Refine
Year of publication
Document Type
- Conference Proceeding (38) (remove)
Language
- English (38)
Has Fulltext
- yes (38)
Keywords
- Automatische Sprachanalyse (13)
- Deutsch (9)
- Beleidigung (7)
- Computerlinguistik (6)
- Frame-Semantik (6)
- Korpus <Linguistik> (6)
- Semantische Analyse (6)
- Annotation (4)
- Beschimpfung (4)
- Propositionale Einstellung (4)
Publicationstate
- Veröffentlichungsversion (38) (remove)
Reviewstate
- Peer-Review (38) (remove)
Publisher
- Association for Computational Linguistics (9)
- German Society for Computational Linguistics & Language Technology und Friedrich-Alexander-Universität Erlangen-Nürnberg (4)
- European Language Resources Association (3)
- European Language Resources Association (ELRA) (2)
- The Association for Computational Linguistics (2)
- Asian Federation of Natural Language Processing (1)
- Austrian academy of sciences (1)
- Berkeley Linguistics Society (1)
- Berkeley Linguistics Society, Inc. (1)
- EDUCatt (1)
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 a testsuite for POS tagging German web data. Our testsuite provides the original raw text as well as the gold tokenisations and is annotated for parts-of-speech. The testsuite includes a new dataset for German tweets, with a current size of 3,940 tokens. To increase the size of the data, we harmonised the annotations in already existing web corpora, based on the Stuttgart-Tübingen Tag Set. The current version of the corpus has an overall size of 48,344 tokens of web data, around half of it from Twitter. We also present experiments, showing how different experimental setups (training set size, additional out-of-domain training data, self-training) influence the accuracy of the taggers. All resources and models will be made publicly available to the research community.
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.
This paper presents Release 2.0 of the SALSA corpus, a German resource for lexical semantics. The new corpus release provides new annotations for German nouns, complementing the existing annotations of German verbs in Release 1.0. The corpus now includes around 24,000 sentences with more than 36,000 annotated instances. It was designed with an eye towards NLP applications such as semantic role labeling but will also be a useful resource for linguistic studies in lexical semantics.
Active learning has been applied to different NLP tasks, with the aim of limiting the amount of time and cost for human annotation. Most studies on active learning have only simulated the annotation scenario, using prelabelled gold standard data. We present the first active learning experiment for Word Sense Disambiguation with human annotators in a realistic environment, using fine-grained sense distinctions, and investigate whether AL can reduce annotation cost and boost classifier performance when applied to a real-world task.
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.
Automatic division of spoken language transcripts into sentence-like units is a challenging problem, caused by disfluencies, ungrammatical structures and the lack of punctuation. We present experiments on dividing up German spoken dialogues where we investigate the impact of task setup and data representation, encoding of context information as well as different model architectures for this task.
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.
Sentiment analysis has so far focused on the detection of explicit opinions. However, of late implicit opinions have received broader attention, the key idea being that the evaluation of an event type by a speaker depends on how the participants in the event are valued and how the event itself affects the participants. We present an annotation scheme for adding relevant information, couched in terms of so-called effect functors, to German lexical items. Our scheme synthesizes and extends previous proposals. We report on an inter-annotator agreement study. We also present results of a crowdsourcing experiment to test the utility of some known and some new functors for opinion inference where, unlike in previous work, subjects are asked to reason from event evaluation to participant evaluation.
Active Learning (AL) has been proposed as a technique to reduce the amount of annotated data needed in the context of supervised classification. While various simulation studies for a number of NLP tasks have shown that AL works well on goldstandard data, there is some doubt whether the approach can be successful when applied to noisy, real-world data sets. This paper presents a thorough evaluation of the impact of annotation noise on AL and shows that systematic noise resulting from biased coder decisions can seriously harm the AL process. We present a method to filter out inconsistent annotations during AL and show that this makes AL far more robust when applied to noisy data.
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.
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.
The classification of verbs in Levin's (1993) English Verb Classes and Alternations: A preliminary Investigation, on the basis of both intuitive semantic grouping and their participation in valence alternations, is often used by the NLP community as evidence of the semantic similarity of verbs (Jing & McKeown 1998; Lapata & Brew 1999; Kohl et al. 1998). In this paper, we compare the Levin classification with the work of the FrameNet project (Fillmore & Baker 2001), where words (not just verbs) are grouped according to the conceptual structures (frames) that underlie them and their combinatorial patterns are inductively derived from corpus evidence. This means that verbs grouped together in FrameNet (FN) might be semantically similar but have different (or no) alternations, and that verbs which share the same alternation might be represented in two different semantic frames.
