Does the century matter? Machine learning methods to attribute historical periods in an Italian literary corpus
Quantitative Approaches to Universality and Individuality in Language
Word Length Distribution and Text Length: Two Important Factors Influencing Properties of Word Length Motifs
Motifs in Language and Text
Material didáctico multimedia para la enseñanza de lingüística y fonética española [Πολυμεσικό υλικό για την εκμάθηση ισπανικής γλωσσολογίας και φωνητικής]
In this paper, we present preliminary results on how an individual's writing style persists even ... more In this paper, we present preliminary results on how an individual's writing style persists even across languages. In other words, what aspects of an individual's writing will persist irrespective of the language in which he or she writes? We argue that cognitive and social traits are likely to persist and demonstrate this by two separate analyses of bilingual corpora using the same individuals. We show that for various measures of linguistic complexity (which we consider to be a cognitive variable) and participation in specific social conventions (a social one), the correlation between scores on the two languages studied is significantly higher than would be expected by chance. We argue that this type of correlation may permit cross-linguistic authorship attribution.
International Conference on the Digital Society, Jul 18, 2021
We present a research focused on the prediction of the author's personality based on natural lang... more We present a research focused on the prediction of the author's personality based on natural language processing techniques applied to essays written in Modern Greek by highschool students. Each writer has been profiled by filling in the Jung Typology Test. In addition, personality prediction is being discussed under the general research framework of author profiling by examining the effectiveness of several stylometric features to predict students' personality types. The feature set we employed was a combination of the word and sentence length, the most frequent part-of-speech tags, most frequent character/word bigrams and trigrams, most frequent words, as well as hapax/dis legomena. Since personality prediction represents a complex multidimensional research problem, we applied various machine learning algorithms to optimize our model's performance after extracting the stylometric features. We compared nine machine learning algorithms and ranked them according to their cross-validated accuracy. The best results were obtained by the Naive Bayes algorithm. According to the personality classification based on the Jung Typology Test, the author's personality prediction accuracy reached 80.7% on Extraversion, 79.9% on Intuition, 68.8% on Feeling, 75.7% on Judging, according to the personality classification. The reported results show a competitive approach to the personality prediction problem. Furthermore, our research revealed new combinations of stylometric features and corresponding computational techniques, giving interesting and satisfying solutions to the problem of the author's personality prediction for Modern Greek.
This paper aims to explore whether cross-linguistic authorship attribution and author's gende... more This paper aims to explore whether cross-linguistic authorship attribution and author's gender identification are feasible using Machine Translation (MT) as a method to bridge the language gap. We designed a series of computational stylistics experiments to explore whether the stylometric signal survives through the MT process. We compiled an extensive blog corpus in Greek containing 100 authors, balanced in gender with 50 texts from each author. Then, we used Google's Neural Machine Translation to automatically translate each text into English. We ran several classification experiments using the Random Forest algorithm in authorship attribution and gender profiling tasks employing different feature groups in both the source language (Greek) and the machine-translated (English) corpora. Moreover, we trained models in the source language and used the texts in the target language as the unseen test set, to simulate a cross-linguistic prediction. The results showed that cross-l...
Sentiment analysis in cross-linguistic context: How can machine translation influence sentiment classification?
Digital Scholarship in the Humanities
In recent years, there has been a lot of interest in cross-language sentiment classification, as ... more In recent years, there has been a lot of interest in cross-language sentiment classification, as the research in sentiment analysis has shifted focus from English to less resourceful languages. Cross-language sentiment classification attempts to leverage the automated machine translation (MT) capability utilizing the infrastructure of languages rich in linguistic resources, mainly English, to help build sentiment analysis systems for low-resource languages. In this study, we explore how MT influences cross-language sentiment classification. To this end, we perform three different experiments, obtaining promising results. In the first experiment, we automatically translate 4,000 positive and negative reviews from English into Greek and Italian, thus obtaining labeled sentiment datasets in these languages. Then, we train a Naive Bayes classifier and compare the performance with the source dataset. In the second experiment, the translated reviews are automatically translated back into ...
