Gerard Lynch

Computational Stylometry: Who's in a Play?

Automatic text classification techniques are applied to the problem of quantifying strength of characterization within plays, using a case study of the works of four sample playwrights that are freely available in machine-readable form. Strong characters are those whose speeches constitute homogeneous categories in comparison with other characters--their speeches are more attributable to themselves than to their play or their author.

UCD : Diachronic Text Classification with Character, Word, and Syntactic N-grams

We present our submission to SemEval-2015 Task 7: Diachronic Text Evaluation, in which we approach the task of assigning a date to a text as a multi-class classification problem. We extract n-gram features from the text at the letter, word, and syntactic level, and use these to train a classifier on date-labeled training data. We also incorporate date probabilities of syntactic features as estimated from a very large external corpus of books. Our system achieved the highest performance of all systems on subtask 2: identifying texts by specific time language use.

Automatic Character Assignation

This article outlines a simple method for parsing an ASCII-format dramatic work from the Project Gutenberg Corpus into separate characters. The motivation for the program is a upcoming study in computational stylistics and characterization in drama. Various previous approaches involving interactive media are examined and the parser is evaluated by comparing the output to data annotated by hand and parsed automatically by the Opensourceshakespeare.org project parser. An acceptable level of accuracy is achieved, and it is identified how to improve accuracy to extremely high levels.

Universum Inference and Corpus Homogeneity

Universum Inference is re-interpreted for assessment of corpus homogeneity in computational stylometry. Recent stylometric research quantifies strength of characterization within dramatic works by assessing the homogeneity of corpora associated with dramatic personas. A methodological advance is suggested to mitigate the potential for the assessment of homogeneity to be achieved by chance. Baseline comparison analysis is constructed for contributions to debates by nonfictional participants: the corpus analyzed consists of transcripts of US Presidential and Vice-Presidential debates from the 2000 election cycle. The corpus is also analyzed in translation to Italian, Spanish and Portuguese. Adding randomized categories makes assessments of homogeneity more conservative.

Linguistically Informed Tweet Categorization for Online Reputation Management

5th Workshop on Computational Approaches to Subjectivity, Sentiment and Social Media Analysis (WASSA 2014), Baltimore, Maryland, USA, 27 June 2014

Following the Trail of Source Languages in Literary Translations

We build on past research in distinguishing English translations from originally English text, and in guessing the source language where the text is deemed to be a translation. We replicate an extant method in relation to both a reconstruction of the original data set and a fresh data set compiled on an analogous basis. We extend this with an analysis of the features that emerge from the combined data set. Finally, we report on an inverse use of the method, not as guessing the source language of a translated text, but as a tool in quality estimation, marking a text as requiring inspection if it is guessed to be a translation, rather than a text composed originally in the language analysed. We obtain c. 80 % accuracy, comparable to results of earlier work in literary source language guessing—this supports the claim of the method’s validity in identifying salient features of source language interference.

The translator’s visibility: Detecting translatorial fingerprints in contemporaneous parallel translations

Abstract We detail the results of experiments towards a fine-grained stylometric analysis, the identification of distinguishing features between contemporaneous literary translations, both parallel works and also translations of non-parallel sets of works by the same author. We examine translations of plays by the Norwegian dramatist Henrik Ibsen with the initial point of focus being the Ibsen drama Ghosts, for which there exists comparable contemporaneous translations by R. Farqhuarson Sharp and William Archer. Consequently, a number of prose translations of Russian author Anton Chekhov by Marian Fell and Constance Garnett are examined in order to validate hypotheses formed from the results of the Ibsen study and investigate possible particularities in translator’s style which may vary according to genre. By carrying out an analysis of these texts using a variety of machine learning approaches such as Support Vector Machines, Simple Logistic Regression, Naive Bayes and Decision Tree classifiers, a number of distinguishing textual features are obtained, and the relative frequency of these features in the texts are compared to their frequencies in reference corpora in order to establish which features can be attributed to stylistic choices by the translators themselves and which features may be due to influence from the source language or the topic or genre of a text. We also use the popular Delta metric from authorship attribution studies to investigate the clustering of texts based on most frequent words and a list of discriminatory terms learned in the supervised machine learning experiments. We find that common word unigrams and bigrams are the most salient features for translator fingerprinting across our two authors and four translators examined and are ultimately successful in our goal of classifying which text originated from a particular translator with accuracy measurements of over 90% on average.

Towards a robust framework for the semantic representation of temporal expressions in cultural legacy data

Date and time descriptors play an important role in cultural record keeping. As part of digital access and information retrieval on heritage databases it is becoming increasingly important that date descriptors are not matched as strings but that their semantics are properly understood and interpreted by man and machine alike. This paper describes a prototype system designed to resolve temporal expressions from English language cultural heritage records to ISO 8601 compatible date expressions. The architecture we advocate calls for a two stage resolution with a “semantic layer” between the input and ISO 8601 output. The system is inspired by a similar system for German language records and was tested on real world data from the National Gallery of Ireland in Dublin. Results from an evaluation with two senior art and metadata experts from the gallery are reported.