Making texts speak: the work of the forensic linguist.

1. Introduction

It is a myth that texts, like Finns are silent -- but like Finns they don't tell their secrets to everyone. Among the tasks of the forensic linguist are: to discover what texts are actually saying, to teach texts to express themselves better, to interpret their meaning or highlight their significance for Courts of Law and to identify the voices of their author(s).

It is now some thirty years since Jan Svartvik published The Evans Statements: A case for forensic linguistics (Svartvik 1968). In his short monograph Svartvik demonstrated that the incriminating parts of a set of four linked statements, purportedly dictated by Timothy Evans to police officers, had a grammatical style measurably different from that of the uncontested parts of the statements. This marked the birth of a new discipline; initially, growth was slow. In unexpected places there appeared isolated articles in which the author, often a distinguished linguist, analysed a disputed confession or commented on the likely authenticity of purported verbatim records of interaction or showed why an accused could not be the person whose voice was recorded on an incriminating tape-recording or identified and evaluated inconsistencies in the language which had been attributed to non-native speakers (Levi 1994a).

In these early days there was, however, no attempt to establish a discipline nor even a methodology -- the work was usually undertaken as an intellectual challenge and almost always required the creation, rather than simply the application, of a method of analysis. In the past ten years, by contrast, there has been a rapid growth in the frequency with which Courts in a number of countries have called on the expertise of linguists; in consequence, methodology is developing rapidly and a growing number of linguists are acting as expert: witnesses, some even on a full time basis (see Levi 1994b; Eades 1994). Forensic linguistics has come of age and, like other mature areas of applied linguistics, is now beginning to raise new and exciting research questions for descriptive linguistics.

2. What do forensic linguists do?

Forensic linguists in the main set out to provide answers to three questions: what does a given text "say", what does it mean and who is its author? In answering these questions they draw on knowledge and techniques derived from one or more of the sub-areas of descriptive linguistics: phonetics and phonology; lexis, syntax, semantics and pragmatics; discourse and text analysis; computational and corpus linguistics.

2.1. What does a text say?

Tape-recordings of interviews, telephone calls and conversations, often of less than satisfactory quality, now constitue important evidence in a large number of criminal trials. The first thing the Court needs to know in such cases is what was actually said -- what was the locution -- before there can be any discussion of the illocutionary value. The forensic phonetician can play a crucial role by enhancing the tape quality and then decoding crucial indistinct words and phrases. For instance, as everyone knows, there can be surprisingly little difference auditorily, in fast conversational speech, between opposite polarity pairs like can and can't even when the sound quality is good -- when a recording is of poor quality the co-operative lay listener or transcriber, trying to make sense of jumbled sounds, may "hear" one thing, where the expert, with a trained ear and the help of sophisticated equipment, will perceive something entirely different. Just one example will suffice: a suspect accused of murder with a strong West Indian accent and some dialect features, was transcribed as saying, in an interview with police officers, that "he got on a train" and then "shot a man to kill"; the forensic phonetician was able to demonstrate that the accused actually said the innocuous and contextually much more plausible "showed a man ticket"!

2.2. What does a word/phrase/sentence mean?

Goddard (1996: 251), quoting Pearce (1974: 1), observes that "about 40% of the work of Australian and English courts requires a ruling on the meaning of an expression in a piece of legislation". Paradoxically, or perhaps not, because the determination of meaning is so central to their function, Judges are reluctant to accept linguists as expert witnesses on meaning; even when they do accept them into court to give evidence, they still reserve the right to ignore their testimony.

Kaplan et al. (1995) report on a case which went to the Supreme Court in 1994. The facts are as follows: a certain Mr Granderson pleaded guilty to a charge of destroying mail, for which the maximum sentence was 6 months imprisonment. In fact the judge decided to fine him and put him on probation for 5 years. Subsequently Mr Granderson violated his probation by being caught in possession of cocaine. In such cases the law instructs the Court to "revoke the sentence of probation and sentence the defendant to not less than one third of the original sentence". This presented the Court with a problem because, if it took "original sentence" to refer to "probation", imposing a sentence of "not less than one third" could in fact have reduced the penalty as he still had more than one third of his probation left. Therefore the judge unwillingly, but feeling he had no option, imposed a sentence of both "20 months" and "in jail", although that in fact was more than three times greater than the original maximum sentence!

