Search

This study was funded by the German Research Foundation (DFG) in the framework of the project 'Information Density and Scientific Literacy in English – Synchronic and Diachronic Perspectives' in the Collaborative Research Center (SFB1102) with the title 'Information Density and Linguistic Encoding' (http://www.sfb1102.uni-saarland.de) and EXC 284: Multimodal Computing and Interaction (www.mmci.uni-saarland.de).

While the production of information in the European early modern period is a well-researched topic, the question how people were engaging with the information explosion that occurred in early modern Europe, is still underexposed. This paper presents the annotations and experiments aimed at exploring whether we can automatically extract media related information (source, perception, and receiver) from a corpus of early modern Dutch chronicles in order to get insight in the mediascape of early modern middle class people from a historic perspective. In a number of classification experiments with Conditional Random Fields, three categories of features are tested: (i) raw and binary word embedding features, (ii) lexicon features, and (iii) character features. Overall, the classifier that uses raw embeddings performs slightly better. However, given that the best F-scores are around 0.60, we conclude that the machine learning approach needs to be combined with a close reading approach for the results to be useful to answer history research questions.

Pretrained language models like BERT have advanced the state of the art for many NLP tasks. For resource-rich languages, one has the choice between a number of language-specific models, while multilingual models are also worth considering. These models are well known for their crosslingual performance, but have also shown competitive in-language performance on some tasks. We consider monolingual and multilingual models from the perspective of historical texts, and in particular for texts enriched with editorial notes: how do language models deal with the historical and editorial content in these texts? We present a new Named Entity Recognition dataset for Dutch based on 17th and 18th century United East India Company (VOC) reports extended with modern editorial notes. Our experiments with multilingual and Dutch pretrained language models confirm the crosslingual abilities of multilingual models while showing that all language models can leverage mixed-variant data. In particular, language models successfully incorporate notes for the prediction of entities in historical texts. We also find that multilingual models outperform monolingual models on our data, but that this superiority is linked to the task at hand: multilingual models lose their advantage when confronted with more semantical tasks.

For the investigation of variation in language use, besides the extra-linguistic context it is also essential to consider variation given the more local linguistic context. This allows us to contribute towards a better understanding of why there is variation. In particular, we are interested in how variation helps to modulate the information content of linguistic units, achieving optimization effects for efficient communication (Jaeger and Levy, 2007; Piantadosi et al., 2011). I will present how we can detect and analyze variation with data-driven methods across different dimensions of variation (i.e. the extra-linguistic context such as time, demographics, registers). We pursue answers to questions such as: how do scientific language change over time and possibly become more optimal for scientific communication; how gender and social class differences impact language use within formal vs. informal contexts; do literary research articles experience an effect of “scientization” over time, e.g. by becoming more standardized? Possible effects of selected linguistic choices on language use are traced at the level of local linguistic context. In accordance with communicative accounts, we focus on tracing effects of variation that modulate the information content transmitted. I will also spend some time on discussing how integrating diverse types of evidence (e.g. corpus-based and experimental) helps to work towards a more comprehensive understanding of variation in language use.

