Heini and Kredens2023). The aim was to investigate the persistence of individual stylistic features across different modes of textual production, a question also explored by Mojedano Batel et al. (Reference Mojedano Batel, Soler Bonafont and Kredens2024). They explored cross-genre idiolectal stability by collecting and analysing Spanish language data from six participants across four discourse types (emails, text messages, an interview, and work meeting transcripts). Through their analysis they found that epistemic modality has potential as an authorship marker as it displayed a degree of both cross-genre consistency and distinctiveness.

While computational approaches to authorship analysis are getting increasingly powerful, these recent findings suggest that there remains a role for qualitative research, particularly in cross-genre comparison and in producing explainable answers to authorship questions.

Emergency Services

While not as widely researched as police and court interactions, there is a small (and growing) body of research into communication with the Emergency Services, which are the first points of contact that users have with the legal system. Garner and Johnson (Reference Johnson and Brown2006) studied emergency call management in England. They were interested in ‘Where are the critical decision-points for the call handler?’ and ‘How can critical information elements be established rapidly?’ They noted that in calls where the caller was ‘argumentative, frightened or upset’, the process took considerably longer and that such calls can later become a crucial part of the evidence in a prosecution case, particularly if they are made by a caller who later becomes identified as a suspect. They also noted that, while contemporaneous record keeping can both structure the interaction and be useful for the call-handler, it can also compromise the flow of the call and even prevent the call-handler from picking up on crucial linguistic clues. The institutionality of the discourse, a key consideration of forensic linguistics more widely, was also noted. They observed that the caller’s definition of an emergency may not meet the institutional criteria that justify deploying the emergency services, so a successful call is one where sufficient, relevant information has been elicited to allow the call-taker to correctly assess the seriousness of the situation. This complex task of matching individual caller claims to institutional criteria was also explored by Drew and Walker (Reference Drew, Walker, Coulthard and Johnson2010) who showed how ‘the request forms selected by citizens calling for police assistance … encode … assessments of their entitlement’ and that there can at times be a mismatch between the caller’s conveyed ‘perception of the seriousness/urgency of the incident … and the assessment of the call taker’.

Some more recent studies have looked at how the communicative challenges of this context interact with particular kinds of emergencies. Stokoe and Richardson (Reference Stokoe and Richardson2023) looked at ways in which callers who have been subject to domestic abuse formulate requests to emergency services, with a particular focus on calls where the perpetrator of that abuse may be present at the time of the call. They used conversation analysis to examine nearly 200 audio recordings, looking at how callers interact with call takers, and how call takers interact with the operator who first receives the call. In this analysis they found both relatively straightforward requests for help and an interesting range of non-lexical resources callers employed to communicate their request for assistance while at the same time trying to conceal their request from the perpetrator.

Atkins et al. (Reference Atkins, Richardson, Traynor and Deamer2025) examined cases of kidnapping and how reports of this ‘crime-in-action’ are communicated by callers and institutionally categorised by call-handlers. They used conversation analysis to examine forty-eight calls, some perceived to have been correctly and some incorrectly categorised at the time of receipt; they focused on points of difficulty in the reporting and categorisation processes. They found two main issues with reporting: first, the call-handlers sometimes had difficulty in eliciting sufficient information, for example, in cases where the callers had important gaps in their knowledge about the event they were reporting, or were ‘unwilling to respond to information-seeking questions’; and, second, with discovering the callers’ level of entitlement to police assistance. Both of these issues were shown to impact the call-handlers’ decision-making regarding the categorisation of calls and the subsequent deployment of police. As Atkins and colleagues point out, this work provides an evidence base for call handler training, and in doing so, demonstrates the practical outcomes of conversation analytical methods when applied in such high-stakes contexts.

Support Services

While calls to emergency services are perhaps unique in the level of response immediacy required, similar issues exist in disclosures of information to, and requests for assistance from, other support services. Sam Larner’s final piece of research before retraining as a barrister, focused on issues of child communication in the legal system; one was a study of children’s disclosures of sexual abuse in online fora. In common with research into police interaction, the focus of this study was not only on how the child spoke, but also on how the practitioner (in this case, someone working for the child counselling service Childline) facilitates the disclosure. Combining move and corpus analyses, he showed how practitioners provide appropriate vocabulary and ‘scaffolding’ for the children’s accounts. While the children might talk about abuse in vague terms, for example, ‘things’ that happened to them, practitioners label these actions (e.g., ‘rape’) much more explicitly, and in doing so, ‘provide the children with specific labels to describe their experience’ (Larner, Reference Larner2022: 284). He also identified a theme which recurs throughout the literature on interaction with legal professionals, namely that there is a tension between the roles of those involved in this interaction and the various purposes of the elicited report. He contrasted the therapeutic responsibilities of the Childline counsellor with the safeguarding responsibilities that other professionals, such as teachers, have. He reflected that, while the therapeutic approach elicits clearer disclosures, the active role that the practitioner plays in supporting its production may undermine its evidential value if it were to be adopted by other professionals.

As we shall see in the following sections, this tension between the way in which answers are elicited and the intended audience for those answers has been a recurring theme in the study of spoken interaction in legal contexts.

Introduction

The longest running subject of forensic linguistic research is the language of police interactions. Some of the earliest studies focused on the disputed authorship of police statements. This is an area in which the linguistic study of the topic has its origins in the 1990s Birmingham Forensic Linguistics Research Group, although later pockets of research emerged across the UK, frequently conducted by researchers in the fields of psychology, criminology, and sociology, interested in communication in this context rather than ‘forensic linguistics’ specifically. In general, due to the limited space, this volume maintains a fairly strict focus on forensic linguistics. However, given the value of the insights that these studies have provided and the often overlapping and interdisciplinary nature of this area of study, this section refers to relevant research from multiple disciplines, though, as ever, foregrounding the linguistic. Coulthard’s (Reference Coulthard and Cotterill2002) paper is a helpful place to start in terms of the examination of police language and why it received so much attention in the early days of forensic linguistics in the UK. Coulthard reports on three cases in which he was called on as an expert witness. In each case the suspect had been ‘verballed’, which is to say their supposedly verbatim statement had been fabricated to some degree by the police (for more on this, refer back to the ‘Early applications of Authorship Analysis in the UK context’ section).

