Showing papers by "University of Mannheim published in 2022"

Be smart, play dumb? A transactional perspective on day-specific knowledge hiding, interpersonal conflict, and psychological strain:

Abstract: Research on knowledge hiding, the intentional attempt to withhold knowledge that others have requested, strikingly shows its detrimental consequences. But, if it has only negative effects, why do e...

Recovery from Work: Advancing the Field Toward the Future

Abstract: Unwinding and recovering from everyday work is important for sustaining employees’ well-being, motivation, and job performance. Accordingly, research on work recovery has grown tremendously in the ...

On cross-lingual retrieval with multilingual text encoders

Abstract: Abstract Pretrained multilingual text encoders based on neural transformer architectures , such as multilingual BERT (mBERT) and XLM, have recently become a default paradigm for cross-lingual transfer of natural language processing models, rendering cross-lingual word embedding spaces (CLWEs) effectively obsolete. In this work we present a systematic empirical study focused on the suitability of the state-of-the-art multilingual encoders for cross-lingual document and sentence retrieval tasks across a number of diverse language pairs. We first treat these models as multilingual text encoders and benchmark their performance in unsupervised ad-hoc sentence- and document-level CLIR. In contrast to supervised language understanding, our results indicate that for unsupervised document-level CLIR—a setup with no relevance judgments for IR-specific fine-tuning—pretrained multilingual encoders on average fail to significantly outperform earlier models based on CLWEs. For sentence-level retrieval, we do obtain state-of-the-art performance: the peak scores, however, are met by multilingual encoders that have been further specialized, in a supervised fashion, for sentence understanding tasks, rather than using their vanilla ‘off-the-shelf’ variants. Following these results, we introduce localized relevance matching for document-level CLIR, where we independently score a query against document sections. In the second part, we evaluate multilingual encoders fine-tuned in a supervised fashion (i.e., we learn to rank ) on English relevance data in a series of zero-shot language and domain transfer CLIR experiments. Our results show that, despite the supervision, and due to the domain and language shift, supervised re-ranking rarely improves the performance of multilingual transformers as unsupervised base rankers. Finally, only with in-domain contrastive fine-tuning (i.e., same domain, only language transfer), we manage to improve the ranking quality. We uncover substantial empirical differences between cross-lingual retrieval results and results of (zero-shot) cross-lingual transfer for monolingual retrieval in target languages, which point to “monolingual overfitting” of retrieval models trained on monolingual (English) data, even if they are based on multilingual transformers.

Walk This Way!

Abstract: RDF2vec is a knowledge graph embedding mechanism which first extracts sequences from knowledge graphs by performing random walks, then feeds those into the word embedding algorithm word2vec for computing vector representations for entities. In this poster, we introduce two new flavors of walk extraction coined e-walks and p-walks, which put an emphasis on the structure or the neighborhood of an entity respectively, and thereby allow for creating embeddings which focus on similarity or relatedness. By combining the walk strategies with order-aware and classic RDF2vec, as well as CBOW and skip-gram word2vec embeddings, we conduct a preliminary evaluation with a total of 12 RDF2vec variants.

Is Constructive Engagement Online a Lost Cause? Toxic Outrage in Online User Comments Across Democratic Political Systems and Discussion Arenas

Abstract: This study is the first to simultaneously investigate country-level and platform-related context factors of toxic outrage, that is, destructive incivility, in online discussions. It compares user comments on the public role of religion and secularism from 2015/16 in four democracies (Australia, United States, Germany, Switzerland) and four discussion arenas on three platforms (News websites, Facebook, Twitter). A novel automated content analysis ( N = 1,236,551) combines LIWC dictionaries with machine learning. The level of toxic outrage is higher in majoritarian than in consensus-oriented democracies and in arenas that afford plural, issue-driven rather than like-minded, preference-driven debates. Yet, toxic outrage is lower in forums that tend to separate public and private conversations than in those that collapse varying contexts. This suggests that user-generated discussions flourish in environments that incentivize actors to strive for compromise, put relevant issues center stage and make room for public debate at a relative distance from purely social conversation.

