Article

What is forensic linguistics?

Forensic LinguisticsLanguage in context. Evidence with limits.

Forensic linguistics applies the study of language to questions arising in investigations and legal settings. The material may be a written message, a recorded interview, a disputed document or the wording of an instruction. The analyst asks a defined question and uses methods appropriate to the material. The field is broader than identifying the writer of an anonymous letter. It includes the analysis of how language is used in interviews, the meaning of disputed expressions, communication across languages and the accessibility of legal texts. Different questions require different expertise.

Four ways to approach the material
Language question
Writingauthorship / overlap
Meaningwording / context
Interactionquestion / response
Speechrecording / comparison
Different questions require different material and methods.

Begin with the question

Consider a fictional email attributed to an employee. Several questions could arise: was its wording copied from a template, does it resemble the employee’s other messages, or is a particular phrase being interpreted reasonably? These are not the same analytical problem. A useful question is narrow enough to investigate. ‘Who wrote this?’ may need to become ‘How do the linguistic patterns in this message compare with a suitable set of known writings?’ The available data may support only a limited answer.

What language evidence can contain

An analyst might examine recurring word combinations, punctuation, spelling, grammatical choices and the way a text performs a task. A speech specialist may investigate aspects of voice and pronunciation. The relevant features depend on the question and material. No isolated spelling error is a personal fingerprint. Writers share conventions, learn from each other, copy wording and change their style for different audiences. A formal complaint should not be compared uncritically with casual messages sent years earlier.

Evidence needs context

Record how a text was obtained, whether it was edited and whether it contains quoted material. A screenshot may omit surrounding messages or change line breaks. A transcript represents an audio recording through a set of transcription decisions; it is not the recording itself. Comparison material should be suitable for the question. Time period, topic, platform, genre and editing can all affect language. An analysis that ignores those differences may mistake context for authorship.

Methods and uncertainty

Qualitative analysis studies patterns in context. Quantitative analysis measures selected features across texts. The approaches can complement each other, but numbers do not remove the need for validation or interpretation. An assessment should explain what was examined, how it was examined and what alternative explanations remain. It should distinguish observations from conclusions. A clear limitation is part of a useful analysis, not an inconvenience to hide.

Learning the field responsibly

Begin with our authorship-analysis guide, corpus exercise and glossary. Use fictional or permission-cleared material while learning. This website is an independent educational publication, not an expert-witness service, an investigative authority or the organiser of the historical summer school previously associated with this domain. For an actual case, the right expertise and legal procedures depend on the jurisdiction and question. Our guides explain concepts and help readers evaluate claims; they do not determine a person’s identity, guilt or credibility.

Three questions, three different studies

Imagine a fictional organisation receives an anonymous complaint: ‘The team keeps changing the dates. Someone should explain what happened.’ An authorship study could compare the complaint with suitable known writing. A meaning study could ask what the expression ‘changing the dates’ refers to in the surrounding correspondence. A communication study could examine how the organisation’s interview questions encourage a particular account of the incident. Each begins with a different question, even though the same document appears in all three.

The choice affects the data required. Authorship comparison needs genuinely attributable writing and relevant background material. Meaning analysis needs the wording in its communicative setting, including plausible competing readings. Interaction analysis needs a sequence of turns rather than a handful of detached quotations. Gathering the wrong kind of data can leave an analyst with many pages but little evidence relevant to the question.

A practical route from document to explanation

For a learning project, start with a source inventory. Give each fictional document an identifier, date, genre and intended audience. Record whether it is original writing or includes a quotation. Preserve the wording before making a separate copy for annotation. This simple habit makes it possible to distinguish a change in the data from a change in your interpretation.

Next, describe something observable. ‘The request uses a modal verb and a question mark’ is an observation. ‘The writer is deliberately evasive’ is an interpretation requiring more evidence. Put observations and interpretations in separate columns. Add a third column for alternatives, such as politeness, institutional convention or an answer to an earlier question. This prevents an attractive explanation from becoming the only explanation considered.

Choose a method that addresses the observation. A concordance can show surrounding uses of an expression. A frequency table can show counts under a stated rule. A sequence analysis can show how a speaker corrects an assumption. None replaces the initial question or supplies a conclusion independently of the data.

Reading a persuasive claim critically

When a news story says that a phrase ‘gave the writer away’, ask how many suitable people use it, where else it occurs and whether the passage was copied. If a report announces a percentage match, ask what the percentage measures and how the method was tested. A figure can describe a classification score, a proportion of shared features or a different quantity entirely. Those meanings cannot be exchanged without explanation.

Also ask what the analysis leaves unresolved. If only one candidate was examined, the result does not automatically exclude everyone else. If the original document is unavailable, formatting features may be uncertain. If known writing spans several genres, a difference may reflect the task. These questions make an explanation stronger by showing which parts deserve confidence.

A first study session

Read the authorship guide, then complete the fictional corpus exercise without looking at its answer table. Compare your counts with the published rules. Finally, write a five-sentence account: the question, the material, one observation, one alternative explanation and the limitation. A useful first result is often modest. Learning to explain a small result precisely builds a better foundation than trying to solve a real dispute from a few messages.

Follow the source

Sources and further reading