We address the task of detecting abusive sentences in which identity groups are depicted as deviating from the norm (e.g. Gays sprinkle flour over their gardens for good luck). These abusive utterances need not be stereotypes or negative in sentiment. We introduce the first dataset for this task. It is created via crowdsourcing and includes 7 identity groups. We also report on classification experiments.

Accounting for different degrees of formality is crucial for producing contextually appropriate language. To assist NLP applications concerned with this problem and formality analysis in general, we present the first dataset of sentences from a wide range of genres assessed on a continuous informal-formal scale via comparative judgments. It is the first corpus with a comprehensive perspective on German sentence-level formality overall. We compare machine learning models for formality scoring, a task we treat as a regression problem, on our dataset. Finally, we investigate the relation between sentence- and document-level formality and evaluate leveraging sentence-based annotations for assessing formality on documents.

We address the task of identifying euphemistic abuse (e.g. “You inspire me to fall asleep”) paraphrasing simple explicitly abusive utterances (e.g. “You are boring”). For this task, we introduce a novel dataset that has been created via crowdsourcing. Special attention has been paid to the generation of appropriate negative (non-abusive) data. We report on classification experiments showing that classifiers trained on previous datasets are less capable of detecting such abuse. Best automatic results are obtained by a classifier that augments training data from our new dataset with automatically-generated GPT-3 completions. We also present a classifier that combines a few manually extracted features that exemplify the major linguistic phenomena constituting euphemistic abuse.

We address the task of distinguishing implicitly abusive sentences on identity groups (“Muslims contaminate our planet”) from other group-related negative polar sentences (“Muslims despise terrorism”). Implicitly abusive language are utterances not conveyed by abusive words (e.g. “bimbo” or “scum”). So far, the detection of such utterances could not be properly addressed since existing datasets displaying a high degree of implicit abuse are fairly biased. Following the recently-proposed strategy to solve implicit abuse by separately addressing its different subtypes, we present a new focused and less biased dataset that consists of the subtype of atomic negative sentences about identity groups. For that task, we model components that each address one facet of such implicit abuse, i.e. depiction as perpetrators, aspectual classification and non-conformist views. The approach generalizes across different identity groups and languages.

The exploding amount of user-generated content has spurred NLP research to deal with documents from various digital communication formats (tweets, chats, emails, etc.). Using these texts as language resources implies complying with legal data privacy regulations. To protect the personal data of individuals and preclude their identification, we employ pseudonymization. More precisely, we identify those text spans that carry information revealing an individual’s identity (e.g., names of persons, locations, phone numbers, or dates) and subsequently substitute them with synthetically generated surrogates. Based on CodE Alltag, a German-language email corpus, we address two tasks. The first task is to evaluate various architectures for the automatic recognition of privacy-sensitive entities in raw data. The second task examines the applicability of pseudonymized data as training data for such systems since models learned on original data cannot be published for reasons of privacy protection. As outputs of both tasks, we, first, generate a new pseudonymized version of CodE Alltag compliant with the legal requirements of the General Data Protection Regulation (GDPR). Second, we make accessible a tagger for recognizing privacy-sensitive information in German emails and similar text genres, which is trained on already pseudonymized data.

We present a new dataset comprising tweets for the novel task of detecting biographically relevant utterances. Biographically relevant utterances are all those utterances that reveal some persistent and non-trivial information about the author of a tweet, e.g. habits, (dis)likes, family status, physical appearance, employment information, health issues etc. Unlike previous research we do not restrict biographical relevance to a small fixed set of pre-defined relations. Next to classification experiments employing state-of-the-art classifiers to establish strong baselines for future work, we carry out a linguistic analysis that compares the predictiveness of various high-level features. We also show that the task is different from established tasks, such as aspectual classification or sentiment analysis.