In this thesis, we elaborate characteristics of evaluative meaning of different scientific disciplines and trace their diachronic linguistic evolution. A main focus lies on newly emerged disciplines, such as computational linguistics, which emerged through contact between two other disciplines, such as computer science and linguistics. Here, we consider (1) whether these newly emerged disciplines have created characteristics of their own over time, showing a process of diversification, and (2) whether they have also adopted characteristics from their disciplines of origin, reflected in a linguistic imprint, and if this might have changed over time. The newly emerged disciplines considered are computational linguistics, bioinformatics, digital construction and microelectronics, which have emerged through contact between computer science and a further discipline (linguistics, biology, mechanical engineering, and electrical engineering, respectively). In terms of theory, this work is grounded in a linguistic theory rooted in sociolinguistics, Systemic Functional Linguistics (SFL; Halliday, 2004), which with its functional perspective on language allowed us to position evaluative meaning within a linguistic theory and to create a model of analysis to trace choices made in the semantic system on the level of lexico-grammar. Moreover, its notion of register, concerned with functional variation, i.e. variation according to language use, combined with the sociolinguistic perspective made it possible to compare the linguistic choices made according to different social contexts, to which the disciplines belong. This allowed us to trace register diversification processes and registerial imprint of evaluative meaning across disciplines. In terms of methods, we apply classification as a data mining technique, taking a macro- and micro-analytic perspective (cf. Jockers, 2013) on the results. Doing so we gain insights on the degree of diversification and imprint (macro-analysis) and the kind of diversification and imprint (micro-analysis). Studies so far have considered either the macro- or the micro-analytic perspective. By considering both, we are able to investigate generalizable trends as well as detailed linguistic characteristics of evaluative meaning across disciplines and time. The approach presented in this thesis draws its strength from being grounded in a linguistic theory, which proved to be extremely useful in defining and testing hypotheses and interpreting results. Moreover, an empirical analysis of evaluative meaning across disciplines and time was possible by combining corpus-based methods with data mining techniques. In der vorliegenden Dissertation werden Bewertungscharakteristiken verschiedener Wissenschaftsdisziplinen erarbeitet und ihre diachrone linguistische Entwicklung untersucht. Ein Hauptfokus liegt auf in neuerer Zeit entstandenen Disziplinen (z. B. Computerlinguistik), die sich durch Kontakt zwischen zwei anderen Disziplinen gebildet haben (z. B. Informatik und Linguistik). In diesem Zusammenhang wird erforscht, (1) ob diese neu entstandenen Disziplinen diachron ihre eigenen Charakteristiken entwickeln und somit einen Diversifikationsprozess aufzeigen und (2) ob sie auch Charakteristiken der Ursprungsdisziplinen übernehmen und somit eine linguistische Prägung aus der Ursprungsdisziplin vorweisen und ob sich diese möglicherweise diachron verändert hat. Die untersuchten relativ neu entstandenen Disziplinen sind die Computerlinguistik, Bioinformatik, Bauinformatik und Mikroelektronik, die durch Kontakt zwischen der Informatik und einer anderen Disziplin entstanden sind, in unserem Fall entsprechend aus der Linguistik, Biologie, dem Maschinenbau und der Elektrotechnik. Die Arbeit basiert auf der soziolinguistischen Theorie der Systemisch Funktionalen Linguistik (SFL; Halliday (2004)). Aufgrund ihrer funktionalen Perspektive auf die Sprache war es uns möglich, das semantische Konzept der Bewertung in eine linguistische Theorie zu positionieren und ein Analysemodel zu entwickeln, um die Auswahl aus dem semantischen System auf der lexicogrammatischen Ebene nachzuverfolgen. Besonders wichtig ist hierbei auch das Registerkonzept aus der SFL, das sich mit funktionaler Variation befasst, d.h. Variation in Bezug auf den Sprachgebrauch. Die Kombination aus funktionaler Variation und soziolinguistischer Perspektive hat es erlaubt, die linguistischen Entscheidungen in Bezug auf Bewertungen, die in unterschiedlichen sozialen Kontexten (d.h. den verschiedenen Disziplinen) gefällt wurden, zu untersuchen und diese zu vergleichen. Dadurch konnten für die untersuchten Disziplinen registerspezifische Diversifikationsprozesse und Prägungen bezüglich Bewertungen ausgemacht werden. Methodisch wurde aus dem Bereich des Data Mining die Klassifikation angewandt, die es erlaubt hat, die Ergebnisse aus einer makro- und mikro-analytischen Perspektive (vgl. Jockers (2013)) zu erforschen. Dadurch konnten Erkenntnisse erlangt werden in Bezug auf den Diversifikations- und Prägungsgrad (Makro-Analyse) sowie der Art der Diversifikation und Prägung (Mikro-Analyse). Studien haben bislang entweder die makro- oder die mikro-analytische Perspektive angewandt. Durch den Einbezug beider Ebenen ist es uns gelungen, sowohl generalisierbare Tendenzen festzustellen als auch detaillierte linguistische Charakteristiken und diachrone Veränderungen von Bewertungsausdrücken in verschiedenen Disziplinen zu untersuchen. Die Stärken des in der vorliegenden Dissertation präsentierten Ansatzes liegen darin, dass er in einer linguistischen Theorie fundiert ist, die sich sehr hilfreich erwiesen hat bei der Hypothesenaufstellung und beim Testen der Hypothesen sowie auch bei der Interpretation der Ergebnisse. Darüber hinaus hat der Ansatz eine empirische Analyse von Bewertungen in wissenschaftlichen Disziplinen durch das Zusammenspiel von korpus-basierten Methoden und Techniken aus dem Data Mining ermöglicht.

This paper studies the use of NMT (neural machine translation) as a normalization method for an early English letter corpus. The corpus has previously been normalized so that only less frequent deviant forms are left out without normalization. This paper discusses different methods for improving the normalization of these deviant forms by using different approaches. Adding features to the training data is found to be unhelpful, but using a lexicographical resource to filter the top candidates produced by the NMT model together with lemmatization improves results.