Scientia Potentia Est—On the Role of Knowledge in Computational Argumentation

Abstract: Abstract Despite extensive research efforts in recent years, computational argumentation (CA) remains one of the most challenging areas of natural language processing. The reason for this is the inherent complexity of the cognitive processes behind human argumentation, which integrate a plethora of different types of knowledge, ranging from topic-specific facts and common sense to rhetorical knowledge. The integration of knowledge from such a wide range in CA requires modeling capabilities far beyond many other natural language understanding tasks. Existing research on mining, assessing, reasoning over, and generating arguments largely acknowledges that much more knowledge is needed to accurately model argumentation computationally. However, a systematic overview of the types of knowledge introduced in existing CA models is missing, hindering targeted progress in the field. Adopting the operational definition of knowledge as any task-relevant normative information not provided as input, the survey paper at hand fills this gap by (1) proposing a taxonomy of types of knowledge required in CA tasks, (2) systematizing the large body of CA work according to the reliance on and exploitation of these knowledge types for the four main research areas in CA, and (3) outlining and discussing directions for future research efforts in CA.

The ‘mixed bag’ of segregation—On positive and negative associations with migrants’ acculturation

Abstract: Many researchers and practitioners consider ethnic segregation in neighborhoods, or schools detrimental to migrants’ acculturation in host societies. Empirically, however, segregation is a “mixed bag” and its effects depend crucially on the investigated acculturation domain (e.g., negative for language skills, positive for well-being). As most prior studies have focused on a restricted spectrum of acculturation, a comprehensive assessment within one single study is needed to establish comparability across different acculturation domains. Among over 8,000 immigrant-background students from four countries, we investigated the association of classroom segregation, defined as opportunities for contact with natives and other migrants, with a broad spectrum of acculturation (academic, attitude-related, identity-related, social, health-related, and psychological criteria). Some findings were consistent (e.g., academic acculturation), some were contrary to prior research (e.g., social acculturation). In sum, our results shed light on the “mixed bag” of segregation and contribute to the understanding of a crucial social issue. This article is protected by copyright. All rights reserved

DBkWik++- Multi Source Matching of Knowledge Graphs

Abstract: Large knowledge graphs like DBpedia and YAGO are always based on the same source, i.e., Wikipedia. But there are more wikis that contain information about long-tail entities such as wiki hosting platforms like Fandom. In this paper, we present the approach and analysis of DBkWik++, a fused Knowledge Graph from thousands of wikis. A modified version of the DBpedia framework is applied to each wiki which results in many isolated Knowledge Graphs. With an incremental merge based approach, we reuse one-to-one matching systems to solve the multi source KG matching task. Based on this alignment we create a consolidated knowledge graph with more than 15 million instances.

Memory as a cognitive requirement for reciprocal cooperation

Abstract: Memory has evolved to guide our decisions in the present and to prepare us for future interactions with the environment. Within the social domain, memory can help to decide with whom to cooperate. This provides a unique opportunity to study memory from a functional perspective. Although several lines of research have demonstrated that many forms of reciprocal cooperation require memory, most of the research does not support the assumption of a highly specialized cheater-detection module that specifically serves to promote the detection of uncooperative interaction partners. Instead, the literature supports the flexible recruitment of domain-general guessing and memory mechanisms that serve to continuously predict the future behavior of others based on situational and person-specific factors and use violations of these expectations to update the predictive models of who can be trusted to cooperate in reciprocal interactions.

The DLCC Node Classification Benchmark for Analyzing Knowledge Graph Embeddings

Abstract: Knowledge graph embedding is a representation learning technique that projects entities and relations in a knowledge graph to continuous vector spaces. Embeddings have gained a lot of uptake and have been heavily used in link prediction and other downstream prediction tasks. Most approaches are evaluated on a single task or a single group of tasks to determine their overall performance. The evaluation is then assessed in terms of how well the embedding approach performs on the task at hand. Still, it is hardly evaluated (and often not even deeply understood) what information the embedding approaches are actually learning to represent. To fill this gap, we present the DLCC (Description Logic Class Constructors) benchmark, a resource to analyze embedding approaches in terms of which kinds of classes they can represent. Two gold standards are presented, one based on the real-world knowledge graph DBpedia and one synthetic gold standard. In addition, an evaluation framework is provided that implements an experiment protocol so that researchers can directly use the gold standard. To demonstrate the use of DLCC, we compare multiple embedding approaches using the gold standards. We find that many DL constructors on DBpedia are actually learned by recognizing different correlated patterns rather than those defined in the gold standard; we further find that specific DL constructors, such as cardinality constraints, are particularly hard to be learned for most embedding approaches.