Abusive language detection is an emerging field in natural language processing which has received a large amount of attention recently. Still the success of automatic detection is limited. Particularly, the detection of implicitly abusive language, i.e. abusive language that is not conveyed by abusive words (e.g. dumbass or scum), is not working well. In this position paper, we explain why existing datasets make learning implicit abuse difficult and what needs to be changed in the design of such datasets. Arguing for a divide-and-conquer strategy, we present a list of subtypes of implicitly abusive language and formulate research tasks and questions for future research.

We present the GermEval 2021 shared task on the identification of toxic, engaging, and fact-claiming comments. This shared task comprises three binary classification subtasks with the goal to identify: toxic comments, engaging comments, and comments that include indications of a need for fact-checking, here referred to as fact-claiming comments. Building on the two previous GermEval shared tasks on the identification of offensive language in 2018 and 2019, we extend this year’s task definition to meet the demand of moderators and community managers to also highlight comments that foster respectful communication, encourage in-depth discussions, and check facts that lines of arguments rely on. The dataset comprises 4,188 posts extracted from the Facebook page of a German political talk show of a national public television broadcaster. A theoretical framework and additional reliability tests during the data annotation process ensure particularly high data quality. The shared task had 15 participating teams submitting 31 runs for the subtask on toxic comments, 25 runs for the subtask on engaging comments, and 31 for the subtask on fact-claiming comments. The shared task website can be found at https://germeval2021toxic.github.io/SharedTask/.

We examine the task of detecting implicitly abusive comparisons (e.g. “Your hair looks like you have been electrocuted”). Implicitly abusive comparisons are abusive comparisons in which abusive words (e.g. “dumbass” or “scum”) are absent. We detail the process of creating a novel dataset for this task via crowdsourcing that includes several measures to obtain a sufficiently representative and unbiased set of comparisons. We also present classification experiments that include a range of linguistic features that help us better understand the mechanisms underlying abusive comparisons.

We propose to use abusive emojis, such as the “middle finger” or “face vomiting”, as a proxy for learning a lexicon of abusive words. Since it represents extralinguistic information, a single emoji can co-occur with different forms of explicitly abusive utterances. We show that our approach generates a lexicon that offers the same performance in cross-domain classification of abusive microposts as the most advanced lexicon induction method. Such an approach, in contrast, is dependent on manually annotated seed words and expensive lexical resources for bootstrapping (e.g. WordNet). We demonstrate that the same emojis can also be effectively used in languages other than English. Finally, we also show that emojis can be exploited for classifying mentions of ambiguous words, such as “fuck” and “bitch”, into generally abusive and just profane usages.

Entity framing is the selection of aspects of an entity to promote a particular viewpoint towards that entity. We investigate entity framing of political figures through the use of names and titles in German online discourse, enhancing current research in entity framing through titling and naming that concentrates on English only. We collect tweets that mention prominent German politicians and annotate them for stance. We find that the formality of naming in these tweets correlates positively with their stance. This confirms sociolinguistic observations that naming and titling can have a status-indicating function and suggests that this function is dominant in German tweets mentioning political figures. We also find that this status-indicating function is much weaker in tweets from users that are politically left-leaning than in tweets by right-leaning users. This is in line with observations from moral psychology that left-leaning and right-leaning users assign different importance to maintaining social hierarchies.

The sentiment polarity of an expression (whether it is perceived as positive, negative or neutral) can be influenced by a number of phenomena, foremost among them negation. Apart from closed-class negation words like “no”, “not” or “without”, negation can also be caused by so-called polarity shifters. These are content words, such as verbs, nouns or adjectives, that shift polarities in their opposite direction, e.g. “abandoned” in “abandoned hope” or “alleviate” in “alleviate pain”. Many polarity shifters can affect both positive and negative polar expressions, shifting them towards the opposing polarity. However, other shifters are restricted to a single shifting direction. “Recoup” shifts negative to positive in “recoup your losses”, but does not affect the positive polarity of “fortune” in “recoup a fortune”. Existing polarity shifter lexica only specify whether a word can, in general, cause shifting, but they do not specify when this is limited to one shifting direction. To address this issue we introduce a supervised classifier that determines the shifting direction of shifters. This classifier uses both resource-driven features, such as WordNet relations, and data-driven features like in-context polarity conflicts. Using this classifier we enhance the largest available polarity shifter lexicon.