Kaplan et al. (1995) argued on linguistic grounds that this interpretation was inadmissible, because the Judge had treated the phrase 'original sentence' as if it could simultaneously have two different meanings: on the one hand he had interpreted it as referring to imprisonment for the purpose of determining the type of punishment, but to the initial imposition of 5 years (of probation) for determining the length of the sentence. They pointed out that this is the linguistic equivalent of a Frenchman taking the phrase Pierre a fait tomber l'avocat to mean 'Pierre did something to a lawyer' and also 'caused an avocado to fall'. In this case the Supreme Court accepted the argument and changed the sentence.

More often, the dispute is not over what the original professional producer of a message intended an item to mean, but rather what a non-expert, the ordinary "man-in-the-street", might reasonably have interpreted it to mean. Prince (1981) reports a case where a 58 year old cement worker sued an insurance company which had refused to pay his disability pension on the grounds that he had lied when he responded to four of the questions on the original proposal form. One of these questions read as follows

Have you any impairments? ... Loss of sight or hearing? ... Loss of arm or leg? ... Are you crippled or deformed? ... If so explain

The insurance company argued that the man had lied when he replied to this question in the negative, since "he was overweight, had a high cholesterol level and occasional backaches", even though they did not dispute his counter-claim that none of these conditions had ever caused him to take time off work -- I suspect that many readers of this article, like its author, would have similarly "lied". In her evidence Prince focused on the vagueness of the word impairment; she outlined a contextual theory of meaning and argued that any co-operative reader would reasonably infer that, given the phrases that followed the word impairment, it was being used in that particular context to mean a relatively severe and incapacitating physical condition. She therefore argued that the man had not lied but rather had answered "no" "appropriately and in good conscience", (Prince 1981: 4), even if the writer of the question had intended a more inclusive meaning for the word. Even so, the judge found in favour of the insurance c ompany.

Such problems with interpretation abound. Dumas (1990) reports a case where what was at issue was whether warnings on cigarette packets in the USA dating from 1970 to 1985 in fact constituted warnings. Two of the dubious warnings she discussed were Cigarette Smoke Contains Carbon Monoxide, Quitting Smoking Now Greatly Reduces Serious Risks to Your Health. Stratrnan and Dahl (1996) looked at the intelligibility of temporary restraining orders and suggested improvements while both Shuy (1997) and Owen (1996) examined the warnings given to suspects on arrest and pointed out communicative deficiencies.

3. Who is the author?

In many cases what is in question is authorship - the linguist is asked to help decide between (usually two) conflicting claims. The phonetician will be asked to decide whether a suspect's voice is the same as that on sample tape-recordings. One of the early cases is one where Labov (1988) rescued an employee of Pan American Airways from a charge of making bomb threats against the company by demonstrating that he spoke with a New York accent, while the speaker on the tape came from the Boston area.

Often the voice identification evidence is provided by amateurs - people involved as witnesses or victims. Traditionally they were presented with a tape-recording and asked whether what they heard was or was not the voice of the criminal. However, it was successfully argued that this was a biased procedure as the witness was being subtly pressured to "confirm" a police suspect rather than identify a voice. For this reason forensic phoneticians have developed the "voice line-up" as a parallel to the long-standing identity parade and at the same time undertaken research into auditory memory. The evidence is encouraging in that it suggests firstly, that emotion in fact heightens auditory memory, secondly, that untrained ears are not significantly worse than trained ears in voice recognition and thirdly, that, if a voice is well-known to the witness, impersonation is less likely to be successful. Even so, there are certainly cases when voice disguise succeeds, particularly if done instrumentally.