A legal development in the UK that had a significant impact on this area of research was the passing into law of the Police and Criminal Evidence Act (or PACE) in 1984, which introduced certain standards and rights around the arrest, detention, investigation and interviewing of suspects. Perhaps most notably for our purposes, Code E of this act made it a requirement to make audio recordings of any significant police interview. Surprisingly, forty years later, the UK remains one of only five countries in which the recording of interviews is required. While this did not fix every problematic aspect of police interviews, it did at least limit the kind of police manipulation of suspect statements that was the focus of early forensic linguistic studies. PACE and later the Criminal Justice and Public Order Act, 1994, also made changes to the Police Caution – the rights that are read out when a suspect is arrested and which are similar to Miranda rights in the USA – changes which are important contextually for the study of police language in the UK. First, the wording of the Caution was standardised. Second, in the revisions made in the 90’s, the Caution was modified to restrict a speaker’s right to silence by allowing inferences of guilt to be made when a speaker refuses to answer potentially incriminating questions. The earlier version of the caution, ‘You do not have to say anything, but anything you do say may be given in evidence’, was expanded to include the caveat that ‘it may harm your defence if you do not mention when questioned something which you later rely on in Court’ (PACE, Code C., Section 10.5, 1984).

Also important to the research into police language were attempts made in the early 1990s to promote improved interview practices, exemplified by the PEACE investigative interviewing guidelines, which were introduced in 1993 with the aim of encouraging more collaborative and less controlling and accusatory interview practices (Bull and Baker, Reference Bull, Baker, Mason and Rock2020). The acronym itself is a little cumbersome but stands for Planning and Preparation, Engage and Explain, Account Clarification and Challenge, Closure, and Evaluation. The first two of these phases relate to establishing rapport and explaining the purpose of the interview. The ‘A’ phase should involve an uninterrupted elicitation followed by clarifying questions. And in the final two phases, the interviewer concludes the interaction and evaluates what has been said.

Researchers in the UK have approached these interactions from various angles. Some studies have focused explicitly on notions such as coercive questioning or the distinction between an ‘interview’ and an ‘interrogation’ (Johnson, Reference Johnson and Brown2006); others focused more on how particular strategies or concepts function, but all engaged to some degree with questions of power (imbalance) and institutional goals (Drew and Heritage, Reference Drew, Heritage, Drew and Heritage1992).

Comprehensibility of Police Procedures (e.g., Caution)

Some of the earliest work on spoken police language, particularly regarding its comprehensibility, was done by Janet Cotterill. She was one of the founding members of the Forensic Linguistics Research Group at the University of Birmingham and, as mentioned above, was later the founder of the Cardiff MA in Forensic Linguistics. Her research interests were many, including trial language (Reference Cotterill2002, Reference Cotterill2004, Reference Cotterill2005) and the language of sexual violence (Reference Cotterill and Cutting2007). Among her earlier publications was a paper in the journal Forensic Linguistics about the language of UK police cautions, specifically how officers, after cautioning suspects by using the fixed formula, set out to explain the Caution in their own words, something that they are expected to be able to do when suspects have difficulty understanding. Crucially, the study focused on the police officers’ delivery rather than on the suspects’ comprehension problems, which up to that point had been foregrounded as the most likely cause of misunderstandings. Cotterill focused on a set of explanations produced by fifty police officers and identified a small set of lexical items which were causing specific comprehension problems. She noted a tendency for police officers to think of the Caution as a ‘warning’ rather than as an advisory notification of rights and also for officers’ explanations to include many of the complex lexical items that they were supposed to be explaining. The article also highlighted the semantic difficulty inherent in key terms in the Caution such as ‘harm’, ‘mention’ and ‘rely on’, particularly when their specific legal meaning differed from that of normal everyday use.

Studies of the comprehensibility of police ‘texts’ are also among the many contributions to the field made by Frances Rock, a colleague of Cotterill’s at Cardiff. In her Reference Cotterill and Cutting2007 book Communicating Rights: The Language of Arrest and Detention, she focused on two texts in particular: the Police Caution (see also Rock, Reference Rock, Solan and Tiersma2012), and the Notice to Detained Persons (now Notice of Rights and Entitlements: A Person’s Rights in Police Detention), both written texts, informing detainees of their rights. In both cases, as well as analysing the text itself, Rock examined the processes by which officers adopt and revise them, and how they are actually used (Reference Cotterill and Cutting2007: 7). The examination of the Notice to Detained Persons demonstrated how the original version of the text contained formal and complex textual and organisational features that limited accessibility, and while it was possible to create more reader-friendly versions, the texts retained an intimidating institutional status that was harder to overcome (Reference Cotterill and Cutting2007: 246). The wording and organisation of the Caution, meanwhile, were shown not only to be problematic but also recognised as such by those using it.

The Caution is perhaps the most significant single police text to receive research attention, but other elements of police procedure have been examined in terms of their comprehensibility. Rock also investigated requests made by police officers for consent to process individuals’ personal details or involve them in investigative procedures, finding them to be ‘inadequate or even nonsensical’ (Reference Rock, Ehrlich, Eades and Ainsworth2016a: i). Also focusing on police delivery (that is, the communicative responsibilities of the police rather than deficits of the hearer), Aldridge-Waddon (Reference Aldridge-Waddon2019) examined the effectiveness of the delivery of the opt-out procedure for child witnesses. This procedure allows children, when giving evidence to a court, to opt out of taped and video-link interviews, measures that were originally introduced into the UK in 1999 as a way to support young and vulnerable witnesses. These apparently well-intentioned measures were made optional in 2009. because it had become apparent that they were not suitable for everyone. The study found that, in general, police officers were not delivering this opt-out procedure effectively, with shortcomings identified in terms of the provision of relevant information, bias in how the information was presented, and the level of language awareness displayed. Greater awareness of the importance of simple linguistic encoding was suggested as the most likely factor to increase comprehensibility.

There is a danger that focusing on the linguistic delivery of legal professionals could be seen as a form of performance review, an unwanted critical opinion from someone who does not really understand the requirements of the job – a concern that Eades et al. (Reference Eades, Fraser and Heydon2023: 10) discuss in relation to the Australian context. However, a valuable thread that has emerged from research in the UK is acknowledging the high level of linguistic complexity involved in some of the tasks that police offices are expected to undertake. And while there has been a tendency to focus on the challenges that legal interactions create for lay and relatively disadvantaged participants, perhaps one of the reasons that collaboration with law enforcement in general, and the police in particular, has been comparatively successful in the UK is the recognition by researchers that it is important to not just criticise practitioners but to support best practice.

Police Interviews

Since the introduction of the systematic recording of police interviews in the UK, the focus of linguistic research has shifted from examining potential fabrication towards studying interviews both as interaction (i.e., looking at how police officers use certain features and strategies, and/or how they co-construct an interview with a witness or suspect), and as text (i.e., how the transcript of the interview is produced, how it is used in court … etc).

