arXiv subjects

Allen Riddell

Publications and source records attributed to Allen Riddell.

5 recordsLinked to original sources

arXiv2022

Reproduction and Replication of an Adversarial Stylometry Experiment

Maintaining anonymity in natural language communication remains a challenging task. Even when the number of candidate authors is large, standard authorship attribution techniques that analyze writing style predict the original author with uncomfortably high accuracy. Adversarial stylometry provides a defense against authorship attribution, helping users avoid unwanted deanonymization. This paper reproduces and replicates experiments from a seminal study of defenses against authorship attribution (Brennan et al., 2012). After reproducing the experiment using the original data, we then replicate the experiment by repeating the online field experiment using the procedures described in the original paper. Although we reach the same conclusion as the original paper, our results suggest that the defenses studied may be overstated in their effectiveness. This is largely due to the absence of a control group in the original study. In our replication, we find evidence suggesting that an entirely automatic method, round-trip translation, warrants re-examination because it appears to reduce the effectiveness of established authorship attribution methods.

cs.CL↗

arXiv2020

What Library Digitization Leaves Out: Predicting the Availability of Digital Surrogates of English Novels

Library digitization has made more than a hundred thousand 19th-century English-language books available to the public. Do the books which have been digitized reflect the population of published books? An affirmative answer would allow book and literary historians to use holdings of major digital libraries as proxies for the population of published works, sparing them the labor of collecting a representative sample. We address this question by taking advantage of exhaustive bibliographies of novels published for the first time in the British Isles in 1836 and 1838, identifying which of these novels have at least one digital surrogate in the Internet Archive, HathiTrust, Google Books, and the British Library. We find that digital surrogate availability is not random. Certain kinds of novels, notably novels written by men and novels published in multivolume format, have digital surrogates available at distinctly higher rates than other kinds of novels. As the processes leading to this outcome are unlikely to be isolated to the novel and the late 1830s, these findings suggest that similar patterns will likely be observed during adjacent decades and in other genres of publishing (e.g., non-fiction).

cs.DL↗

arXiv2019

Common Library 1.0: A Corpus of Victorian Novels Reflecting the Population in Terms of Publication Year and Author Gender

Research in 19th-century book history, sociology of literature, and quantitative literary history is blocked by the absence of a collection of novels which captures the diversity of literary production. We introduce a corpus of 75 Victorian novels sampled from a 15,322-record bibliography of novels published between 1837 and 1901 in the British Isles. This corpus, the Common Library, is distinctive in the following way: the shares of novels in the corpus associated with sociologically important subgroups match the shares in the broader population. For example, the proportion of novels written by women in 1880s in the corpus is approximately the same as in the population. Although we do not, in this particular paper, claim that the corpus is a representative sample in the familiar sense--a sample is representative if "characteristics of interest in the population can be estimated from the sample with a known degree of accuracy" (Lohr 2010, p. 3)--we are confident that the corpus will be useful to researchers. This is because existing corpora--frequently convenience samples--are conspicuously misaligned with the population of published novels. They tend to over-represent novels published in specific periods and novels by men. The Common Library may be used alongside or in place of these non-representative convenience corpora.

cs.DL↗

arXiv2018

Reassembling the English novel, 1789-1919

The absence of an exhaustive bibliography of novels published in the British Isles and Ireland during the 19th century blocks several lines of research in sociologically-inclined literary history and book history. Without a detailed account of novelistic production, it is difficult to characterize, for example, the population of individuals who pursued careers as novelists. This paper contributes to efforts to develop such an account by estimating yearly rates of new novel publication in the British Isles and Ireland between 1789 and 1919. This period witnessed, in aggregate, the publication of between 40,000 and 63,000 previously unpublished novels. The number of new novels published each year counts as essential information for researchers interested in understanding the development of the text industry between 1789 and 1919.

cs.DL↗

arXiv2017

Evaluating prose style transfer with the Bible

In the prose style transfer task a system, provided with text input and a target prose style, produces output which preserves the meaning of the input text but alters the style. These systems require parallel data for evaluation of results and usually make use of parallel data for training. Currently, there are few publicly available corpora for this task. In this work, we identify a high-quality source of aligned, stylistically distinct text in different versions of the Bible. We provide a standardized split, into training, development and testing data, of the public domain versions in our corpus. This corpus is highly parallel since many Bible versions are included. Sentences are aligned due to the presence of chapter and verse numbers within all versions of the text. In addition to the corpus, we present the results, as measured by the BLEU and PINC metrics, of several models trained on our data which can serve as baselines for future research. While we present these data as a style transfer corpus, we believe that it is of unmatched quality and may be useful for other natural language tasks as well.

cs.CL↗