3.1. The linguistic investigation of authorship

The linguist approaches the problem of authorship from the theoretical position that every native speaker has a distinct and individual version of the language they speak and write, their own idiolect. This allows for the possibility that linguists might be able to devise a method of linguistic fingerprinting -- in other words that the linguistic "impressions" created by a given speaker/writer could be used, just like a signature, to uniquely identify them. So far, however, practice is a long way behind theory and no one has even begun to speculate about how much and what kind of data would be needed to characterise an idiolect, nor how the data once collected would be analysed and stored -- indeed work on the very much simpler task of identifying the linguistic characteristics or "fingerprints" of genres is still in its infancy (Biber 1988).

In reality, the concept of the linguistic fingerprint is an unhelpful if not actually misleading metaphor, at least when used in the context of forensic investigations of authorship, because it leads us to imagine the creation of massive databanks consisting of representative linguistic samples (or summary analyses) of enormous numbers of idiolects, against which a given text could be matched and tested. In reality such an enterprise is, and for the foreseeable future will continue to be, impractical if not impossible. The value of the physical fingerprint is that each sample is both identical and complete, that is it contains all the necessary and unique information, whereas, by contrast, any text sample provides only very partial information about its creator's idiolect -- a situation compounded by the fact that many of the texts which the forensic linguist is asked to examine are very short indeed -- most suicide notes and many threatening letters, for example, are well under 200 words long.

However, the situation is not as bad as it might at first seem, because forensic texts are usually accompanied by information or clues which massively restrict the number of possible authors. Thus, the task of the linguistic detective is never one of uniquely identifying an author from millions of candidates on the basis of the linguistic evidence alone, but rather of selecting (or of course deselecting) one author from a very small number of candidates, usually fewer than a dozen and in many cases only two. In what follows I will exemplify from some of my own cases which are reported in more detail in Coulthard (1992, 1993, 1994a, 1994b, 1997).

3.2. Fabricated texts

There are many occasions when someone claims that a text is in part or completely falsified -- i.e. that the real author is different from the purported author. In this context the fabricator, whether he is creating an interview record, a confession statement or a suicide note is acting as an amateur dramatist or novelist imagining what the purported speaker/author would have produced in the same circumstances. As with any fabrication, be it bank notes or written texts, the quality of the finished product will depend on the degree of understanding that the falsifier has of the nature of what he is falsifying. Depending on the nature of the text being examined different linguistic approaches are suitable. I will give a few examples.

3.3. Spoken and written language

The first case concerns a disputed statement, in which the accused had apparently confessed to involvement in a terrorist murder. He claimed that some of what was contained in the statement had been accurately recorded, but he denied having dictated a substantial proportion of the statement, in particular the very incriminating first sentence, which he said had been invented by the police officers who were questioning him.

It is now well established within linguistics (Halliday 1989) that spoken and written language have different principles of organisation and can usually be distinguished both grammatically and lexically. As a generalisation spoken language tends to have short clauses, a low ratio of lexical to grammatical words and represents what happened as process by the use of verbs, whereas written language tends to have longer clauses, a higher lexical density and represents what happened as product by the use of nominalisations. For example, the following sentence, which the accused admitted to having said, displays the short co-ordinated clauses and very low lexical density that are typical of spoken narrative:

I drove down to the flats & I saw him up on the roof & I shouted to him & he said that he would be down in a couple of minutes.

We notice that this sentence contains thirty two words, only s even of them lexical, and is divided into five clauses, giving an average of 6.4 words per clause and a lexical density of 1.4 words per clause. The disputed first sentence, presented below, is in marked contrast consisting, as it does, of a mere three clauses which contain forty seven words, (I have conservatively treated '1987' and 'ABC' as single words), 25 of which are lexical, giving an average clause length of 15.7 and a lexical density of 8.3:

I wish to make a further statement explaining my complete involvement in the hijacking of the Ford Escort van from John Smith on Monday 28 May 1987 on behalf of the A.B.C. which was later used in the murder of three person (sic) in Newton that night.

