To cut a long story short:an analysis of formulaic sequences in short written narratives and their potential as markers of authorship

Previous research into formulaic language has focussed on specialised groups of people (e.g. L1 acquisition by infants and adult L2 acquisition) with ordinary adult native speakers of English receiving less attention. Additionally, whilst some features of formulaic language have been used as evidence of authorship (e.g. the Unabomber’s use of you can’t eat your cake and have it too) there has been no systematic investigation into this as a potential marker of authorship. This thesis reports the first full-scale study into the use of formulaic sequences by individual authors. The theory of formulaic language hypothesises that formulaic sequences contained in the mental lexicon are shaped by experience combined with what each individual has found to be communicatively effective. Each author’s repertoire of formulaic sequences should therefore differ. To test this assertion, three automated approaches to the identification of formulaic sequences are tested on a specially constructed corpus containing 100 short narratives. The first approach explores a limited subset of formulaic sequences using recurrence across a series of texts as the criterion for identification. The second approach focuses on a word which frequently occurs as part of formulaic sequences and also investigates alternative non-formulaic realisations of the same semantic content. Finally, a reference list approach is used. Whilst claiming authority for any reference list can be difficult, the proposed method utilises internet examples derived from lists prepared by others, a procedure which, it is argued, is akin to asking large groups of judges to reach consensus about what is formulaic. The empirical evidence supports the notion that formulaic sequences have potential as a marker of authorship since in some cases a Questioned Document was correctly attributed. Although this marker of authorship is not universally applicable, it does promise to become a viable new tool in the forensic linguist’s tool-kit.

Linguistic identifiers of L1 Persian speakers writing in English:NLID for authorship analysis

This research focuses on Native Language Identification (NLID), and in particular, on the linguistic identifiers of L1 Persian speakers writing in English. This project comprises three sub-studies; the first study devises a coding system to account for interlingual features present in a corpus of L1 Persian speakers blogging in English, and a corpus of L1 English blogs. Study One then demonstrates that it is possible to use interlingual identifiers to distinguish authorship by L1 Persian speakers. Study Two examines the coding system in relation to the L1 Persian corpus and a corpus of L1 Azeri and L1 Pashto speakers. The findings of this section indicate that the NLID method and features designed are able to discriminate between L1 influences from different languages. Study Three focuses on elicited data, in which participants were tasked with disguising their language to appear as L1 Persian speakers writing in English. This study indicated that there was a significant difference between the features in the L1 Persian corpus, and the corpus of disguise texts. The findings of this research indicate that NLID and the coding system devised have a very strong potential to aid forensic authorship analysis in investigative situations. Unlike existing research, this project focuses predominantly on blogs, as opposed to student data, making the findings more appropriate to forensic casework data.

An analysis of police interview discourse and its role(s) in the judicial process

This study analyses the current role of police-suspect interview discourse in the England & Wales criminal justice system, with a focus on its use as evidence. A central premise is that the interview should be viewed not as an isolated and self-contained discursive event, but as one link in a chain of events which together constitute the criminal justice process. It examines: (1) the format changes undergone by interview data after the interview has taken place, and (2) how the other links in the chain – both before and after the interview – affect the interview-room interaction itself. It thus examines the police interview as a multi-format, multi-purpose and multi-audience mode of discourse. An interdisciplinary and multi-method discourse-analytic approach is taken, combining elements of conversation analysis, pragmatics, sociolinguistics and critical discourse analysis. Data from a new corpus of recent police-suspect interviews, collected for this study, are used to illustrate previously unaddressed problems with the current process, mainly in the form of two detailed case studies. Additional data are taken from the case of Dr. Harold Shipman. The analysis reveals several causes for concern, both in aspects of the interaction in the interview room, and in the subsequent treatment of interview material as evidence, especially in the light of s.34 of the Criminal Justice and Public Order Act 1994. The implications of the findings for criminal justice are considered, along with some practical recommendations for improvements. Overall, this study demonstrates the need for increased awareness within the criminal justice system of the many linguistic factors affecting interview evidence.

Candidate knowledge? Exploring epistemic claims in scientific writing:a corpus-driven approach

In this article I argue that the study of the linguistic aspects of epistemology has become unhelpfully focused on the corpus-based study of hedging and that a corpus-driven approach can help to improve upon this. Through focusing on a corpus of texts from one discourse community (that of genetics) and identifying frequent tri-lexical clusters containing highly frequent lexical items identified as keywords, I undertake an inductive analysis identifying patterns of epistemic significance. Several of these patterns are shown to be hedging devices and the whole corpus frequencies of the most salient of these, candidate and putative, are then compared to the whole corpus frequencies for comparable wordforms and clusters of epistemic significance. Finally I interviewed a ‘friendly geneticist’ in order to check my interpretation of some of the terms used and to get an expert interpretation of the overall findings. In summary I argue that the highly unexpected patterns of hedging found in genetics demonstrate the value of adopting a corpus-driven approach and constitute an advance in our current understanding of how to approach the relationship between language and epistemology.

On stopwords, filtering and data sparsity for sentiment analysis of Twitter

Sentiment classification over Twitter is usually affected by the noisy nature (abbreviations, irregular forms) of tweets data. A popular procedure to reduce the noise of textual data is to remove stopwords by using pre-compiled stopword lists or more sophisticated methods for dynamic stopword identification. However, the effectiveness of removing stopwords in the context of Twitter sentiment classification has been debated in the last few years. In this paper we investigate whether removing stopwords helps or hampers the effectiveness of Twitter sentiment classification methods. To this end, we apply six different stopword identification methods to Twitter data from six different datasets and observe how removing stopwords affects two well-known supervised sentiment classification methods. We assess the impact of removing stopwords by observing fluctuations on the level of data sparsity, the size of the classifier's feature space and its classification performance. Our results show that using pre-compiled lists of stopwords negatively impacts the performance of Twitter sentiment classification approaches. On the other hand, the dynamic generation of stopword lists, by removing those infrequent terms appearing only once in the corpus, appears to be the optimal method to maintaining a high classification performance while reducing the data sparsity and substantially shrinking the feature space

Comparing sentence-level features for authorship analysis in Portuguese

In this paper we compare the robustness of several types of stylistic markers to help discriminate authorship at sentence level. We train a SVM-based classifier using each set of features separately and perform sentence-level authorship analysis over corpus of editorials published in a Portuguese quality newspaper. Results show that features based on POS information, punctuation and word / sentence length contribute to a more robust sentence-level authorship analysis. © Springer-Verlag Berlin Heidelberg 2010.