Studies of Police Interviews as Interaction

A relatively early examination of police language in the UK context was Auburn et al.’s (Reference Auburn, Drake and Willig1995) study of ‘discourses of violence’ in police interviews. They looked at eleven police-suspect interviews and carried out an inductive thematic analysis of discussions of violence by both participants. They identified two types of violence (‘disorderly’ and ‘justificatory’), a generic structure (or ‘grammar’ in their terms) of accusatory descriptions of violence, along with a range of ways in which speakers position themselves in relation to those descriptions in their questions and responses. They also found that institutional role has an impact on how these types of violence are employed, noting that ‘although in principle, both participants … have [the] right to construct events using these discourses, the recognised authority of the police facilitates their use of the accusatory grammar and the ability to call others to account’ (Reference Auburn, Drake and Willig1995: 384). This focus on institutional goals and the relative power that police have in controlling interactions has been a recurring theme in this area of study.

Drawing on some of the same corpus data and methods available at the University of Birmingham that were so game-changing in authorship analysis research (see Section 2), Alison Johnson (now May, Reference Johnson and Cotterill2002) examined the use of ‘so’-prefaced questioning in police interviews with child witnesses. The study compared spoken data in the COBUILD corpus with a corpus of police interviews with child witnesses and adult suspects in abuse cases, using a combination of corpus analysis, conversation analysis and semantic-pragmatically focused discourse analysis. The study found that so-prefacing was much more frequent in interview data than in general use. It fulfilled a range of functions from supporting and providing a ‘scaffold’ to help young interviewees to share their accounts, to summarising and evaluating adult defendant accounts in the interviewer’s terms.

The ways in which speakers control these interactions have been shown to go beyond questioning. Johnson (Reference Johnson2008b) also explored institutional goals in suspect interviews, looking at how lay suspects’ narratives were ‘transformed and institutionalised’ through negotiation in interactions with the police. Three interviews and four types of negotiation of the narrative were examined: negotiation of framing and audience, responsibility within the narrative, summarising and reformulating narrative details to have more evidential value and negotiation of their role within the interview (Reference Johnson2008b: 331). While a complex picture of these negotiations emerges with interviewees shown to be exercising some control over their own narratives, ‘most of the negotiating resources are in the control of interviewers (Reference Johnson2008b: 346), and these resources are used to reshape narratives towards institutional goals. Holt and Johnson (Reference Holt, Johnson, Coulthard and Johnson2010) provide a summary of some of the research, including their own, into the ways in which stories are formulated in legal settings, focusing on reported speech, repetition formulations, and contrasts.

While these approaches to controlling interactions in legal settings have typically focused on the legal professional, conversation analysis has also been used to examine suspect resistance to questioning. Among Kate Haworth’s earlier studies was an examination of power and resistance in an interview involving police and the serial killer Harold Shipman; specifically, how they were realised through topic, question type, question and answer sequences and allusions to institutional status (Reference Haworth2006: 743). In each of these areas, the suspect is shown to have resisted (and even co-opted) typical strategies employed by the police to assert control. (N.B. The same strategies were also the subject of a paper by Newbury and Johnson (Reference Johnson and Brown2006), while Johnson in a 2008 paper explored evaluation patterns and the possibility for both interviewers and interviewees to adapt their roles and strategies.) Haworth noted that it is only when the police interviewer ‘[gives] up his powerful discursive position in order to allow [the suspect] to speak freely’ that the suspect effectively self-incriminates (Reference Haworth2006: 755) and therefore suggests that best interview practice should consider the importance of at times relinquishing control.

The benefits of a less controlling approach were also demonstrated in Kelly Benneworth-Gray’s qualitative discourse analyses of investigative interviews with suspects accused of sexual assault (Reference Benneworth2009, Reference Benneworth, Coulthard and Johnson2010, Reference Benneworth-Grey2015). She identified two approaches taken by police officers: one ‘closed’, characterised by restrictive accusatory questioning, sexually explicit language and the presentation of a ‘definitive’ account which requires little input from the suspect; the other ‘open’, characterised by open questioning and by providing opportunities for the suspect to tell their story (Reference Benneworth, Coulthard and Johnson2010: 152). While the latter has long been the recommended approach to investigative interviewing in the UK from an ethical standpoint, Benneworth-Gray demonstrated how it is also the more effective in eliciting information and even admissions of guilt.

Benneworth-Gray also used this interview data to examine how police interviewers invoke the ideas of ‘truth’ and ‘honesty’. These concepts were used towards the start of interviews as part of a relatively complex reciprocal arrangement in which promises of truthfulness were offered and elicited. This agreement was then drawn on later when encountering discrepancies in accounts of events. As Benneworth-Gray noted, the key contribution here was to look at the observable interactional function of truth invocations, rather than the more attractive but practically problematic question of whether someone is lying (Reference Benneworth-Grey2015: 270). A similar approach was taken by Rock (2013b) when examining how trust and distrust factor into statement-taking interactions between police and witnesses, and looking at ‘key moments in which trust … becomes salient’ (2013b: 203)

Research carried out predominantly at Loughborough University by Elizabeth Stokoe and other members of the Discourse and Rhetoric Group (DARG) has provided, and continues to provide, a great deal of insight into institutional interactions, including those involving the police and the language they use. For example, Stokoe and Edwards have investigated the use of racial insults in neighbour complaints and police interviews (Reference Stokoe and Edwards2007), the use of ‘silly questions’ in police-suspect interrogations (Reference Stokoe and Edwards2008), how lawyers intervene to take turns in police-suspect interrogations (Reference Stokoe, Edwards, Coulthard and Johnson2010), and the use of ‘no comment’ responses in investigative police interviews (Reference Stokoe, Edwards and Edwards2016), to name just a few.

Despite the assumption that there must be a conflict between the objectives of police officer and suspect, Haworth (Reference Haworth2017) drew upon elements of conversation analysis to highlight, in a sample interview with a rape suspect, a striking example when the police officer appeared to draw upon less incriminating details elicited during the interview to, in essence, construct a defence case for the suspect (Reference Haworth2017: 221). While no judgements are made as to the facts of the case in question, it is a powerful reminder that controlling questioning can be employed just as easily to support an assumption of innocence as it can to support one of guilt.

In a less troubling examination of co-operation in police interviews, Tkacukova (now Grieshofer) and Oxburgh looked at co-operation in interactions with suspected sex offenders and highlighted the importance of the second interviewer in clarifying answers and acting as ‘a mechanism for quality assurance’ (Reference Rock, Mason and Rock2020: 149) Another aspect of co-operation was examined by Rock (Reference Rock and Preece2016b) when she investigated not just how interviewees (in this case witnesses) construct their side of the interview, but also at how the ‘identity’ they assume as ‘novice’ or ‘expert’ relates to their testimony. She also assessed the advantages that each approach offers: while obviously contrasting, both identities are shown to offer ways of managing social distance and presenting oneself as a reliable witness. Rock (Reference Rock, Coulthard and Johnson2010) also looked at the exploitation of identity, expressed using the Conversation Analysis concept of ‘membership categorisation’, in interviews between police and suspects accused of violence towards women. In particular, she showed how suspects invoked the category of ‘men who hit women’ specifically in order to distance themselves from it.

