Refine
Year of publication
Document Type
- Conference Proceeding (47)
- Part of a Book (9)
- Article (3)
- Doctoral Thesis (1)
- Master's Thesis (1)
Language
- English (61)
Has Fulltext
- yes (61)
Keywords
- Computerlinguistik (34)
- Natürliche Sprache (29)
- Information Extraction (24)
- Maschinelles Lernen (21)
- Text Mining (16)
- Sentimentanalyse (15)
- Lebensmittel (11)
- Polarität (11)
- Automatische Sprachanalyse (10)
- Beleidigung (8)
Publicationstate
- Veröffentlichungsversion (44)
- Zweitveröffentlichung (12)
- Postprint (2)
Reviewstate
Publisher
- Association for Computational Linguistics (14)
- European Language Resources Association (6)
- Springer (6)
- The Association for Computational Linguistics (6)
- German Society for Computational Linguistics & Language Technology und Friedrich-Alexander-Universität Erlangen-Nürnberg (3)
- Incoma Ltd. (3)
- Asian Federation of Natural Language Processing (2)
- European language resources association (ELRA) (2)
- Northern European Association for Language Technology (2)
- Universitätsverlag Hildesheim (2)
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.
Implicitly abusive language – What does it actually look like and why are we not getting there?
(2021)
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 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.
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.
Alleviating pain is good and abandoning hope is bad. We instinctively understand how words like alleviate and abandon affect the polarity of a phrase, inverting or weakening it. When these words are content words, such as verbs, nouns, and adjectives, we refer to them as polarity shifters. Shifters are a frequent occurrence in human language and an important part of successfully modeling negation in sentiment analysis; yet research on negation modeling has focused almost exclusively on a small handful of closed-class negation words, such as not, no, and without. A major reason for this is that shifters are far more lexically diverse than negation words, but no resources exist to help identify them. We seek to remedy this lack of shifter resources by introducing a large lexicon of polarity shifters that covers English verbs, nouns, and adjectives. Creating the lexicon entirely by hand would be prohibitively expensive. Instead, we develop a bootstrapping approach that combines automatic classification with human verification to ensure the high quality of our lexicon while reducing annotation costs by over 70%. Our approach leverages a number of linguistic insights; while some features are based on textual patterns, others use semantic resources or syntactic relatedness. The created lexicon is evaluated both on a polarity shifter gold standard and on a polarity classification task.
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.
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.
A Supervised learning approach for the extraction of opinion sources and targets from German text
(2019)
We present the first systematic supervised learning approach for the extraction of opinion sources and targets on German language data. A wide choice of different features is presented, particularly syntactic features and generalization features. We point out specific differences between opinion sources and targets. Moreover, we explain why implicit sources can be extracted even with fairly generic features. In order to ensure comparability our classifier is trained and tested on the dataset of the STEPS shared task.
We present a descriptive analysis on the two datasets from the shared task on Source, Subjective Expression and Target Extraction from Political Speeches (STEPS), the only existing German dataset for opinion role extraction of its size. Our analysis discusses the individual properties of the three components, subjective expressions, sources and targets and their relations towards each other. Our observations should help practitioners and researchers when building a system to extract opinion roles from German data.
We present the second edition of the GermEval Shared Task on the Identification of Offensive Language. This shared task deals with the classification of German tweets from Twitter. Two subtasks were continued from the first edition, namely a coarse-grained binary classification task and a fine-grained multi-class classification task. As a novel subtask, we introduce the classification of offensive tweets as explicit or implicit.
The shared task had 13 participating groups submitting 28 runs for the coarse-grained
task, another 28 runs for the fine-grained task, and 17 runs for the implicit-explicit
task.
We evaluate the results of the systems submitted to the shared task. The shared task homepage can be found at https://projects.fzai.h-da.de/iggsa/
Knowledge Acquisition with Natural Language Processing in the Food Domain: Potential and Challenges
(2012)
In this paper, we present an outlook on the effectiveness of natural language processing (NLP) in extracting knowledge for the food domain. We identify potential scenarios that we think are particularly suitable for NLP techniques. As a source for extracting knowledge we will highlight the benefits of textual content from social media. Typical methods that we think would be suitable will be discussed. We will also address potential problems and limits that the application of NLP methods may yield.
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.
In this article, we examine the effectiveness of bootstrapping supervised machine-learning polarity classifiers with the help of a domain-independent rule-based classifier that relies on a lexical resource, i.e., a polarity lexicon and a set of linguistic rules. The benefit of this method is that though no labeled training data are required, it allows a classifier to capture in-domain knowledge by training a supervised classifier with in-domain features, such as bag of words, on instances labeled by a rule-based classifier. Thus, this approach can be considered as a simple and effective method for domain adaptation. Among the list of components of this approach, we investigate how important the quality of the rule-based classifier is and what features are useful for the supervised classifier. In particular, the former addresses the issue in how far linguistic modeling is relevant for this task. We not only examine how this method performs under more difficult settings in which classes are not balanced and mixed reviews are included in the data set but also compare how this linguistically-driven method relates to state-of-the-art statistical domain adaptation.
