Refine
Year of publication
Document Type
- Conference Proceeding (35)
- Part of a Book (8)
- Article (1)
Language
- English (44)
Has Fulltext
- yes (44)
Keywords
- Computerlinguistik (19)
- Natürliche Sprache (15)
- Information Extraction (11)
- Sentimentanalyse (11)
- Text Mining (11)
- Automatische Sprachanalyse (10)
- Maschinelles Lernen (10)
- Beleidigung (8)
- Deutsch (7)
- Korpus <Linguistik> (7)
Publicationstate
- Veröffentlichungsversion (44) (remove)
Reviewstate
- Peer-Review (38)
- (Verlags)-Lektorat (3)
Publisher
- Association for Computational Linguistics (13)
- The Association for Computational Linguistics (6)
- German Society for Computational Linguistics & Language Technology und Friedrich-Alexander-Universität Erlangen-Nürnberg (3)
- European Language Resources Association (2)
- European language resources association (ELRA) (2)
- Incoma Ltd. (2)
- Northern European Association for Language Technology (2)
- AAAI Press (1)
- Asian Federation of Natural Language Processing (1)
- Austrian Academy of Sciences (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 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.
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 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.
This paper presents a survey on the role of negation in sentiment analysis. Negation is a very common linguistic construction that affects polarity and, therefore, needs to be taken into consideration in sentiment analysis.
We will present various computational approaches modeling negation in sentiment analysis. We will, in particular, focus on aspects such as level of representation used for sentiment analysis, negation word detection and scope of negation. We will also discuss limits and challenges of negation modeling on that task.
Automatic Food Categorization from Large Unlabeled Corpora and Its Impact on Relation Extraction
(2014)
We present a weakly-supervised induction method to assign semantic information to food items. We consider two tasks of categorizations being food-type classification and the distinction of whether a food item is composite or not. The categorizations are induced by a graph-based algorithm applied on a large unlabeled domain-specific corpus. We show that the usage of a domain-specific corpus is vital. We do not only outperform a manually designed open-domain ontology but also prove the usefulness of these categorizations in relation extraction, outperforming state-of-the-art features that include syntactic information and Brown clustering.
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.
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.
Bootstrapping Supervised Machine-learning Polarity Classifiers with Rule-based Classification
(2010)
In this paper, we explore the effectiveness of bootstrapping supervised machine-learning polarity classifiers using the output of domain-independent rule-based classifiers. The benefit of this method is that no labeled training data are required. Still, this method allows to capture in-domain knowledge by training the supervised classifier on in-domain features, such as bag of words.
We investigate how important the quality of the rule-based classifier is and what features are useful for the supervised classifier. The former addresses the issue in how far relevant constructions for polarity classification, such as word sense disambiguation, negation modeling, or intensification, are important for this self-training approach. We not only compare how this method relates to conventional semi-supervised learning but also examine how it performs under more difficult settings in which classes are not balanced and mixed reviews are included in the dataset.
We compare several different corpus- based and lexicon-based methods for the scalar ordering of adjectives. Among them, we examine for the first time a low- resource approach based on distinctive- collexeme analysis that just requires a small predefined set of adverbial modifiers. While previous work on adjective intensity mostly assumes one single scale for all adjectives, we group adjectives into different scales which is more faithful to human perception. We also apply the methods to both polar and non-polar adjectives, showing that not all methods are equally suitable for both types of adjectives.
Opinion holder extraction is one of the important subtasks in sentiment analysis. The effective detection of an opinion holder depends on the consideration of various cues on various levels of representation, though they are hard to formulate explicitly as features. In this work, we propose to use convolution kernels for that task which identify meaningful fragments of sequences or trees by themselves. We not only investigate how different levels of information can be effectively combined in different kernels but also examine how the scope of these kernels should be chosen. In general relation extraction, the two candidate entities thought to be involved in a relation are commonly chosen to be the boundaries of sequences and trees. The definition of boundaries in opinion holder extraction, however, is less straightforward since there might be several expressions beside the candidate opinion holder to be eligible for being a boundary.
In this paper, we explore different linguistic structures encoded as convolution kernels for the detection of subjective expressions. The advantage of convolution kernels is that complex structures can be directly provided to a classifier without deriving explicit features. The feature design for the detection of subjective expressions is fairly difficult and there currently exists no commonly accepted feature set. We consider various structures, such as constituency parse structures, dependency parse structures, and predicate-argument structures. In order to generalize from lexical information, we additionally augment these structures with clustering information and the task-specific knowledge of subjective words. The convolution kernels will be compared with a standard vector kernel.
