Idiolect: Individual Language Variation

• Native speakers possess individual language versions, termed idiolects, influencing speech and writing choices.
• Bloch is credited with the first use of the term idiolect in 1948.
• The concept traces back to Coleridge's early nineteenth-century work, Biographia Literaria.
• Speakers develop large vocabularies over time, leading to variations in word selection preferences.
• Speakers tend to make typical and individuating co-selections of preferred words.
• The concept of a linguistic fingerprint is a misleading metaphor in the context of authorship investigations.
• Linguistic samples provide only partial information about the creator, even large samples.
• Forensic linguists often analyze short texts, like suicide notes and ransom demands.
• Texts often contain clues that restrict the number of possible authors.
• The task involves selecting or deselecting from a small number of candidate authors.

The Unabomber Case

• The Unabomber case provides an example of idiolectal coselection.
• From 1978 to 1995, the Unabomber sent bombs through the mail, targeting university and airline workers.
• In 1995, six national publications received a 35,000-word manuscript, Industrial Society and its Future, from the Unabomber along with an offer to stop sending bombs if the manuscript were published.
• The Washington Post published the manuscript, and someone recognized the writing style as his brother's.
• 'Cool-headed logician' was identified as a distinct idiolectal preference.
• The FBI found a 300-word newspaper article written a decade earlier with similar language, indicating common authorship.
• Robin Lakoff argued that shared vocabulary does not confirm authorship.
• Lakoff singled out 12 common words and phrases that could occur in any argumentative text.
• The FBI found millions of documents containing one or more of the 12 items on the internet.
• Narrowing the search to documents with all 12 items resulted in only 69 documents.
• All 69 documents proved to be versions of the manifesto.
• This was a rejection of open authorial choice, and example of idiolectal coselection

Early Interest in Authorship Attribution

• The question of authorship has been interrogated for over 2,000 years.
• Davis humorously reported an early unsuccessful attempt in Greece in the fourth century BC.
• Philosophers Heraklides and Dionysius had a falling out, and Dionysius wrote a tragedy and presented it as a recently discovered work.
• Heraklides, renowned for literary criticism, affirmed the play was written by Sophocles.
• Dionysius revealed his authorship, but Heraklides insisted on his initial judgment.
• Dionysius asked how the acrostic of the first letters of the first eight lines of the play spell out the name of his lover.
• Heraklides points out that the acrostic of the first letters of the next lines form a couplet.
• The next lines contain another acrostic, where 'Heraklides knows nothing whatsoever about literature'.

Linguistic Regularities

• De Morgan created a proposal in 1851 on how to solve authorship questions by accessing individual linguistic regularities.
• Average word length, measured by letters per word, was hypothesized to be writer-specific and virtually constant.
• Mendenhall counted the length of words from the Pauline letters, works by Shakespeare, Marlowe, and Bacon in 1887.
• Word length scores for Marlowe's later plays correlated more closely with Shakespeare's plays.
• Neither Mendenhall nor anyone else re-used or developed the method.
• Word length was one of 11 authorship markers that survived Grant's reliability tests.
• In 1938, Yule proposed average sentence length as a marker likely to discriminate well.
• A study by Winter and Woolls in 1996 combined this measure with lexical richness.
• Both of these markers were among those approved by Grant.
• Winter and Woolls were challenged by a literature colleague to distinguish between the style of two late-Victorian authors who had both written a novel.
• Provided with 1,000 words from the first five and last six chapters, along with 2,500 words from another novel.
• Potential markers that were analyzed were average sentence length and lexical richness.
• The suggestion was that the sentence boundary acts as an interaction point, or clarifies a writer's contentious point.
• Pace is a feature under the writer's direct control, relating new content to new vocabulary.
• This is amplified by elegant variation, which is the decision to relexicalize while discussing the same topic.

Honoré (1979) on Vocabulary Choice

• Honoré conducted an early study on vocabulary choice
• Honore stated the frequency of hapax legomena is a measure of vocabulary richness.
• He created a formula: 100 × log N/(1 – V₁/ V) - N is total length, V₁ is hapaxes, V is vocabulary types.
• Honoré confl ated open and closed set by measuring lexical and grammatical items together.
• The four most frequent grammatical items account for 14% of written text.
• Winter and Woolls resolved to measure only lexical richness for comparing texts.
• Results of individual 1,000 word extracts showed sentence length and lexical richness.