In other words, this sentence has the high lexical density, massive subordination and frequent nominalisation -- for example statement, involvement, hijacking and murder -- typical of written texts. After I had given evidence on these features the police officer/scribe conceded that the statement may not after all have been verbatim, although he resolutely maintained that all the words had indeed been spoken by the accused, although "perhaps not in that exact order"!

3.4. Register features

Linguists have long recognised that the language that any given individual uses varies according to the contexts in which, and the topics for which, s/he is using it -- thus, at its simplest a policeman at work will have a series of linguistic options which mark him as a policeman. When a text is being falsified there is always the possibility that the real author will allow idiolectal or register features of his own usage to escape into the text.

To illustrate this I will focus on a confession statement taken from a case dating from the 1950's and made internationally famous by a film with the title Let him have it Chris. Two teenagers, Craig and Bentley were caught trying to break into a warehouse -- Craig shot and killed a policeman and Bentley, although under arrest at the time that the policeman was shot, was also convicted of murder and subsequently hanged. There is an ongoing campaign to get Bentley a posthumous pardon and the analysis outlined below was made to support this campaign. At his trial Bentley claimed that his statement was in fact a composite document, not only written down but also in part authored by police officers. I will focus on one small linguistic item -- obviously a full analysis would focus on a whole series.

3.4.1. "then"

One of the marked features of Derek Bentley's confession is the frequent use of the word then in its temporal meaning -- 10 occurrences in 582 words. This may not, at first, seem at all remarkable given that Bentley is reporting a series of sequential events and that one of the obvious requirements of a witness statement is accuracy about time. However, a cursory glance at a series of other witness statements suggested to me that Bentley's usage was at the very least atypical, and thus a potential intrusion of a specific feature of policeman register deriving from a professional concern with the accurate recording of temporal sequence.

To test this hypothesis I created two small corpora, the first composed of three ordinary witness statements, one from a woman involved in the Bentley case itself and two from men involved in another unrelated case, which totalled some 930 words of text, the second composed of statements by three police officers, two of whom were involved in the Bentley case, and the third in another unrelated case, which totalled some 2270 words.

The results were startling: whereas in the ordinary witness statements there is only one occurrence of then in 930 words, by contrast then occurs 29 times in the police officers' statements, that is on average once every 78 words. Thus, Bentley's usage of temporal then, once every 58 words, groups his statement firmly with those produced by the police officers. In this case I was fortunate in being able to check the representativeness of my "ordinary witness" data against a reference corpus, the Corpus of Spoken English, a subset of the COBUILD Bank of English, which, at that time, consisted of some 1.5 million running words. Then in all its meanings proved to occur a mere 3,164 times, that is, on average, only once every 500 words, which supported the representativeness of the witness data and the claimed specialness of the data from the police and Bentley.

What was perhaps even more striking about the Bentley statement was the frequent post-positioning of the then's, as can be seen in the two sample sentences below, selected from a total of 7 occurrences in the 582 word text:

Chris then jumped over and I followed.

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

This has an odd feel because not only do ordinary speakers use then much less frequently than policemen, they also use it in a structurally different way -- for instance, in the COBUILD spoken data then I was ten times more frequent than I then; indeed the structure I then occurred a mere 9 times in the whole of the spoken sample, in other words only once every 165,000 words. However, the phrase occurs 3 times in Bentley's short statement, that is once every 194 words, a frequency almost a thousand times greater. In addition, not only does this I then structure, as one might predict from the corpus data, not occur at all in any of the three witness statements, but by contrast there are 9 occurrences in one single 980 word police statement, as many as in the entire 1.5 million word spoken corpus. Taken together the average occurrence in the three police statements is once every 119 words. Thus, the structure I then does appear to be a feature of policeman's (written) register. More generally, it is in fact the structure Subject (+Verb) followed by then which is typical of policeman's register -- it occurs 26 times in the statements of the three officers and 7 times in Bentley's own statement.