Stylometric analysis-the study of the writing style of the author of a document, either to determ... more Stylometric analysis-the study of the writing style of the author of a document, either to determine his/her identity or personal characteristics-is an important problem in text analysis and information retrieval, with many important real-world applications. It is generally limited by a need for reference documents that are representative of the unknown documents to be analyzed. This paper addresses the issue of analysis with highly unrepresentative documents, and specially the question of whether elements of writing style can be shown to vary systematically with the individual irrespective of the language of writing. We identify fourteen Twitter users who post bilingually in both Spanish and English. An analysis of several standard linguistic and Twitter-specific extra-linguistic variables show both that there is a substantial amount of individual variation along these variables, but (more importantly), that the variations correlate very strongly across languages. In other words, a...
This paper presents the development of the first parliamentary corpus of written questions in the... more This paper presents the development of the first parliamentary corpus of written questions in the Hellenic Parliament. Moreover, we discuss a well-defined end-toend process that has been streamlined and optimised to produce high-quality open text data based on parliamentary documents. Based on the above methodology, a representative sample of 2,000 questions from four parliamentary periods in the Hellenic Parliament has been extracted, validated, and placed into an open data repository. Furthermore, open data production is analysed, and several degrees of freedom in its application in alternative data sets are proposed and discussed. Consequently, the authors argue that this method constitutes a transferable and scalable practice that can be used by other representative institutions for the creation and subsequent study of their open data.
Modern Greek is one of the least quantitatively studied modern European languages and the goal of... more Modern Greek is one of the least quantitatively studied modern European languages and the goal of this paper is to fill this relative void. We use the Hellenic National Corpus (HNC), which is a growing corpus that currently includes 33 million words. The corpus and all the tools used in our work were developed by the Institute for Language and Speech Processing (ILSP). In this paper we focus on three main areas: the lists of the 1000 most common words and lemmas, word length and letter frequency. We also make some comparisons with earlier work, in which we had used the previous 13 million word edition of the HNC.
The aim of this paper is to investigate the major quantitative parameters related to the definiti... more The aim of this paper is to investigate the major quantitative parameters related to the definition of the optimum text size in Modern Greek corpus development. Using the Hellenic National Corpus (HNC) (Hatzigeorgiu et al., 2000) as a reference point we estimated a number of critical statistical measures regarding feature counting in different text sizes. The results indicate that frequent linguistic features behave differently from the medium frequency and the rare ones and the text size increase do not affect them uniformly. 1.
The vast amount of electronic texts in the web facilitates the creation of megacorpora at a very ... more The vast amount of electronic texts in the web facilitates the creation of megacorpora at a very short time. However, as the corpus size increases, the complexity of its management is becoming a serious problem limiting its functionality. Furthermore, a great deal of corpus linguistics research is based on the
Internet offers technologies and means for creating an efficient and collaborative learning envir... more Internet offers technologies and means for creating an efficient and collaborative learning environment for on-line lessons. Second language acquisition research has shown that collaboration facilitates language acquisition [1] and related cognitive development [2]. In addition, it changes the structure of communication and social relationships developed in the classroom framework [3]. Moreover, it allows authentic experience in learning environments [4], and development of students’ responsibility for their own learning. Furthermore, we have noticed a positive effect in writing (advanced complexity in sentence structure) and an encouragement for enhanced production and effective linguistic feedback, via collaboration writing [5].
The aim of this study is to explore authorship attribution methods in Greek tweets. We have devel... more The aim of this study is to explore authorship attribution methods in Greek tweets. We have developed the first Modern Greek Twitter corpus (GTC) consisted of 12,973 tweets crawled from 10 Greek popular users. We used this corpus in order to study the effectiveness of a specific document representation called Author’s Multilevel N-gram Profile (AMNP) and the impact of different methods on training data construction for the task of authorship attribution. In order to address the above research questions we used GTC to create 4 different datasets which contained merged tweets in texts of different sizes (100, 75, 50 and 25 words). Results were evaluated using authorship attribution accuracy both in 10-fold crossvalidation and in an external test set compiled from actual tweets. AMNP representation achieved significant better accuracies than single feature groups across all text sizes.
[email protected] ABSTRACT The aim of this research is to describe tools and methods applied in... more [email protected] ABSTRACT The aim of this research is to describe tools and methods applied in the stylistic analysis of Modern Greek newswire corpora. We present a new method of selecting Author-Specific Words that can be used as reliable authorship discriminators. Furthermore, we explore the impact of different kinds of various other stylometric variables using Discriminant Function Analysis.
Uploads
Papers by George Mikros