Winter and Woolls

• Winter and Woolls indicated a stylistic difference between the two authors being measured..
• The penultimate chapter was found to have scores comparable with those of chapters 2 and 4.
• The scores for chapter 33 seemed to fit with those that had been gathered in chapters 1, 3, and 5.
• Winter and Woolls suggested the authors collaborated on writing them due to the scores for the remaining four chapters, 28–31.
• The analysis indicated the authors also shared the same author because of the single author novel and the comparable values.
• The book was revealed to be Adrian Rome, with authors Arthur Moore and Edward Dowson, and Dowson also authored Souvenirs of an Egoist.
• Letters confirmed chapter authorship and timeline.
• Coulthard reports using the methodology to compare the style of six 1,000-word extracts.
• Clemit and Woolls investigated the authorship of eighteenth-century pamphlets, considering lexical richness and hapax dislegomena.
• The analysis assigned the texts to William Godwin.

Cusum

• Morton and Michaelson (1990) describe a statistical approach to authorship to be used in court.
• The approach revives de Morgan's claim on accurate 'habits'.
• Counter-intuitive to linguists, habits measured were nouns, words beginning with a vowel, and words consisting of two-to-four letters in each sentence.
• A CUmulative SUM of the deviation of both from the average for the whole text was created in a calculation that was made separately.
• The method itself was thus labelled, CUSUM.
• Graphs are superimposed on one another from the resulting scores.
• To exemplify the method, a text is imagined with an average of 12 words per sentence and an average of five two-and three-letter words per sentence.
• For the first three sentences of the text we have 20, 12 and 6 words long and contain respectively 7, 5 and 3 two- and three-letter words.
• Assumption: The assumption is that as habits are constant, the graphs made for multiple measurements will shadow each other.

Method Limitations

• The method, which appeared to work without a linguistic basis, was disturbing for linguists.
• Farringdon in 1996 compared the method to fingerprint analysis.
• Academic psychologists tested the method and yielded devastating results, proving the initial assumption to be incorrect.
• Morton in 1995 rejected their observations.
• Canter and Chester in 1997 set out to test the ability of CUSUM to reliably detect whether a text had a single or multiple authors.
• Results were promising as unaltered texts were classified as written by a single author; unfortunately so were all of the multiple author texts.
• Hardcastle, a Home Office trained document examiner, concluded forensics scientists turn their attention to other methods.

More Authorship Investigation Methods

• Specific analyses center on markers that permeate all sections of text, but methods can focus on smaller sections looking for differential use.
• Mosteller and Wallace in 1964 analyzed the Federalist Papers, 85 essays published anonymously in 1787-8.
• The 85 essays were written by Alexander Hamilton, James Madison, and John Jay.
• A study assumed there were unique selection preferences involved, analyzing texts and finding the items with differentiating usages between them.
• Hamilton used 'upon' more than Madison, and Madison used ‘also’ more than Hamilton.
• This analysis assigned all 12 of texts that were in question to Madison.
• The Federalist Papers have been a testing ground for the latest stylometric methods since then.
• The computer program "learns" from its mistakes because one of its models is the human brain.
• Training via a neural network has been used to discern the plays of both Shakespeare and Marlowe.
• Shakespeare adopted the play Henry VI, Part 3 from an earlier work by Marlowe.
• Studies provide evidence for potential differential usage, with Kredens indicating the difference between Smith and Morrissey in 3 out of 5 discourse markers.

Error and Mistake

• Authorship studies all work under the assumption that all speakers/writers are unique in their linguistic selections.
• The investigations have concerns with variation, both intra-speaker and inter-speaker.
• One of the distinctive features of the author's texts when the account has no spell-checker is 'teh' instead of the and '-ign' for '-ing'.
• Majority of items are checked, but imperfection causes errors.
• Corder in 1973 suggested to categorize problems of language learners by creating mistake and error categories.
• The mistakes are instances where the producer knows of the deviation and can possibly be corrected.
• Unlike mistakes, errors are when the producer has acquired a different rule from the standard system.
• Unique markers are rare, with authorship being the identification of a collection of markers.
• McMenamin found 300 style markers that has been used in some 80 authorship cases.
• Classified as: Text Format, Numbers and Symbols, Abbreviations, Punctuation, Capitalization, Spelling,Word Formation, Syntax, Discourse, Errors and Correction, and High Frequency Words and Phrases.
• McMenamin states there are two major authorship questions: looking for consistency, and ‘looking for resemblance'.
• The first will determine if a text or collection of texts have the same author while the second investigates a case where authorship is unknown.
• Qualitative traits are identified and described, while quantitative indicators are identified and then measured through relative frequency.
• McMenamin (2002:77) exemplifies the qualitative approach with a case in which the questioned author consistently spelled the name Mary Ann as two words.
• For the quantitative approach an assessment of rarity and significance is carried out.
• The results showed that the version Ca. as opposed to Ca., CA., Ca, ca and ca. occurred in 11 percent of the 686 addresses examined.