Vulnerable Participants and Police Interaction

Communicating in legal contexts is challenging at the best of times, but there can be additional factors that make participating even harder. These are recognised under UK law as ‘vulnerabilities’, and the linguistic experiences of vulnerable participants form another thread of UK forensic linguistics research.

Under UK law, there are a number of factors that are recognised as constituting a vulnerability. They include, but are not limited to:

1. (a) Age (being under eighteen years of age);

2. (b) Communication or language difficulties (including literacy);

3. (c) Physical disability or impairment and health condition;

4. (d) Mental health condition or significant impairment of any aspect of their intelligence or social functioning (including learning difficulties);

5. (e) The impact of the subject matter of, or facts relevant to, the case (an example being having witnessed a traumatic event relating to the case);

6. (f) Relationship to an involved party or witness (examples being sexual assault, domestic abuse or intimidation (actual or perceived));

7. (g) Social, domestic or cultural circumstances. (https://shorturl.at/IDbHW)

It is also often the case that these vulnerabilities overlap and intersect. There has historically been an unfortunate tendency in terms of policy and professional practice to blame communication difficulties in legal contexts on interviewee ‘deficits’ (Cotterill, Reference Cotterill2000: 7; Stephenson et al., Reference Stephenson, Giller and Brown2011: 88; Case et al., Reference Case, Lorenzo-Dus and Morton2023: 44). While we are keen to avoid doing so here, it is nonetheless worth noting that the vulnerabilities listed above can make an already challenging communicative task even more challenging, and the studies listed here have attempted to explore how.

In addition to the general reforms and recommendations for improving police practice that have been made across the UK in the past four decades, this period has also seen the introduction of measures intended to improve interview practice specifically for vulnerable witnesses. Most relevant to the police interview context is the Memorandum of Good Practice for Video-Recorded Interviews with Child Witnesses, published by the Home Office in 1992, and the Achieving Best Evidence (ABE) interview guidelines on preparing and supporting vulnerable witnesses in court (originally Davies and Bull, 2002, revised by Milne, Reference Milne2007, Reference Milne2011, and finally Smith and Shaw, Reference Smith and Shaw2022). See Aldridge-Waddon (Reference Aldridge-Waddon, Coulthard and Johnson2010: 296–298) for an excellent summary.

An important subfield of linguistic research into legal interactions with vulnerable participants concerns victims of sexual assault and rape. Nicci MacLeod’s work on police interviews with rape victims is a good example of how micro-level linguistic analysis can point to subtly conveyed ideological stances about sexual violence. Using a corpus of video-recorded police interviews to examine interactional control over content, MacLeod (Reference MacLeod2009) looked at, on the one hand, the use of the discourse markers ‘well’ and ‘so’ by officers to attach importance to particular aspects of witness accounts, while, on the other hand, choosing to omit from the final record some of the information provided by the victim. MacLeod also noted that interviewing officers often opted to use language more closely associated with romantic relationships than with sexual violence when reformulating the rape victims’ accounts of events, for example, asking whether the suspect ‘made a move’ or whether they were ‘affectionate towards’ the victim. In later work, MacLeod (Reference MacLeod2016) found linguistic evidence showing that rape victims in fact anticipate blame in police interviews and the presupposition that they were in some way responsible for the attack. Victims attempted to pre-empt conjecture around reckless behaviour or lack of caution, drugs and alcohol, prior relationships with suspects and ‘appropriate’ resistance to the rape; all themes related to well-entrenched cultural myths that are typically used to downgrade the seriousness of the sexual encounter.

Forensic linguistic research in this area has also looked at the impact that the special measures have had on vulnerable participants’ experience of the legal system. Aldridge and Luchjenbroers (Reference Aldridge and Luchjenbroers2008) examined some ‘unforeseen consequences’ of the introduction of video-taped interviews in cases involving vulnerable witnesses, by analysing narrative and framing. In particular, they focused on the contrasting purposes and skill sets of the police conducting the interviews and the lawyers presenting the evidence in court. They found that the police officers’ elicitation of free narratives, while positive in the sense of promoting open communication, can become problematic when the interview is used as evidence in chief, because it has the potential to introduce inconsistencies into the testimony. They also demonstrated that video-recorded police interviews are not designed to elicit and frame details of evidence in a way that supports the case, as questioning in open court by a prosecution lawyer would. Instead, the questioning not only failed to present the testimony adequately, but at times may even have undermined it by employing the previously referred to framing found in popular rape myths. In a Reference Aldridge-Waddon, Luchjenbroers, Freeman and Smith2013 chapter, Aldridge and Luchjenbroers go further, exploring a range of unhelpful stereotypes that are invoked in video-taped police interviews with child witnesses, because of a questioning style which they characterise as ‘suggesting scepticism and doubt’ (Reference Aldridge-Waddon, Luchjenbroers, Freeman and Smith2013: 308). Ultimately given all of the issues that they uncovered, they concluded that such interviews should not be used as a child’s Evidence in Chief.

As part of their wider body of research on police interaction, conversation analysts Charles Antaki, Emma Richardson and Elizabeth Stokoe worked with clinical psychologist Sara Willott on a number of studies analysing a set of twenty police interviews with victims of sexual assault who also had intellectual disabilities, comparing written guidance with the actual interactional practice. They looked at the use of probing questions, to see whether they were used to clarify, in line with the ABE guidelines, or in a challenging manner, and found that while ‘much of the interviewers’ questioning is in line with the guidelines’, nearly all of the interviews included an element of challenge, sometimes also implying blame and occurring particularly around key details of the offence (Reference Antaki, Richardson, Stokoe and Willott2015a: 24). In another study from the same year, they found that interviewees varied in their ability to resist the implications of probing questions (Reference Antaki, Richardson, Stokoe and Willott2015b), implications that can later seriously undermine the complainant’s case in court.

This question of police practice and the degree to which it limits interviewees’ ability to answer effectively was revisited a few years later in relation to the knotty concepts of ‘truth’ and ‘lies’ (Richardson et al., Reference Richardson, Stokoe and Antaki2018). While noting some positive findings regarding police interview practice, they also observed practices that went against ABE guidelines. These included eliciting confirmations rather than demonstrations of understanding of ‘truth’ and ‘lies’, asking for multiple demonstrations (implying that the first was not sufficient), and/or requiring demonstrations of understanding later in the interview (implying that something up to that point had suggested a lack of understanding).

