The social sciences and digital humanities have recently adopted the machine learning technique of topic modeling to address research questions in their fields. This is problematic in a number of ...ways, some of which have not received much attention in the debate yet. This paper adds epistemological concerns centering around the interface between topic modeling and linguistic concepts and the argumentative embedding of evidence obtained through topic modeling. It concludes that topic modeling in its present state of methodological integration does not meet the requirements of an independent research method. It operates from relevantly unrealistic assumptions, is non-deterministic, cannot effectively be validated against a reasonable number of competing models, does not lock into a well-defined linguistic interface, and does not scholarly model topics in the sense of themes or content. These features are intrinsic and make the interpretation of its results prone to apophenia (the human tendency to perceive random sets of elements as meaningful patterns) and confirmation bias (the human tendency to perceptually prefer patterns that are in alignment with pre-existing biases). While partial validation of the statistical model is possible, a conceptual validation would require an extended triangulation with other methods and human ratings, and clarification of whether statistical distinctivity of lexical co-occurrence correlates with conceputal topics in any reliable way.
We argue for a perspective on bilingual heritage speakers as native speakers of both their languages and present results from a large-scale, cross-linguistic study that took such a perspective and ...approached bilinguals and monolinguals on equal grounds. We targeted comparable language use in bilingual and monolingual speakers, crucially covering broader repertoires than just formal language. A main database was the open-access RUEG corpus, which covers comparable informal vs. formal and spoken vs. written productions by adolescent and adult bilinguals with heritage-Greek, -Russian, and -Turkish in Germany and the United States and with heritage-German in the United States, and matching data from monolinguals in Germany, the United States, Greece, Russia, and Turkey. Our main results lie in three areas. (1) We found non-canonical patterns not only in bilingual, but also in monolingual speakers, including patterns that have so far been considered absent from native grammars, in domains of morphology, syntax, intonation, and pragmatics. (2) We found a degree of lexical and morphosyntactic inter-speaker variability in monolinguals that was sometimes higher than that of bilinguals, further challenging the model of the streamlined native speaker. (3) In majority language use, non-canonical patterns were dominant in spoken and/or informal registers, and this was true for monolinguals and bilinguals. In some cases, bilingual speakers were leading quantitatively. In heritage settings where the language was not part of formal schooling, we found tendencies of register leveling, presumably due to the fact that speakers had limited access to formal registers of the heritage language. Our findings thus indicate possible quantitative differences and different register distributions rather than distinct grammatical patterns in bilingual and monolingual speakers. This supports the integration of heritage speakers into the native-speaker continuum. Approaching heritage speakers from this perspective helps us to better understand the empirical data and can shed light on language variation and change in native grammars. Furthermore, our findings for monolinguals lead us to reconsider the state-of-the art on majority languages, given recurring evidence for non-canonical patterns that deviate from what has been assumed in the literature so far, and might have been attributed to bilingualism had we not included informal and spoken registers in monolinguals and bilinguals alike.
In this paper, we present corpus data that questions the concept of native speaker homogeneity as it is presumed in many studies using native speakers (L1) as a control group for learner data (L2), ...especially in corpus contexts. Usage-based research on second and foreign language acquisition often investigates quantitative differences between learners, and usually a group of native speakers serves as a control group, but often without elaborating on differences within this group to the same extent. We examine inter-personal differences using data from two well-controlled German native speaker corpora collected as control groups in the context of second and foreign language research. Our results suggest that certain linguistic aspects vary to an extent in the native speaker data that undermines general statements about quantitative expectations in L1. However, we also find differences between phenomena: while morphological and syntactic sub-classes of verbs and nouns show great variability in their distribution in native speaker writing, other, coarser categories, like parts of speech, or types of syntactic dependencies, behave more predictably and homogeneously. Our results highlight the necessity of accounting for inter-individual variance in native speakers where L1 is used as a target ideal for L2. They also raise theoretical questions concerning a) explanations for the divergence between phenomena, b) the role of frequency distributions of morphosyntactic phenomena in usage-based linguistic frameworks, and c) the notion of the individual adult native speaker as a general representative of the target language in language acquisition studies or language in general.