Analysing Cases

• An ideal forensic world would have a substantial amount of known text to work with.
• Forensic world = texts are unhelpfully short.
• McMenamin (2002: 181-205) examplifies a analysis with the in the 1996 JonBenét Ramsey murder case.
• McMenamin was asked to compare the ransom note with a set of writings.
• Analysis indicated a series of idiosyncrasies.
• As the samples of comparable text from the suspects was limited, McMenamin decided to elicit rich data.
• Elicited both versions where he dictated the text, and had them copy.
• He found differences between the ransom and father's note and the note and mother's, while observing the stylistic differences were consistent for pre-crime writings.
• Quantitative: number of writers in a population have the profile of variables identified in the note.

Comparing Styles

• Compared style features of ransom and corpus of handwritten texts from American Writing Project.
• Isolated six variables because they were frequent in the corpus and Mrs. Ramsey differed.
• The likelihood of the six co-occurring in the same text by chance was one in 10,000.
• Both qualitative and quantitative measures, thus, support the opinion that neither Mr nor Mrs Ramsey had written the ransom note.
• Report of a letter of similar analysis which lead to husband being found to be the writer.
• It is not unusual for the expert to use more than one approach.
• The case also illustrates two more approaches to possible multiple authorship.
• The police was there to arrest 2 teenagers, Derek Bentley aged 19 and Chris Craig, when Craig starting shooting and killing a policeman.
• Craig and Bentley were both trialed with murder, and Bentley was jointly charged.
• A trial lasted 2 days, and the two were found guilty, despite Craig shooting during the police apprehension.
• Bentley's relatives sued to drop the verdict.
• The prosecution said Derek did kill by claiming he had a gun.
• Exact 8.1 indicates some evidence as it relates to the events that happened during that time.
• Police officers were to ask no substantive questions during the event.
• Three police officers said it was an unaided monologue.

Interpreting Narrative

• One example of interpreting narrative from monologue involved what happened during the crime, an observation on their knowledge, what they thought and what had happened.
• It would have formed some meta narrative, and it most likely came from a series of clarificatory questions.
• What can be deduced with narrative questions requires an intricate process that can only be seen through clear textual evidence.
• It means that Bentley really didn’t see anything any more, according to the police report, and there were clear events that had occurred.
• It’s apparent from just the available, textual evidence, that the actual context of the situation must be analyzed due to the fact what did not, was not included.
• There would need to be evidence to prove some sort of way to have inferred to people about certain events, and a great example is Bentley telling his friends, which must have been a clear detail based on the actual event.
• Bentley’s knowledge about the gun would come out in full capacity due to the questioning by the police officer about it.
• The knowledge was presented in Extract 8.2, and if knowledge of it would need to be discussed, the loaded gun part would come in advance too.
• Bentley stated it with an explanation on the logic, sequence and information that was available to tell.
• A Corpus analyzed with register indicated how Bentley had used what information was available.

Corpus Assisted Analysis

• One of the features that was indicated was the meaning of “then", temporal meaning.
• It was apparent through looking at the witness statements.
• Bentley’s unusual usage seemed unusual at the time.
• The investigation could derive with the way of accurate temporal meaning
• With smaller portions that was used by officer and police.
• The comparative result meant there was information not said which can be stated through 78 words.
• A reference corpus had over 1.5 million running words.
• It was more remarkable to put evidence in that area with this Bentley.
• The phrase gave an odd feel, not ordinary or speaking.
• It has shown to include a structure on the verb, typically police register.
• It helps find the structure with Craig and it happened by the officer.
• All are examples from police style.
• Added support was brought forth it the police confirmed Bentley for the joint authorship and that it all was undermined with “let him at Craig!”
• The Lord Chief was able to put evidence and prove that things would be showed had the conviction been allowed.