We discuss the impact of data bias on abusive language detection. We show that classification scores on popular datasets reported in previous work are much lower under realistic settings in which this bias is reduced. Such biases are most notably observed on datasets that are created by focused sampling instead of random sampling. Datasets with a higher proportion of implicit abuse are more affected than datasets with a lower proportion.

We examine the new task of detecting derogatory compounds (e.g. “curry muncher”). Derogatory compounds are much more difficult to detect than derogatory unigrams (e.g. “idiot”) since they are more sparsely represented in lexical resources previously found effective for this task (e.g. Wiktionary). We propose an unsupervised classification approach that incorporates linguistic properties of compounds. It mostly depends on a simple distributional representation. We compare our approach against previously established methods proposed for extracting derogatory unigrams.

Naming and titling have been discussed in sociolinguistics as markers of status or solidarity. However, these functions have not been studied on a larger scale or for social media data. We collect a corpus of tweets mentioning presidents of six G20 countries by various naming forms. We show that naming variation relates to stance towards the president in a way that is suggestive of a framing effect mediated by respectfulness. This confirms sociolinguistic theory of naming and titling as markers of status.

We address the detection of abusive words. The task is to identify such words among a set of negative polar expressions. We propose novel features employing information from both corpora and lexical resources. These features are calibrated on a small manually annotated base lexicon which we use to produce a large lexicon. We show that the word-level information we learn cannot be equally derived from a large dataset of annotated microposts. We demonstrate the effectiveness of our (domain-independent) lexicon in the cross-domain detection of abusive microposts.

In this paper we use methods for creating a large lexicon of verbal polarity shifters and apply them to German. Polarity shifters are content words that can move the polarity of a phrase towards its opposite, such as the verb “abandon” in “abandon all hope”. This is similar to how negation words like “not” can influence polarity. Both shifters and negation are required for high precision sentiment analysis. Lists of negation words are available for many languages, but the only language for which a sizable lexicon of verbal polarity shifters exists is English. This lexicon was created by bootstrapping a sample of annotated verbs with a supervised classifier that uses a set of data- and resource-driven features. We reproduce and adapt this approach to create a German lexicon of verbal polarity shifters. Thereby, we confirm that the approach works for multiple languages. We further improve classification by leveraging cross-lingual information from the English shifter lexicon. Using this improved approach, we bootstrap a large number of German verbal polarity shifters, reducing the annotation effort drastically. The resulting German lexicon of verbal polarity shifters is made publicly available.

We study German affixoids, a type of morpheme in between affixes and free stems. Several properties have been associated with them – increased productivity; a bleached semantics, which is often evaluative and/or intensifying and thus of relevance to sentiment analysis; and the existence of a free morpheme counterpart – but not been validated empirically. In experiments on a new data set that we make available, we put these key assumptions from the morphological literature to the test and show that despite the fact that affixoids generate many low-frequency formations, we can classify these as affixoid or non-affixoid instances with a best F1-score of 74%.

We present a major step towards the creation of the first high-coverage lexicon of polarity shifters. In this work, we bootstrap a lexicon of verbs by exploiting various linguistic features. Polarity shifters, such as “abandon”, are similar to negations (e.g. “not”) in that they move the polarity of a phrase towards its inverse, as in “abandon all hope”. While there exist lists of negation words, creating comprehensive lists of polarity shifters is far more challenging due to their sheer number. On a sample of manually annotated verbs we examine a variety of linguistic features for this task. Then we build a supervised classifier to increase coverage. We show that this approach drastically reduces the annotation effort while ensuring a high-precision lexicon. We also show that our acquired knowledge of verbal polarity shifters improves phrase-level sentiment analysis.