A factor that comes up repeatedly in their research is the balance between institutional, forensic, and communicative goals. Antaki et al. (Reference Antaki, Richardson, Stokoe and Willott2015c) looked at the seemingly unempathetic responses that reports of distress in descriptions of sexual assault received from police interviewers. These ‘cold’ responses are explained in terms of the need to secure admissible evidence which cannot be seen to be prompted by the interviewer. They reached similar conclusions in their examination of rapport building (Stokoe et al., Reference Stokoe, Antaki, Richardson, Willott, Mason and Rock2020), arguing that the need to produce a forensically robust interview is prioritised. While the size of the dataset for these studies is a limitation, the availability of data for this kind of study will always be limited, and by applying fine-grained analysis, it has been possible to obtain a range of valuable insights into police interview guidance and practice and into how these interact with a specific set of overlapping vulnerabilities.

The UK legal system considers children vulnerable by default. In 1998, following the publication of the Memorandum of Good Practice, Aldridge and Wood produced a practical guide to interviewing children. The study combines recommendations from the academic literature with illustrations from interview transcripts with 100 children aged three to twelve years. It remains a seminal study and one of only a few to draw on an extensive database of child interview data, probably because it is so difficult to obtain. Due to their classification as vulnerable participants, children are automatically entitled to various special measures in legal processes, including the mandatory presence of an appropriate adult during a police interview. But once an individual reaches the age of 18, vulnerable status is lost, along with the entitlement to support.

Annina Heini, while at Aston University, was the first to problematise the fact that this significant change in legal status occurs essentially overnight. Heini (Reference Heini2020) used Conversation Analysis – a fine-grained, data-driven examination of transcribed interaction – to examine police interviews with suspects aged seventeen and eighteen in order to explore the language and interactional dynamics at either side of this fine boundary. Findings illustrate both consistency and differences across the age gap. For instance, suspects at both seventeen and eighteen years of age used the term ‘kid’ as a self-reference as a means of invoking the characteristic of vulnerability, but police interviewers would refer to ‘adults’ outside of the interview only when questioning seventeen-year-olds, a marker of their treatment of eighteen-year-old suspects as ‘sensible adults’ versus ‘naive children’. This work also found that seventeen-year-old suspects routinely interpreted interviewing officers’ non-verbal responses to their accounts as conveying scepticism or disbelief, and would reactively seek corroboration from their appropriate adult, for example, ‘didn’t I tell you that, granddad?’ (Reference Heini2020: 270); a strategy indeed unavailable to the eighteen-year-olds.

Likely due to the additional ethical considerations involved in studying the effect of any form of vulnerability in the legal system, studies in this area remain limited. Given the formative impact that these early interactions with the legal system can have on young people, and the way in which additional vulnerabilities can impact experiences and the ability to operate within the legal system more widely, we hope that researchers continue to find ethical and responsible ways to expand this area of research.

Studies of Police Interviews as Text

In addition to the research that has focused on the interactants in legal contexts, another branch of research in the UK has examined police interview documents as ‘text’, focusing on their production and use, and the ‘textual travel’ (Heffer et al., Reference Heffer, Rock and Conley2013) that they undergo from spoken word to a text with legal/evidential value. This research draws on notions of entextualisation (the creation of a text), intertextuality (how texts relate to one another and the people or institutions that create them), and recontextualisation (the effect of moving a text produced in one context to another) to track this ‘travel’ of texts.

The use of texts has been examined at various stages of interaction with the police. Looking at the role of texts in police interviews, Rock (Reference Rock, Coulthard and Johnson2010) examined the written texts which feed into the interview (most notably the Caution, and how its form, meaning, and function are drawn on by officers), and the written texts that are produced as a result of the interview (the ‘entextualised’ statement). Citing an earlier study (Rock, Reference Rock2001), she outlines the four-stage process by which information is elicited, probed, checked and formulated into an account, and draws on illustrative examples to demonstrate how this is done. In Rock (Reference Rock, Candlin and Crichton2013a), the process of textual travel is outlined in more detail with a full range of the functions that intertextuality and recontextualisation enable.

Rock (Reference Rock2017) revisited witness statements yet again to look at the ‘four discursive means’ that police officers draw on when producing them (writing aloud, proposing wordings, reading back what’s been written, and metatextual references). She labelled these discursive means ‘Frontstage Entextualisation’, the process by which the text is co-produced with the active involvement of the interviewee (Reference Rock, Heffer, Rock and Conley2017: 437) and in a way that makes apparent the text’s production. Rock (Reference Rock, Mason and Rock2020) went on to explore further this idea of drawing attention to the process of entextualisation through transmodal metalanguage in police-witness interviews, and how this metalanguage, ‘can be seen to give laypersons at least some insight into the texts written for them and their motivations’ (Reference Rock, Mason and Rock2020: 324)

The texts that are produced in police interviews may go on to be used in subsequent parts of the legal process, especially in court. Kate Haworth, a former barrister and alumnus of the University of Nottingham and, until recently, one of the longest-serving members of the Aston Institute for Forensic Linguistics, completed her doctoral thesis on the role of police interview discourse in the judicial process (Reference Haworth2009), and this has remained a key research focus. Like Heffer and Rock, she has also looked at the idea of textual travel (Reference Haworth, Coulthard and Johnson2010, Reference Haworth2018, Reference Haworth, Coulthard, May and Sousa-Silva2020) observing significant dangers of ‘distortion and contamination unintentionally built in’ to the collection and use of verbal evidence, and contrasting this with the meticulous preservation practices that are applied to physical evidence (Reference Haworth2018: 428). She highlighted a number of problematic elements in the process and the questionable underlying assumptions, such as that the written text is an accurate representation of (and so can be used unproblematically as a re-presentation of) the spoken interview. She also problematised the assumptions that a consistent account is the same as an honest account and that interview responses are elicited in a free and non-coercive way (Reference Rock, Mason and Rock2020: 147–148). She noted that, despite audio recordings being technically the evidence, this evidence is not simply typically retextualised as a written transcript (a process which we have already seen is potentially problematic), but that, rather than play the audio recording in court, this transcript is typically retextualised back into speech, with the ‘script’ being ‘performed’ in the court by a lawyer and a policeman. Each step introduces further potential for distortion and inaccuracy. Haworth’s Reference Haworth2018 paper looked in depth at the processes of production and reproduction of the interview, focusing on audibility, transcription, editing and courtroom presentation, while in 2020 she examined the same interview context, but this time with a focus on the effect of format, context, and audience.