In this article, we explore the feasibility of extracting suitable and unsuitable food items for particular health conditions from natural language text. We refer to this task as conditional healthiness classification. For that purpose, we annotate a corpus extracted from forum entries of a food-related website. We identify different relation types that hold between food items and health conditions going beyond a binary distinction of suitability and unsuitability and devise various supervised classifiers using different types of features. We examine the impact of different task-specific resources, such as a healthiness lexicon that lists the healthiness status of a food item and a sentiment lexicon. Moreover, we also consider task-specific linguistic features that disambiguate a context in which mentions of a food item and a health condition co-occur and compare them with standard features using bag of words, part-of-speech information and syntactic parses. We also investigate in how far individual food items and health conditions correlate with specific relation types and try to harness this information for classification.
Sentiment Analysis is the task of extracting and classifying opinionated content in natural language texts. Common subtasks are the distinction between opinionated and factual texts, the classification of polarity in opinionated texts, and the extraction of the participating entities of an opinion(-event), i.e. the source from which an opinion emanates and the target towards which it is directed. With the emerging Web 2.0 which describes the shift towards a highly user-interactive communication medium, the amount of subjective content on the World Wide Web is steadily increasing. Thus, there is a growing need for automatically processing this type of content which is provided by sentiment analysis. Both natural language processing, which is the task of providing computational methods for the analysis and representation of natural language, and machine learning, which is the task of building task-specific classification models on the basis of empirical data, may be instrumental in mastering the challenges of the automatic sentiment analysis of written text. Many problems in sentiment analysis have been proposed to be solved with machine learning methods exclusively using a fairly low-level feature design, such as bag of words, containing little linguistic information. In this thesis, we examine the effectiveness of linguistic features in various subtasks of sentiment analysis. Thus, we heavily draw from the insights gained by natural language processing. The application of linguistic features can be applied on various classification methods, be it in rule-based classification, where the linguistic features are directly encoded as a classifier, in supervised machine learning, where these features complement basic low-level features, or in bootstrapping methods, where these features form a rule-based classifier generating a labeled training set from which a supervised classifier can be trained. In this thesis, we will in particular focus on scenarios where the combination of linguistic features and machine learning methods is effective. We will look at common text classification tasks, both coarse-grained and fine-grained, and extraction tasks.
While good results have been achieved for named entity recognition (NER) in supervised settings, it remains a problem that for low resource languages and less studied domains little or no labelled data is available. As NER is a crucial preprocessing step for many natural language processing tasks, finding a way to overcome this deficit in data remains of great interest. We propose a distant supervision approach to NER that is both language and domain independent where we automatically generate labelled training data using gazetteers that we previously extracted from Wikipedia. We test our approach on English, German and Estonian data sets and contribute further by introducing several successful methods to reduce the noise in the generated training data. The tested models beat baseline systems and our results show that distant supervision can be a promising approach for NER when no labelled data is available. For the English model we also show that the distant supervision model is better at generalizing within the same domain of news texts by comparing it against a supervised model on a different test set.
Opinion holder extraction is one of the most important tasks in sentiment analysis. We will briefly outline the importance of predicates for this task and categorize them according to part of speech and according to which semantic role they select for the opinion holder. For many languages there do not exist semantic resources from which such predicates can be easily extracted. Therefore, we present alternative corpus-based methods to gain such predicates automatically, including the usage of prototypical opinion holders, i.e. common nouns, denoting for example experts or analysts, which describe particular groups of people whose profession or occupation is to form and express opinions towards specific items.
In der natürlichen Sprachverarbeitung haben Frage-Antwort-Systeme in der letzten Dekade stark an Bedeutung gewonnen. Vor allem durch robuste Werkzeuge wie statistische Syntax-Parser und Eigennamenerkenner ist es möglich geworden, linguistisch strukturierte Informationen aus unannotierten Textkorpora zu gewinnen. Zusätzlich werden durch die Text REtrieval Conference (TREC) jährlich Maßstäbe für allgemeine domänen-unabhängige Frage-Antwort-Szenarien definiert. In der Regel funktionieren Frage-Antwort-Systeme nur gut, wenn sie robuste Verfahren für die unterschiedlichen Fragetypen, die in einer Fragemenge vorkommen, implementieren. Ein charakteristischer Fragetyp sind die sogenannten Ereignisfragen. Obwohl Ereignisse schon seit Mitte des vorigen Jahrhunderts in der theoretischen Linguistik, vor allem in der Satzsemantik, Gegenstand intensive Forschung sind, so blieben sie bislang im Bezug auf Frage-Antwort-Systeme weitgehend unerforscht. Deshalb widmet sich diese Diplomarbeit diesem Problem. Ziel dieser Arbeit ist zum Einen eine Charakterisierung von Ereignisstruktur in Frage-Antwort Systemen, die unter Berücksichtigung der theoretischen Linguistik sowie einer Analyse der TREC 2005 Fragemenge entstehen soll. Zum Anderen soll ein Ereignis-basiertes Antwort-Extraktionsverfahren entworfen und implementiert werden, das sich auf den Ergebnissen dieser Analyse stützt. Informationen von diversen linguistischen Ebenen sollen daten-getrieben in einem uniformen Modell integriert werden. Spezielle linguistische Ressourcen, wie z.B. WordNet und Subkategorisierungslexika werden dabei eine zentrale Rolle einnehmen. Ferner soll eine Ereignisstruktur vorgestellt werden, die das Abpassen von Ereignissen unabhängig davon, ob sie von Vollverben oder Nominalisierungen evoziert werden, erlaubt. Mit der Implementierung eines Ereignis-basierten Antwort-Extraktionsmoduls soll letztendlich auch die Frage beantwortet werden, ob eine explizite Ereignismodellierung die Performanz eines Frage-Antwort-Systems verbessern kann.