Unknown words are a challenge for any NLP task, including sentiment analysis. Here, we evaluate the extent to which sentiment polarity of complex words can be predicted based on their morphological make-up. We do this on German as it has very productive processes of derivation and compounding and many German hapax words, which are likely to bear sentiment, are morphologically complex. We present results of supervised classification experiments on new datasets with morphological parses and polarity annotations.

This paper presents a survey on hate speech detection. Given the steadily growing body of social media content, the amount of online hate speech is also increasing. Due to the massive scale of the web, methods that automatically detect hate speech are required. Our survey describes key areas that have been explored to automatically recognize these types of utterances using natural language processing. We also discuss limits of those approaches.

In this paper, we describe MLSA, a publicly available multi-layered reference corpus for German-language sentiment analysis. The construction of the corpus is based on the manual annotation of 270 German-language sentences considering three different layers of granularity. The sentence-layer annotation, as the most coarse-grained annotation, focuses on aspects of objectivity, subjectivity and the overall polarity of the respective sentences. Layer 2 is concerned with polarity on the word- and phrase-level, annotating both subjective and factual language. The annotations on Layer 3 focus on the expression-level, denoting frames of private states such as objective and direct speech events. These three layers and their respective annotations are intended to be fully independent of each other. At the same time, exploring for and discovering interactions that may exist between different layers should also be possible. The reliability of the respective annotations was assessed using the average pairwise agreement and Fleiss' multi-rater measures. We believe that MLSA is a beneficial resource for sentiment analysis research, algorithms and applications that focus on the German language.

We present a gold standard for semantic relation extraction in the food domain for German. The relation types that we address are motivated by scenarios for which IT applications present a commercial potential, such as virtual customer advice in which a virtual agent assists a customer in a supermarket in finding those products that satisfy their needs best. Moreover, we focus on those relation types that can be extracted from natural language text corpora, ideally content from the internet, such as web forums, that are easy to retrieve. A typical relation type that meets these requirements are pairs of food items that are usually consumed together. Such a relation type could be used by a virtual agent to suggest additional products available in a shop that would potentially complement the items a customer has already in their shopping cart. Our gold standard comprises structural data, i.e. relation tables, which encode relation instances. These tables are vital in order to evaluate natural language processing systems that extract those relations.

In recent years, text classification in sentiment analysis has mostly focused on two types of classification, the distinction between objective and subjective text, i.e. subjectivity detection, and the distinction between positive and negative subjective text, i.e. polarity classification. So far, there has been little work examining the distinction between definite polar subjectivity and indefinite polar subjectivity. While the former are utterances which can be categorized as either positive or negative, the latter cannot be categorized as either of these two categories. This paper presents a small set of domain independent features to detect indefinite polar sentences. The features reflect the linguistic structure underlying these types of utterances. We give evidence for the effectiveness of these features by incorporating them into an unsupervised rule-based classifier for sentence-level analysis and compare its performance with supervised machine learning classifiers, i.e. Support Vector Machines (SVMs) and Nearest Neighbor Classifier (kNN). The data used for the experiments are web-reviews collected from three different domains.

One problem of data-driven answer extraction in open-domain factoid question answering is that the class distribution of labeled training data is fairly imbalanced. In an ordinary training set, there are far more incorrect answers than correct answers. The class-imbalance is, thus, inherent to the classification task. It has a deteriorating effect on the performance of classifiers trained by standard machine learning algorithms. They usually have a heavy bias towards the majority class, i.e. the class which occurs most often in the training set. In this paper, we propose a method to tackle class imbalance by applying some form of cost-sensitive learning which is preferable to sampling. We present a simple but effective way of estimating the misclassification costs on the basis of class distribution. This approach offers three benefits. Firstly, it maintains the distribution of the classes of the labeled training data. Secondly, this form of meta-learning can be applied to a wide range of common learning algorithms. Thirdly, this approach can be easily implemented with the help of state-of-the-art machine learning software.