Abstract
Quantitative approaches are gaining popularity in German legal research. The analysis of large corpora of legal text may be supported by text mining methods. In this study, we employ topic ...modeling—which aims at retrieving the “topics” of a corpus—to identify words related to certain areas of law present in the case law of the German Federal Constitutional Court (FCC). This information is then evaluated by legal experts and used to show significant content-related differences between the two most frequent types of proceedings before the FCC. Technical and somewhat unstable areas of law, such as tax law, social law, and civil service law, are significantly overrepresented in referrals for judicial review, whereas areas of law characterized by well-developed case law and judicial doctrine appear substantially more often in constitutional complaints. This insight may come as a surprise due to the fact that the Court’s material scope of review is identical in both types of proceedings. Our considerations do not, however, seem to apply to private law. Though we recognize the methodological and epistemological concerns regarding the heuristic nature of topic modeling, this study exemplifies its productive use in complementing, rather than replacing, more traditional techniques of analysis in legal studies.
The present study analyzes morphological productivity for complex verbs in second language acquisition by analyzing a corpus of German as a Foreign Language (GFL). It shows that advanced learners of ...GFL use prefix and particle verbs relatively frequently and productively but less so than native speakers do and discusses these findings in the light of different linguistic models and acquisition theories. It argues that corpus data must be evaluated against good models and that it is necessary to make the categorization decisions available as annotations.
Die korpuslinguistische Arbeit untersucht den Erwerb von Koselektionsbeschränkungen bei Lerner*innen des Deutschen als Fremdsprache in einem quasi-longitudinalen Forschungsdesign anhand des ...Kobalt-Korpus. Neben einigen statistischen Analysen wird vordergründig eine graphbasierte Analyse entwickelt, die auf der Graphmetrik Louvain-Modularität aufbaut. Diese wird für diverse Subkorpora nach verschiedenen Kriterien berechnet und mit Hilfe verschiedener Samplingtechniken umfassend intern validiert. Im Ergebnis zeigen sich eine Abhängigkeit der gemessenen Modularitätswerte vom Sprachstand der Teilnehmer*innen, eine höhere Modularität bei Muttersprachler*innen, niedrigere Modularitätswerte bei weißrussischen vs. chinesischen Lerner*innen sowie ein U-Kurven-förmiger Erwerbsverlauf bei weißrussischen, nicht aber chinesischen Lerner*innen. Unterschiede zwischen den Gruppen werden aus typologischer, kognitiver, diskursiv-kultureller und Registerperspektive diskutiert. Abschließend werden Vorschläge für den Einsatz von graphbasierten Modellierungen in kernlinguistischen Fragestellungen entwickelt. Zusätzlich werden theoretische Lücken in der gebrauchsbasierten Beschreibung von Koselektionsphänomenen (Phraseologie, Idiomatizität, Kollokation) aufgezeigt und ein multidimensionales funktionales Modell als Alternative vorgeschlagen.
The thesis located in corpus linguistics analyzes the acquisition of coselectional constraint in learners of German as a second language in a quasi-longitudinal design based on the Kobalt corpus. Supplemented by a number of statistical analyses, the thesis primarily develops a graph-based analysis making use of Louvain modularity. The graph metric is computed for a range of subcorpora chosen by various criteria. Extensive internal validation is performed through a number of sampling techniques. Results robustly indicate a dependency of modularity on language acquisition progress, higher modularity in L1 vs. L2, lower modularity in Belarusian vs. Chinese learners, and a u-shaped learning development in Belarusian, but not in Chinese learners. Group differences are discussed from a typological, cognitive, cultural and cultural discourse, and register perspective. Finally, future applications of graph-based modeling in core-linguistic research are outlined. In addition, some gaps in the theoretical discussion of coselection phenomena (phraseology, idiomaticity, collocation) in usage-based linguistics are discussed and a multidimensional and functional model is proposed as an alternative.
Die Sprache von Lerner/-innen einer Fremdsprache unterscheidet sich auf allen linguistischen Ebenen von der Sprache von Muttersprachler/-innen. Seit einigen Jahrzehnten werden Lernerkorpora gebaut, ...um Lernersprache quantitativ und qualitativ zu analysieren. Hier argumentieren wir anhand von drei Fallbeispielen (zu Modifikation, Koselektion und rhetorischen Strukturen) für eine linguistisch informierte, tiefe Phänomenmodellierung und Annotation sowie für eine auf das jeweilige Phänomen passende formale und quantitative Modellierung. Dabei diskutieren wir die Abwägung von tiefer, mehrschichtiger Analyse einerseits und notwendigen Datenmengen für bestimmte quantitative Verfahren andererseits und zeigen, dass mittelgroße Korpora (wie die meisten Lernerkorpora) interessante Erkenntnisse ermöglichen, die große, flacher annotierte Korpora so nicht erlauben würden.