We present a method and a software tool, the FrameNet Transformer, for deriving customized versions of the FrameNet database based on frame and frame element relations. The FrameNet Transformer allows users to iteratively coarsen the FrameNet sense inventory in two ways. First, the tool can merge entire frames that are related by user-specified relations. Second, it can merge word senses that belong to frames related by specified relations. Both methods can be interleaved. The Transformer automatically outputs format-compliant FrameNet versions, including modified corpus annotation files that can be used for automatic processing. The customized FrameNet versions can be used to determine which granularity is suitable for particular applications. In our evaluation of the tool, we show that our method increases accuracy of statistical semantic parsers by reducing the number of word-senses (frames) per lemma, and increasing the number of annotated sentences per lexical unit and frame. We further show in an experiment on the FATE corpus that by coarsening FrameNet we do not incur a significant loss of information that is relevant to the Recognizing Textual Entailment task.
We present a quantitative approach to disambiguating flat morphological analyses and producing more deeply structured analyses. Based on existing morphological segmentations, possible combinations of resulting word trees for the next level are filtered first by criteria of linguistic plausibility and then by weighting procedures based on the geometric mean. The frequencies for weighting are derived from three different sources (counts of morphs in a lexicon, counts of largest constituents in a lexicon, counts of token frequencies in a corpus) and can be used either to find the best analysis on the level of morphs or on the next higher constituent level. The evaluation shows that for this task corpus-based frequency counts are slightly superior to counts of lexical data.
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.
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.
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.
Semantic argument structures are often incomplete in that core arguments are not locally instantiated. However, many of these implicit arguments can be linked to referents in the wider context. In this paper we explore a number of linguistically motivated strategies for identifying and resolving such null instantiations (NIs). We show that a more sophisticated model for identifying definite NIs can lead to noticeable performance gains over the state-of-the- art for NI resolution.
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 crossdomain detection of abusive microposts.
We introduce a system that learns the participants of arbitrary given scripts. This system processes data from web experiments, in which each participant can be realized with different expressions. It computes participants by encoding semantic similarity and global structural information into an Integer Linear Program. An evaluation against a gold standard shows that we significantly outperform two informed baselines.
We present MaJo, a toolkit for supervised Word Sense Disambiguation (WSD), with an interface for Active Learning. Our toolkit combines a flexible plugin architecture which can easily be extended, with a graphical user interface which guides the user through the learning process. MaJo integrates off-the-shelf NLP tools like POS taggers, treebank-trained statistical parsers, as well as linguistic resources like WordNet and GermaNet. It enables the user to systematically explore the benefit gained from different feature types for WSD. In addition, MaJo provides an Active Learning environment, where the
system presents carefully selected instances to a human oracle. The toolkit supports manual annotation of the selected instances and re-trains the system on the extended data set. MaJo also provides the means to evaluate the performance of the system against a gold standard. We illustrate the usefulness of our system by learning the frames (word senses) for three verbs from the SALSA corpus, a version of the TiGer treebank with an additional layer of frame-semantic annotation. We show how MaJo can be used to tune the feature set for specific target words and so improve performance for these targets. We also show that syntactic features, when carefully tuned to the target word, can lead to a substantial increase in performance.
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 an approach to the new task of opinion holder and target extraction on opinion compounds. Opinion compounds (e.g. user rating or victim support) are noun compounds whose head is an opinion noun. We do not only examine features known to be effective for noun compound analysis, such as paraphrases and semantic classes of heads and modifiers, but also propose novel features tailored to this new task. Among them, we examine paraphrases that jointly consider holders and targets, a verb detour in which noun heads are replaced by related verbs, a global head constraint allowing inferencing between different compounds, and the categorization of the sentiment view that the head conveys.
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/
We examine predicative adjectives as an unsupervised criterion to extract subjective adjectives. We do not only compare this criterion with a weakly supervised extraction method but also with gradable adjectives, i.e. another highly subjective subset of adjectives that can be extracted in an unsupervised fashion. In order to prove the robustness of this extraction method, we will evaluate the extraction with the help of two different state-of-the-art sentiment lexicons (as a gold standard).
