Core Linguistic Concepts for Forensic Work

Register is the systematic way language adapts to the situation of use. A doctor's clinical notes are terse, passive, and full of abbreviations. The same doctor's email to a patient explains the same information in complete sentences and avoids jargon. A text message to a friend the same evening might use sentence fragments and emoji. All three are the same person, in the same language, but the texts look radically different.

The linguist M. A. K. Halliday described register variation along three dimensions: field (what the communication is about), tenor (the relationship between the participants), and mode (the channel and form, written versus spoken, formal versus informal). Each of these dimensions independently shapes the language that appears. A mismatch on any one of them is enough to make a comparison unreliable.

Register also interacts with medium. Written language typically has more subordinate clauses, longer sentences, and more varied vocabulary than speech. When a spoken statement is typed up by a third party, even without deliberate alteration, the transition from spoken to written mode introduces register features that were not in the original utterance. This is one of the reasons Svartvik's analysis of the Evans statements was so significant: the shift in register within the statements was a textual fact, not just an impression, and it had to be explained.

The concept of the idiolect is appealing as a forensic tool precisely because it seems to promise what forensic science always wants: a way to tie evidence uniquely to one person. Everyone's language is shaped by their particular history of exposure: the region they grew up in, the schools they attended, the registers they habitually use, the books they have read, the errors they absorbed from people around them and never corrected. The combination of all those influences is unique at the level of the individual's experience.

The problem is that uniqueness of experience does not translate into uniqueness of output in a forensically useful sense. Language is a shared system. Two people from the same city, similar age, similar education, and similar reading habits will have substantially overlapping idiolects. The features that distinguish them, the particular combination of vocabulary preferences, minor syntactic habits, and spelling choices, may be real, but they appear at frequencies too low to be statistically discriminating from a sample of two or three pages of text.

The honest position in court is that idiolectal evidence supports a claim of consistency or inconsistency between two texts, and that the strength of that claim depends on how many features are compared and how rare those features are in a relevant population. What it rarely supports is a claim that only one person on earth could have produced the disputed text. That is the individualisation standard of DNA or friction-ridge evidence, and language, as currently understood and analysed, does not meet it.

Dialect is the group-level dimension of variation. Where an idiolect is what makes one speaker's language distinct from another individual, a dialect is what makes speakers from a particular region or social group systematically different from speakers elsewhere. Dialects differ in phonology (how words sound), vocabulary (what words are used for common concepts), grammar (which syntactic constructions are the default), and orthography (regional spelling conventions in informal writing).

Dialect analysis is used for linguistic profiling: characterising an unknown author's likely background when there is no identified suspect. A ransom note with consistent use of vocabulary, spelling, and grammatical features associated with a specific dialect community gives investigators a characterisation they can use to guide their search, even if it cannot name a specific person. The important caution is that dialect features are suppressed by formal register, so a highly educated author writing in a deliberately formal style may not show dialect features even if their everyday speech is strongly marked.

The corpus turn in linguistics, the shift from analysing single constructed examples to systematically studying large collections of natural-language data, is what gave forensic linguistics its most reliable methodological tools. A corpus is simply a principled collection of texts compiled for analysis. Principled means the collection is defined by explicit criteria (all of the suspect's work emails from a particular period; all published novels in a genre; all police interview transcripts from a particular force) rather than by convenience or cherry-picking.

The power of corpus analysis in forensic work comes from frequency. A feature used once in a disputed text can be coincidence. The same feature appearing at twice the rate in a suspect's known writing versus a comparison corpus, consistently across multiple text samples, begins to constitute evidence. The statistical question is whether the observed difference in frequency is larger than would be expected by chance, and if so, by how much.

• Known samples corpus: texts confirmed to have been produced by the suspected author. Should be large enough for reliable frequency estimates and should match the register of the disputed text.
• Comparison corpus: a reference body representing the general population of writers who might plausibly have produced the disputed text. This is the baseline against which idiolectal features are evaluated.
• Disputed text: the text whose authorship is in question. Features are extracted and their frequencies compared against both the known samples and the comparison corpus.

A text is not just a sequence of sentences. It is organised above the sentence level into larger structural units: paragraphs, episodes, moves, turns. Discourse structure is the study of that organisation. In ordinary conversation, speakers take turns, repair misunderstandings, use specific moves to open and close topics, and hedge or confirm claims in patterned ways. In written texts, arguments proceed through stages, narratives follow conventional episode structures, and documents in particular genres have recognisable formats.

Discourse structure is forensically relevant in several ways. In analysis of police interviews, the structure of questions and answers can reveal whether an interviewer was using suggestive or leading questions that shaped the suspect's responses. In confession analysis, the structural organisation of a statement, whether it reads as a coherent first-person narrative or as a series of answers to implicit questions, can indicate whether the text was produced by the named speaker or shaped by another party.

Discourse connectives, words like 'then', 'so', 'but', 'because', are a specific aspect of discourse structure that has proved productive in forensic analysis. These are the hinges between clauses and sentences. Their frequency and their positioning within a text are partly habitual and partly register-dependent, making them useful markers in authorship comparison, particularly because they are processed below conscious attention.

Style in linguistics is not about literary elegance. It is the sum of the habitual choices a person makes at every level of language: which word to choose when two options are available, whether to use active or passive constructions, how long sentences typically run, whether argument is organised deductively or inductively, how hedges and intensifiers are distributed. Most of these choices are made without awareness, and that is precisely what makes them forensically useful.

The forensic significance of style has a practical ceiling. Style is variable: the same person's style shifts across registers, across time, and in response to audience. A writer's formal style in the 1990s may differ from their current informal style in ways that would reduce similarity measures between the two. A person who has read extensively in a particular genre tends to absorb stylistic features from that genre, blurring the distinctiveness of their idiolect in that register.

• Deliberate style disguise: a sophisticated author can consciously alter surface vocabulary, punctuation habits, and sentence length. What they cannot easily alter is function-word frequency or discourse structure, because these operate below conscious awareness.
• Style borrowing: a forger attempting to mimic another person's style tends to over-apply the features they consciously notice, producing an exaggeration of the target style rather than a faithful replica. This over-application is itself a detectable pattern.
• Cross-genre inconsistency: features that are highly distinctive in one genre may be unremarkable in another. A stylistic feature is only evidentially significant if it is rare in the relevant comparison population.

Taken together, register, idiolect, dialect, corpus methods, discourse structure, and style form an integrated set of tools. None of them is independently conclusive. Used together, systematically and transparently, they allow a forensic linguist to make evidential claims that have a known basis, a testable method, and explicit limits. That combination is what distinguishes forensic linguistics from what a careful but untrained reader would say if asked to compare two texts.