Johnson (now May) (Reference Johnson, Conley, Heffer and Rock2013) explored some of the same issues regarding intertextuality and recontextualisation to look at the embedding of extracts from transcripts of interviews with Harold Shipman both in his trial and in the media coverage around it. She looked at how the prosecution introduced snippets of these interviews in their opening arguments to set up the case for there being inconsistency between accounts and how they were able to ‘ventriloquize’ (Tannen, Reference Tannen2007: 22) Shipman by embedding his voice in the presentation of their case, and thereby draw, selectively and evaluatively, on elements of his narrative to highlight points in their case against him.

Finally, by means of a corpus analysis of the transcripts, Johnson was able to highlight the use of a range of very negative evaluative collocates in order to ‘build up a semantic picture of a guilty defendant in denial and distress’ (Reference Johnson, Conley, Heffer and Rock2013: 158). Like Haworth (Reference Haworth, Coulthard and Johnson2010, Reference Haworth2018, Reference Haworth, Coulthard, May and Sousa-Silva2020), Johnson commented on the unusual practice of performing transcripts in court (despite the availability of the audio recordings, which could have been played) and in doing so ‘[making] the defendant speak for the prosecution’ (Reference Johnson, Conley, Heffer and Rock2013: 166). Finally, she commented on the inclusion of extracts from the interview transcripts in the judge’s summing up, and on the powerful institutional effect of embedding the interview at every stage of the legal process.

Through linguistic analysis, then, it has been possible to bring attention to the ways in which texts are produced and reproduced, and the communicative and institutional purposes that these textual travels serve.

Improving Police Practice

Reflections on Working With the Police in the UK Context

In an ideal world forensic linguists would collaborate with professionals in all branches of the legal system to improve the delivery of justice. But in the UK, linguists have had most success in working with the police, partly because, unlike in many countries, the police have welcomed collaboration with academics. There have always been graduates in the upper ranks of the police force, and the proportion has gradually increased until, in 2020, policing in the UK became a totally graduate profession, so police officers do not see academics as a threat and indeed have welcomed academic input. As a consequence, we as linguists have been able to have a positive impact in the areas outlined above.

Introduction

We start this section by noting its relative shortness compared to the preceding section on Forensic Linguistics and Police Discourse, some of the reasons for which will be discussed here. The studies that have been done, however, have produced valuable insights into the ‘complex genre’ (Heffer, Reference Heffer2005) of legal trials, and other communicative tasks that engagement with court processes require.

Trial Discourse

Much of the early research into trial discourse was conducted by researchers who we have already mentioned as being instrumental in establishing forensic linguistic research at Cardiff University. In his influential study in this area, Heffer (Reference Heffer2005) analysed the English jury trial as consisting of a series of speech events: jury selection, indictment, opening address, prosecution and defence evidence, closing speeches, judges summing-up, jury deliberation, judgement and sentencing, while Cotterill (Reference Cotterill2003) added that these highly structured events involve one of two discourse modes: monologic, where one speaker is addressing the court, as in opening and closing statements by the lawyers, or the judge instructing the jury; and dialogic, where two speakers are interacting, as during the examination and cross-examination of witnesses.

A good deal of courtroom interaction is conducted by means of narrative. Stories are central to legal cases, and indeed Cotterill (Reference Cotterill2003) suggested that Labov’s (Reference Labov1972) narrative categories of abstract, complicating action, evaluation, resolution and coda can be directly mapped at a macro-level onto the structural components of the trial. Opening statements relate to abstract and orientation; witness examination and cross-examination to complicating action; closing arguments to evaluation (although examples of evaluation are in fact omnipresent throughout the trial), verdict to resolution and sentencing to coda.

As well as focusing on pragmatic and structural elements of trial discourse, researchers have also focused on the importance of lexical choices. Cotterill goes on to discuss ‘the role of strategic lexical choices in constructing the prosecution and defence narrative frameworks in their opening statements. She notes that in the O.J. Simpson trial, the prosecution’s case was that he was violent, while the defence was that his wife was manipulative and promiscuous. So, the prosecution chose words such as encounter and control, with their negative ‘semantic prosodies’ realised by collocates such as prejudice, problems, opposition, risks, and hazards, in order to portray Simpson as a violent man capable of murder. By contrast, the defence set out to defuse this image by presenting the violence as ‘verbal rather than physical’ and supported this with lexical choices with neutral or positive semantics like incident, dispute, discussion and conversation to lexicalise verbal interaction in the Simpson household. Cotterill also discusses the way that the differing storytelling abilities of the lawyers affected the jurors and their verdict. She quotes one of the jurors who praised the defence lawyers’ storytelling abilities, saying that while the defence constituted ‘a show’ with ‘power and charisma’, the prosecution ‘never knew how to present [their case], they couldn’t keep it sharp and simple’.

We saw in the discussion of police language that question type/strategy was shown to be a powerful way to control the interaction and achieve institutional goals. Looking at this question in the court context, Johnson (Reference Johnson, Coulthard and Johnson2007: 102) noted that ‘a friendly lawyer will use his institutional role to produce questioning turns that often simply require confirmation, leading their witnesses through straightforward parts of their stories’. These questions are based on information gathered from the witness’s police statement and/or a previous meeting and are essentially supportive, aimed at eliciting evidence with minimum effort. So, Johnson notes that fifteen of the first nineteen questions put to Harold Shipman by his defence barrister required only confirmation. While the aim of the friendly counsel is to establish agreement on important facts, the aim of the cross-examiner is to challenge the reliability and/or significance of those facts and to present alternatives. As Coulthard et al. note, ‘cross-examining counsel build negative evaluation into their questions to attack the defendant’s character and/or to undermine their story’ (Reference Coulthard, Johnson and Wright2017: 88).

Also focusing on cross-examination, Johnson and Clifford (Reference Johnson and Clifford2011) analysed the use of (im)politeness strategies in cross-examination using data from the David Irving versus Penguin Books Ltd and Deborah Lipstadt libel trial. They divided their analysis into examinations of direct impoliteness strategies, such as the use of interruption and taboo language, and indirect impoliteness, wherein politeness strategies are used to mask otherwise impolite or face-threatening actions such as accusing or undermining through sarcasm. They found that maintaining this utterance-level locutionary politeness allows cross-examiners to deliver challenging illocutionary meanings which are difficult to counter due, in part, to their superficial politeness.

Exchanges between lawyers and witnesses are perhaps the main type of interaction that we think of when we think of court trials. Forensic linguistic research in the UK has also explored the role of judges when communicating with the jury. Jurors are, most of the time, silent receivers of evidence, although all the talk is specifically for their benefit, but occasionally the judge addresses them directly, most of the time unproblematically. However, for the past twenty years Chris Heffer has been writing about a still-unresolved problem that judges have: how to convey the ‘burden and standard of proof’ to juries in criminal trials. The jury needs to know the degree of proof required for them to convict the defendant, but the problem is that if the jury’s understanding of the required degree of proof is too severe, they will fail to convict a guilty defendant, whereas if their understanding of the required proof is too low, then an innocent defendant could be convicted.