Introducing grip force as a nonverbal measure of bilingual feelings

Abstract: Abstract Bilinguals’ emotions can vary in intensity with the language of a stimulus. Yet, extant research has somewhat surprisingly accepted inconsistent results from implicit nonverbal and explicit verbal emotion measures. To date, it is unclear if this inconsistency recurs to conceptual or methodological differences. We therefore investigated if squeezing a handheld dynamometer is a valid nonverbal, “visceral” alternative to self-reported language-dependent feelings by comparing explicit ratings to neuro-physiological emotional reactions. We replicated two pupillometry experiments inducing language-dependent emotions through sentence reading (Study 1) and listening to narrative video commercials (Study 2) of low and high emotionality in the first or second language. Pupillometry confirmed that bilinguals are more sensitive to the low-high emotionality contrast in their first than second language. Grip force (but not duration) mirrored these findings, whereas verbal ratings did not. We thus recommend grip force as a new attentional, nonverbal measure for bilingualism research.

Memory as a cognitive requirement for reciprocal cooperation

Abstract: Memory has evolved to guide our decisions in the present and to prepare us for future interactions with the environment. Within the social domain, memory can help to decide with whom to cooperate. This provides a unique opportunity to study memory from a functional perspective. Although several lines of research have demonstrated that many forms of reciprocal cooperation require memory, most of the research does not support the assumption of a highly specialized cheater-detection module that specifically serves to promote the detection of uncooperative interaction partners. Instead, the literature supports the flexible recruitment of domain-general guessing and memory mechanisms that serve to continuously predict the future behavior of others based on situational and person-specific factors and use violations of these expectations to update the predictive models of who can be trusted to cooperate in reciprocal interactions.

Structural Transformation of Occupation Employment

Abstract: We use census data to show that structural transformation reflects a fundamental reallocation of labour from goods to services, instead of a relabelling that occurs when goods-producing firms outsource their in-house service production. The novelty of our approach is that it categorizes labour by occupations, which are invariant to outsourcing. We find that the reallocation of labour from goods-producing to service-producing occupations is a robust feature in censuses from around the world and different time periods. To understand the underlying forces, we propose a tractable model in which uneven occupation-specific technological change generates structural transformation of occupation employment.

Jumping on the Bandwagon: The Role of Voters’ Social Class in Poll Effects in the Context of the 2021 German Federal Election

Abstract: Abstract Published findings of opinion polls are an important part of the political coverage before elections. Thus, researchers have long investigated whether the perceived popularity of political parties can lead to even more voters following this majority. However, empirical findings on this so-called political bandwagon effect are mixed. In the present paper, we integrate theories from political science and social psychology to explain these inconsistencies through social class as a potential moderating variable. Based on previous findings regarding consumer decisions, we hypothesized that bandwagon effects are greater among voters with lower social class. To investigate this hypothesis, we combined data from the German Longitudinal Election Study (GLES) Rolling Cross-Section 2021, which was collected over the 55 days before the 2021 German federal election, with the results of published preelection polls. Using separate multilevel models for each of the parties, we found no evidence for bandwagon effects. Only for the Social Democratic Party were poll results related to voting intentions assessed on the following day, suggesting that polls might have contributed to the party’s electoral success. However, there was no evidence for a moderation of bandwagon effects by voters’ social class. Accordingly, we could not resolve the mixed findings in this field of research. Our results point to important open questions in research on bandwagon effects in multiparty systems as well as on effects of social class in Germany.