3.5. Contextual variation

The basic facts of the next case are as follows: an armed robbery took place at a sub-branch of the Halifax building society. A man with a record of previous, though less serious, offences was arrested on suspicion. When questioned he denied any involvement; he admitted to having been in the area at the time, but claimed that he had left the immediate vicinity some ten to fifteen minutes before the robbery and gone to a "bookies" not very far away where he spent the rest of the afternoon.

The police invited him to go out in a police car in order to show them his previous day's itinerary. As they drove, one officer wrote down notes, about where they went and what was said during the journey, on both sides of a single sheet of lined paper secured to a clipboard resting on his knee. At trial the accused claimed that the record of the car journey as presented to the Court had been substantially altered; indeed he asserted that while the original text produced in the car had been written on alternate lines, the spaces between had been filled in with incriminating text after he had signed it.

Analysis showed that there were significant linguistic differences between the claimed authentic and the disputed parts of the text. From the authentic parts it was possible to derive a "note-taking grammar" with the following features: verbs without subjects; an absence of definite and indefinite articles; verbless clauses; verb forms restricted to the present imperative or simple past:

"Pointed out Halifax"; "Left back to P(olice) S(tation)";

By contrast, the claimed additions were characterised by: verbs with explicit subjects; the use of definite articles; clauses with verbs; the use of continuous and future tenses; the occurrence of subordinate clauses of time and place; modality:

"where the woman saw me"; "where I spent some of the money"; "I might have looked";

Thus it was possible to identify two markedly different styles arguably appropriate to the two different contexts of composition, the first on a knee in a moving car, the second on a desk in the relative tranquillity of a police station. The police initially denied this claim, but later conceded that some of the text, at least although not as much as the accused claimed, had indeed been produced later - however, they still resolutely insisted that all the text had been written before the record was signed by the accused. The Appeal Court judges accepted that there was a problem with the text, but they refused to rule that this was a case of deliberate falsification, with all that implied - instead they chose simply to ignore the material it contained (Coulthard 1997).

3.6. Vocabulary choice

A topic that is of great interest to layman and professional alike is plagiarism or the theft of text. Johnson (1997) outlines an approach which points the way towards an automated test for plagiarism. Johnson takes three student essays, all written on the same topic, which on a first reading seemed to have been produced collaboratively. The student authors denied collaboration and counter-suggested that the similarity came from their having answered the same question and referred to the same source texts. (The opening sentences of the three essays are presented below side by side as Figure 1 -- items in bold occur identically in another essay, those in italic are closely paraphrased.)

Johnson selected three other essays from the same batch for comparison. The initial suspicion had come from noting a great similarity between a few phrases, but often the skilful plagiarist will alter crucial words while maintaining the structure or alter the structure while maintaining the vocabulary. Given that content is carried essentially by the lexical vocabulary, Johnson chose to concentrate on that and set out to examine the degree of overlap. She focused on the opening paragraph(s) of the essays, roughly the first 500 words. What Johnson found was that the three randomly selected essays shared only 13 lexical words, items like policy, school, bilingual, multilingual, language(s) children, which were central to the question set -- "Discuss the kind. of policy a primary school should have towards bilingualism and multilingualism". Together the occurrences of these 13 shared lexical words constituted some 19% of the total lexical tokens in the three unrelated essays; by comparison, the suspect group sha red 74 lexical words whose occurrences accounted for almost half (49,3%) of all the lexical tokens.

A second analysis concentrated on the unique vocabulary -- insofar as each essay had unique content one would expect that content to be expressed through unique lexis -- sure enough each of the three non-suspect essays was found to have unique lexical vocabulary making up between 54% and 60% of the total -- by contrast two of the suspect essays had only 17% and 15% unique vocabulary; the third at 39%, contained more, but still considerably less than the independent essays. On being confronted with these results

...the writer of text 3 [the less similar of the two suspected of copying] admitted that collaboration was such that she could no longer say that the piece was independently written. The writers of texts 1 and 2 strongly denied plagiarism, although it transpired that text 1 was completed first and a draft of some of the text was seen by the other writers (and actually taken away on paper, says the writer of text 3). Furthermore, the writer of text 1 typed and corrected text 3 for its writer. Even so, no admission of collaboration was made (Johnson 1997: 233).