The legal concept was traditionally conveyed through the phrase ‘beyond (a) reasonable doubt’. Originally, this expression was based on the concept that the jury must reason or argue their way to a conclusion about which they had no doubt. However, in 1950’s England, it was clear that jurors were struggling with the phrase because the ordinary language meaning of ‘reasonable’ had changed from ‘according to reason’ to the much milder ‘sensible’ or ‘moderate’ (Heffer, Reference Heffer2006). So the then Lord Chief Justice, Lord Goddard, suggested that it would be better to tell jurors that they had to be ‘satisfied so that you are sure’. The use of the everyday word ‘sure’ seemed to have major advantages and the word has been generally used in various forms ever since to try to convey the concept, with the Judicial Studies Board settling, in 1999, on the formulation ‘the Prosecution must make you sure’.

However, there was the underlying pretence that ‘sure’ continued as an ordinary language word when, in reality, it was now actually functioning as a technical legal term, but a term which judges did not define. So, when juries actually ask for a definition, the inherent problems become plain. ‘How sure?’ they ask, assuming that legal ‘sure’, like everyday ‘sure’, is gradable, but in fact legal ‘sure’, just like the term ‘beyond reasonable doubt’ which it is being used to replace, is not gradable. An accused is either guilty or not guilty, so the jury must be ‘sure’ or not ‘sure’, not 100 per cent sure or 90 per cent sure. Heffer and Coulthard (Reference Heffer and Coulthard2022) show that the problem is ongoing.

Vulnerable Participants and Courtroom Interaction

Just as with spoken interactions with police, a strand of UK forensic linguistic research has focused on court interactions involving vulnerable participants and the particular challenges they face. As explored in our discussion of police interactions, there are several personal circumstances which, under UK law, are deemed to constitute a vulnerability – relating to age, mental capacity, nature of crime … and so on. It is recognised that these factors may impact an individual’s ability to participate in the legal process, and that certain adjustments may need to be made to accommodate such individuals.

The Youth Justice and Criminal Evidence Act of 1999 enabled special measures to be adopted for children and other vulnerable witnesses in court. These included enabling private and video-recorded interviews, the use of screens to shield witnesses, and, more unusually, the removal of barristers’s wigs and gowns in court in an attempt to make them less intimidating. While broadly welcomed, in time, these measures were also found not to suit everyone, and the Coroners and Justice Act of 2009 made it possible for children to opt out of special measures if they prefer.

Part of what linguists have been able to offer is an insight as to whether the special measures introduced actually help vulnerable participants. Aldridge-Waddon (Reference Aldridge-Waddon2019) examined the effectiveness of the delivery of the opt-out procedure for child witnesses looking to avoid special measures when giving evidence to a court. The study found that officers were not generally effective at delivering this opt-out procedure, with shortcomings identified in terms of provisions of relevant information, biases in how information was presented, and the levels of language awareness displayed.

The cross-examination of witnesses is a notoriously challenging and adversarial communicative activity. It is perhaps not surprising then that it has been the focus of several studies of courtroom interaction, particularly with vulnerable participants. In 1992, conversation analyst Paul Drew looked at cross-examination in rape cases, in particular the ways in which contrasting versions of events were offered by lawyers and witnesses. While these contrasts are present in both sets of speakers, the picture that emerges is of witness modification of the accounts presented by lawyers and strategic undermining of witness accounts by lawyers.

In 2007 Aldridge-Waddon carried out a comparison of questioning strategies in children’s initial police interviews and their cross examination in court, finding the former to be a context in which children were generally permitted to give their account, whereas the adversarial style of court interaction left little opportunity for the child to respond. Building on the work (detailed in Section 3.3) on videotaped police interviews with children and their unsuitability as Evidence in Chief, Aldridge-Waddon and Luchjenbroers (Reference Aldridge-Waddon and Luchjenbroers2007a, Reference Aldridge-Waddon and Luchjenbroers2007b, Reference Aldridge-Waddon and Luchjenbroers2017) went on to look at the questioning strategies of lawyers in cases involving sexual abuse. Their research highlighted ways in which lawyers are able to manipulate witness accounts and undermine their credibility through the use of metaphor, framing, conceptual mapping, and information smuggling, activating listener preconceptions and signalling to them how they should understand the account. Their Reference Aldridge-Waddon and Luchjenbroers2017 study looked at how many of the unhelpful stereotypes detailed in their examination of police interviews (Reference Aldridge-Waddon, Luchjenbroers, Freeman and Smith2013) are also present in trial discourse and media representations thereof.

Most recently, Wright et al.’s 2020–2023 ‘Mapping the Changing Face of Cross-Examination in Criminal Trials’ looked at cross-examination of vulnerable witnesses, comparing best practice guidance with actual practice. The project found evidence of adjustments made for child witnesses, but less so for vulnerable adult witnesses when it comes to the number of questions posed, turn length and question type (e.g., closed ‘statement’ type questions). They also make the point that ‘vulnerable’ witnesses are not a homogenous group and note that adult victims of rape and sexual assault are perhaps least supported through special measures. Furthermore, there is evidence that adjustments are not made for vulnerable defendants, as opposed to vulnerable complainants or witnesses (Cooper et al., Reference Cooper, Doak, Jackson, Saunders and Wright2024). The research also showed that some elements that are present both in trials and good practice guidelines, such as signposting topic change, could be used to disorient as well as orient vulnerable participants.

The Involvement of Lay Participants

There are other factors which, while not recognised as ‘vulnerabilities’, nevertheless, can have a significant impact on communication in legal/courtroom contexts. One such factor is the relative lack of knowledge regarding procedure and terminology that lay participants have as compared to the legal professional with whom they interact. Cuts to Legal Aid in the UK during this period have seen a massive increase in the number of people representing themselves. During this period (in 2018 specifically), the UK Supreme Court also made clear that litigants in person (people who represent themselves) would receive no ‘special treatment’ or allowances for their relative lack of legal expertise.