Reframing FrameNet Data
(2004)
The Berkeley FrameNet Project (http://www.icsi.berkeley.edu/~framenet) is building an on-line lexical resource for contemporary English. The database provides information about the semantic and syntactic combinatorial possibilities (valences) of each item analyzed. This paper describes the conceptual basis for what has been called reframing of data in the FrameNet database and exemplifies two new frame-to-frame relations, Causative_of and Inchoative_of, the implementation of which came about as a result of reanalysis of certain frames and lexical units. The new relations are characterized with respect to a triple of frames involving the notion of attaching, and entering them into the database is demonstrated using the Frame Relations Editor. The two relations allow FrameNet to make frame-wise distinctions that capture fairly systematic semantic relationships across sets of lexical units. While the Inheritance and Subframe relations are of particular interest to the NLP research community, Causative_of and Inchoative_of may be more relevant to lexicography.
Current work on sentiment analysis is characterized by approaches with a pragmatic focus, which use shallow techniques in the interest of robustness but often rely on ad-hoc creation of data sets and methods. We argue that progress towards deep analysis depends on a) enriching shallow representations with linguistically motivated, rich information, and b) focussing different branches of research and combining ressources to create synergies with related work in NLP. In the paper, we propose SentiFrameNet, an extension to FrameNet, as a novel representation for sentiment analysis that is tailored to these aims.
We describe the SemEval-2010 shared task on “Linking Events and Their Participants in Discourse”. This task is an extension to the classical semantic role labeling task. While semantic role labeling is traditionally viewed as a sentence-internal task, local semantic argument structures clearly interact with each other in a larger context, e.g., by sharing references to specific discourse entities or events. In the shared task we looked at one particular aspect of cross-sentence links between argument structures, namely linking locally uninstantiated roles to their co-referents in the wider discourse context (if such co-referents exist). This task is potentially beneficial for a number of NLP applications, such as information extraction, question answering or text summarization.
Both for psychology and linguistics, emotion concepts are a continuing challenge for analysis in several respects. In this contribution, we take up the language of emotion as an object of study from several angles. First, we consider how frame semantic analyses of this domain by the FrameNet project have been developing over time, due to theory-internal as well as application-oriented goals, towards ever more fine-grained distinctions and greater within-frame consistency. Second, we compare how FrameNet’s linguistically oriented analysis of lexical items in the emotion domain compares to the analysis by domain experts of the experiences that give rise (directly or indirectly) to the lexical items. And finally, we consider to what extent frame semantic analysis can capture phenomena such as connotation and inference about attitudes, which are important in the field of sentiment analysis and opinion mining, even if they do not involve the direct evocation of emotion.
In the paper we investigate the impact of data size on a Word Sense Disambiguation task (WSD). We question the assumption that the knowledge acquisition bottleneck, which is known as one of the major challenges for WSD, can be solved by simply obtaining more and more training data. Our case study on 1,000 manually annotated instances of the German verb drohen (threaten) shows that the best performance is not obtained when training on the full data set, but by carefully selecting new training instances with regard to their informativeness for the learning process (Active Learning). We present a thorough evaluation of the impact of different sampling methods on the data sets and propose an improved method for uncertainty sampling which dynamically adapts the selection of new instances to the learning progress of the classifier, resulting in more robust results during the initial stages of learning. A qualitative error analysis identifies problems for automatic WSD and discusses the reasons for the great gap in performance between human annotators and our automatic WSD system.
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.
Who is we? Disambiguating the referents of first person plural pronouns in parliamentary debates
(2021)
This paper investigates the use of first person plural pronouns as a rhetorical device in political speeches. We present an annotation schema for disambiguating pronoun references and use our schema to create an annotated corpus of debates from the German Bundestag. We then use our corpus to learn to automatically resolve pronoun referents in parliamentary debates. We explore the use of data augmentation with weak supervision to further expand our corpus and report preliminary results.
This paper presents an annotation scheme for English modal verbs together with sense-annotated data from the news domain. We describe our annotation scheme and discuss problematic cases for modality annotation based on the inter-annotator agreement during the annotation. Furthermore, we present experiments on automatic sense tagging, showing that our annotations do provide a valuable training resource for NLP systems.