Despite the continuing denial these results seem to confirm that an analysis of shared vocabulary is one fruitful way of getting at shared content.

4. Concluding remarks

As I said at the beginning of this article forensic linguistics is still developing its methodology, but it is now in a position to move towards the creation of a battery of computerised measures which will provide the forensic linguist with an initial profile of the style of both the questioned text(s) and the authenticated samples of the candidate authors. Results so far suggest that the following measures are useful ones to begin with.

i) lexical density

As we saw earlier in the Northern Ireland terrorist case, lexical density can be used to distinguish spoken and written genres, although as Stubbs (1996) demonstrates the relationship is not as simple as Halliday had implied.

ii) lexical novelty

One of the interesting discoveries of computational linguistics is that in any text or corpus of whatever length, be it 150, 150,000 or even 150,000,000 words long, roughly half of all the words (types not tokens) occur only once. However, there are sometimes interesting individual deviations from this norm: on the one hand Winter (personal communication) has shown that a highly articulate speaker/writer may use a disproportionately, and therefore distinctively, large percentage of once-only words, while on the other hand the disputed Northern Ireland statement analysed briefly above, whose main function is to convict by frequent reference to and repetition of the main features of the crime, displays an unusually low proportion of lexical novelty. Early studies (Woolls -- Coulthard 1988) suggest that a measure of 'lexical richness based on the frequency of 'once only' lexical items may be a way of distinguishing authorial style at last in some cotexts.

iii) collocation 

A concordancing programme allows the investigator to examine not simply the frequencies of individual words like then, but also frequent and/or idiosyncratic collocations and colligations. Thus in the Bentley case the occurrences of then I were much more significant than the frequency of then alone, while in a suicide/murder case currently going to appeal it is the frequent co-occurrence of the items cause and pain with heartache and suffering which is potentially significant.

iv) stylistic structures

Occasionally an examination of frequent words or collocations throws up distinctive stylistic structures. Thus, in this same suicide case a list of the stylistic preferences of one of the candidate authors includes the frequent use of the paired-item structure "X and Y", e.g., hurt and suffering, hurt and pain, hurt and greaf (sic), pain and heartache, lied and cheated, physically and emotionally.

Any branch of applied linguistics depends essentially upon, but also raises interesting questions for, descriptive linguistics. Thus the future of forensic linguistics is inextricably linked to the development, by descriptive and in particular corpus linguists, of more sophisticated means of identifying and evaluating regularities in texts. Nevertheless, this new discipline has begun well and looks set to continue life as one of the most exciting areas of applied linguistics.

Figure 1. Openings of three suspect student essays (Johnson 1997: 214)

It is essential for all teachers to understand the history of Britain as a multi-racial, multi-cultural nation. Teachers, like anyone else, can be influenced by age old myths and beliefs However, it is only by having an Understanding of the past that we can begin to comprehend the present

In order for teachers to competently acknowledge the ethnic minority, it is essential to understand the history of Britain as a multi-racial, multi-cultural nation. Teachers are prone to believe popular myths and beliefs; however, it is only by understanding and appreciating past theories that we can begin to anticipate the present

It is very important for us as educators to realise that Britain as a nation has become both multi-racial and multi-cultural. Clearly it is vital for teachers and associate teachers to ensure that popular myths and stereotypes held by the wider community do not influence their teaching. By examining British history this will assist our understanding and in that way be better equipped to deal with the present and the future

REFERENCES

Biber, Douglas

1988 Variation across speech and writing. Cambridge: CUP

Coulthard, R. Malcolm

1992 "Forensic discourse analysis", in: R. Malcolm Coulthard (ed.), 242-257.