Tatiana Grieshofer (formerly Tkačuková) has led research in this area in the UK context for a number of years and has recently published a volume in this Cambridge Elements series on ‘Legal–Lay Discourse and Procedural Justice in Family and County Courts’ (Reference Grieshofer2024). In it she explores methodological and theoretical frameworks for analysing court discourse and ways in which they can be applied to improve communicative practices in court. This builds on other work she has done in the area of lay-legal language from (1) the nature and quality of legal advice provided by lay advisers on social media (Tkačuková, Reference Tkacukova2020; Grieshofer, Reference Grieshofer2022), to (2) judicial language and institutional roles in litigant in-person cases (Tkačuková, Reference Tkačuková2015), to (3) the issues that lay participants face when conducting cross-examination (Tkačuková, Reference Tkacukova, Coulthard and Johnson2010, Reference Tkacukova2011, Reference Tkacukova2014), and (4) linguistic complexity in UK court forms (Grieshofer et al., Reference Grieshofer, Gee and Morton2021). This latter research question was explored using corpus-driven lexico-grammatical and syntactic analyses and it highlighted issues surrounding structural complexity, missing information, and the lack of coherence. Communication and Legal Practice, a collection edited with Kate Haworth, was published in early Reference Grieshofer and Haworth2026.

Lay-legal language was also the topic of Lucia Busso’s study of lexico-grammatical features in Italian (Reference Busso2024), in which she combined a construction grammar approach with corpus analysis to identify ‘idiosyncratic’ features which represent a blend of specialist and non-specialist registers. While the majority of studies in this area look at the difficulties lay participants encounter as, in some sense, reluctant participants in its heightened linguistic registers, recent work by Griffin (Reference Griffin2023) and Griffin and Roemling (Reference Griffin and Roemling2025) looked at the proactive adoption of pseudo-legal language in court filings by anti-government conspiracy theorists in the Sovereign Citizen Movement. In a comparison with legitimate legal documents, they found that while the use of legal signifiers was ‘heightened’ in the pseudo-legal texts, and the content is at times ‘outlandish’, these authors do nevertheless manage to appropriate features of legitimate legal texts into their arguments (Reference Griffin and Roemling2025: 1429).

Interpreting

Another factor which can complicate participation in the legal system is doing so in a language that is not your first. When one hears that Mozambique has twenty-five native languages, but that the justice system works almost exclusively in Portuguese, which is spoken by less than half the population and fluently by only a sixth, it sounds like absolute chaos. But the situation in London is arguably much worse: there are some 300 languages spoken on a daily basis, but for only a small proportion of these are there qualified UK-based interpreters.

Yvonne Fowler, who used to train legal interpreters and Eva Ng, an experienced court interpreter, have investigated the problems facing non-native speakers caught up in the legal system. In fact, the British police come out well, insisting, unlike forces in many other countries, that interviews with non-natives must be conducted through an interpreter, even when the interviewee claims competence in English. As practical input to ameliorate the situation, Krzysztof Kredens (Reference Kredens2016) has run courses for some police forces on how to successfully conduct interpreter-mediated interviews and has published about interpretating problems in both police interviews (Reference Kredens2017) and other interactions with emergency services (2020).

Fowler et al. (Reference Fowler, Ng and Coulthard2012) observed that ideally, not only will the interpreter be competent in English and experienced in legal contexts, but also a native speaker of the non-native language. However, very frequently, there is no competent interpreter available and there are more than isolated cases of court clerks asking unqualified individuals – relatives and even children – to interpret for the defendant. And to complicate matters further, Fowler and Ng observed there are court personnel, judges and lawyers included, who do not understand the role of the interpreter, let alone appreciate the complexities of interpreting. And then there are unrepresented defendants, who see the interpreter as their compatriot or even their saviour and expect them to act as legal advisors. Yet another serious problem for the justice system is that normally the only permanent record of both interpreted police interviews and interpreted court-room interactions is in the language of the legal process, so the interactions in the non-native’s language are lost and thus any claimed interpreting errors are subsequently unprovable.

Sadly, as Fowler et al. (Reference Fowler, Ng and Coulthard2012) observed, the situation had got worse rather than better through attempts to save money; first, with growing frequency, the interpreter is no longer present in court, working instead by video link in order to save the cost of travelling-time and this situation is ever more frequently exacerbated when, to save even more money, the accused is also not present in court. Second, large agencies that recruit less expensive but also less qualified interpreters are increasingly being used.

Writing thirteen years after Fowler et al.’s study, following a prolonged period of cuts to public services, there seems little hope that any kind of co-ordinated government-funded support for those people who face additional linguistic barriers in the legal system in the UK will be in place any time soon. What these studies have done, and can continue to do, is identify the nature of the challenges, improve practice at a local level, and make recommendations to those in a position to affect more widespread change.

Reflections on Working With Legal Professionals in the UK Context

Forensic linguists in the UK have not had great success in their attempts to work more closely with legal professionals: solicitors, barristers, magistrates and judges. When the International Journal of Speech, Language and the Law was in the planning stage, Peter French and Malcolm Coulthard courageously invited two senior lawyers, the high-profile defence barrister Michael Mansfield and the solicitor Sir David Napley, to join the editorial board and to attend a planning meeting in the offices of the publisher, Routledge. The barrister came by bike, the solicitor in his chauffeur-driven Rolls-Royce. The barrister correctly predicted it would be hard to elicit articles from barristers, the solicitor asked, ‘What will you pay the chaps to write?’ and then concluded they would also almost certainly be unwilling to contribute.

Later attempts by the Aston Institute for Forensic Linguistics to appoint barristers, solicitors and judges to the Advisory Board, so that they could then suggest useful research areas have equally failed miserably. And, unlike in Australia, forensic linguists have had no input into the training of lawyers or judges. Even the very important work by Chris Heffer questioning the nature of the linguistic encoding of the burden of proof in criminal trials – should judges use ‘Beyond reasonable doubt’, ‘Sure’ or some other formulation – which is a central topic for the criminal justice system and about which all judges have reservations, has not been taken up by the judiciary. In 2022, Heffer and Coulthard submitted a report on this point in an Appeal Court hearing. Their points were considered by three Appeal Court Judges, but they declined to engage with what the plaintiff’s barrister described as a ‘closely argued case’, that is, that the ‘Sure’ direction is fatally flawed and that the Judiciary must not simply acknowledge the problem, but also actively engage and seek a better solution. Sadly, the relationship of forensic linguists with legal professionals continues restricted to that of hired guns – expert witnesses – rather than collaborators in the joint process of improving the delivery of justice.

Conclusion

We conclude by restating that, while there clearly have been a number of important and impactful pieces of research into UK courtroom discourse, forensic linguistics has perhaps not had the kind of impact on improving communication in the courtroom in the UK that it has had on police interview practice. As outlined in their contribution to this Elements series (Eades et al., Reference Eades, Fraser and Heydon2023), our Australian counterparts have had rather more success in this area. However, the literature summarised here details a number of key contributions that have been made, and areas in which we hope to see and indeed some areas where we can already see further developments.