1993 "Beginning the study of forensic texts: Corpus, concordance, collocation", in: Michael Peter Hoey (ed.), 86-97.

1994a "Powerful evidence for the defence: An exercise in forensic discourse analysis", in: John Gibbons (ed.), 414-442.

1994b "On the use of corpora in the analysis of forensic texts", Forensic Linguistics: The International Journal of Speech, Language and the Law 1, 1: 27-43.

1997 "A failed appeal", Forensic Linguistics: The international Journal of Speech, Language and the Law 4, 2: 287-302.

Coulthard, R. Malcolm (ed.)

1992 Advances in spoken discourse analysis. London: Routledge.

Dumas, Bethany

1990 "Adequacy of cigarette package warnings", in: Judy Levi -- Anne G. Walker (eds.), 309-352.

Eades, Diana

1994 "Forensic linguistics in Australia: An overview", Forensic Linguistics: The International Journal of Speech, Language and the Law 1, 2:113-132.

Gibbons, John

1994 Language and the law. London: Longman.

Goddard, Cliff

1996 "Can linguists help judges know what they mean?", Forensic Linguistics: The International Journal of Speech, Language and the Law 3, 2: 250-272.

Halliday, Michael A. K.

1989 Spoken and written language. (2nd edition.) Oxford: OUP.

Hoey, Michael Peter (ed.)

1993 Data description discourse. London: HarperCollins.

Johnson, Alison

1997 "Textual kidnapping -- A case of plagiarism among three student texts", Forensic Linguistics: The International Journal of Speech, Language and the Law 4, 2: 210-225.

Kaplan, Jeff P. - Georgia M. Green - Clark D. Cunningham - Judy N. Levi

1995 "Bringing linguistics into judicial decision making: Semantic analysis submitted to the US Supreme Court", Forensic Linguistics: The International Journal of Speech, Language and the Law 2, 1: 81-98.

Kniffka, Hannes - Susan A. Blackwell - R. Malcolm Coulthard (eds.)

1996 Recent developments in forensic linguistics. Frankfurt: Peter Lang.

Labov, William 

1988 "The judicial testing of linguistic theory", in: Deborah Tannen (ed.), 159-182.

Levi, Judy N.

1994a Language and the law: A bibliographical guide to social science research in the USA. Chicago: American Bar Association.

1994b "Language as evidence: The linguist as expert witness in North American Courts", Forensic Linguistics: The International Journal of Speech, Language and the Law 1, 1:1-26.

Levi, Judy N. - Anne G. Walker (eds.)

1990 Language in the judicial process. New York: Plenum.

Owen, Charles

1996 "Readability theory and the rights of detained persons", Hannes Kniffka - Susan A. Blackwell -- R. Malcolm Coulthard (eds.), 279-299.

Prince, Ellen

1981 "Language and the law: A case for linguistic pragmatics", Working Papers in Sociolinguistics. 112-160. Austin: Southwest Educational Development Laboratory.

Pearce, Dennis C.

1974 Statutory interpretation in Australia. Sydney: Butterworths.

Shuy, Roger

1997 "Ten unanswered questions about Miranda", Forensic Linguistics: The International Journal of Speech, Language and the Law 4, 2: 175-196.

Stratman, James F -- Patricia Dahl

1996 "Readers' comprehension of temporary restraining orders in domestic violence cases: A missing link in abuse prevention?", Forensic Linguistics: The International Journal of Speech, Language and the Law 3, 2: 211-231.

Stubbs, Michael

1996 Text and corpus analysis. Oxford: Blackwell.

Svartvik, Jan

1968 The Evans Statements: A case for forensic linguistics. Goteborg: University of Goteborg Press.

Tannen, Deborah (ed.)

1988 Linguistics in context. New Jersey: Ablex.

Woolls, David -- Malcolm Coulthard

1998 "Tools for the trade", Forensic Linguistics: The International Journal of Speech, Language and the Law 5, 2: 1-28.