In this paper, we examine methods to automatically extract domain-specific knowledge from the food domain from unlabeled natural language text. We employ different extraction methods ranging from surface patterns to co-occurrence measures applied on different parts of a document. We show that the effectiveness of a particular method depends very much on the relation type considered and that there is no single method that works equally well for every relation type. We also examine a combination of extraction methods and also consider relationships between different relation types. The extraction methods are applied both on a domain-specific corpus and the domain-independent factual knowledge base Wikipedia. Moreover, we examine an open-domain lexical ontology for suitability.
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.
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.
Negation is an important contextual phenomenon that needs to be addressed in sentiment analysis. Next to common negation function words, such as not or none, there is also a considerably large class of negation content words, also referred to as shifters, such as the verbs diminish, reduce or reverse. However, many of these shifters are ambiguous. For instance, spoil as in spoil your chance reverses the polarity of the positive polar expression chance while in spoil your loved ones, no negation takes place. We present a supervised learning approach to disambiguating verbal shifters. Our approach takes into consideration various features, particularly generalization features.
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%.
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.
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.
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.
The sentiment polarity of a phrase does not only depend on the polarities of its words, but also on how these are affected by their context. Negation words (e.g. not, no, never) can change the polarity of a phrase. Similarly, verbs and other content words can also act as polarity shifters (e.g. fail, deny, alleviate). While individually more sparse, they are far more numerous. Among verbs alone, there are more than 1200 shifters. However, sentiment analysis systems barely consider polarity shifters other than negation words. A major reason for this is the scarcity of lexicons and corpora that provide information on them. We introduce a lexicon of verbal polarity shifters that covers the entirety of verbs found in WordNet. We provide a fine-grained annotation of individual word senses, as well as information for each verbal shifter on the syntactic scopes that it can affect.
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.
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 for modeling German negation in open-domain fine grained sentiment analysis. Unlike most previous work in sentiment analysis, we assume that negation can be conveyed by many lexical units (and not only common negation words) and that different negation words have different scopes. Our approach is examined on a new dataset comprising sentences with mentions of polar expressions and various negation words. We identify different types of negation words that have the same scopes. We show that already negation modeling based on these types largely outperforms traditional negation models which assume the same scope for all negation words and which employ a window-based scope detection rather than a scope detection based on syntactic information.
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 present an approach for opinion role induction for verbal predicates. Our model rests on the assumption that opinion verbs can be divided into three different types where each type is associated with a characteristic mapping between semantic roles and opinion holders and targets. In several experiments, we demonstrate the relevance of those three categories for the task. We show that verbs can easily be categorized with semi-supervised graphbased clustering and some appropriate similarity metric. The seeds are obtained through linguistic diagnostics. We evaluate our approach against a new manually-compiled opinion role lexicon and perform in-context classification.
Opinion Holder and Target Extraction for Verb-based Opinion Predicates – The Problem is Not Solved
(2015)
We offer a critical review of the current state of opinion role extraction involving opinion verbs. We argue that neither the currently available lexical resources nor the manually annotated text corpora are sufficient to appropriately study this task. We introduce a new corpus focusing on opinion roles of opinion verbs from the Subjectivity Lexicon and show potential benefits of this corpus. We also demonstrate that state-of-the-art classifiers perform rather poorly on this new dataset compared to the standard dataset for the task showing that there still remains significant research to be done.
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.
In recent years, theoretical and computational linguistics has paid much attention to linguistic items that form scales. In NLP, much research has focused on ordering adjectives by intensity (tiny < small). Here, we address the task of automatically ordering English adverbs by their intensifying or diminishing effect on adjectives (e.g. extremely small < very small). We experiment with 4 different methods: 1) using the association strength between adverbs and adjectives; 2) exploiting scalar patterns (such as not only X but Y); 3) using the metadata of product reviews; 4) clustering. The method that performs best is based on the use of metadata and ranks adverbs by their scaling factor relative to unmodified adjectives.
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 present the pilot edition of the GermEval Shared Task on the Identification of Offensive Language. This shared task deals with the classification of German tweets from Twitter. It comprises two tasks, a coarse-grained binary classification task and a fine-grained multi-class classification task. The shared task had 20 participants submitting 51 runs for the coarse-grained task and 25 runs for the fine-grained task. Since this is a pilot task, we describe the process of extracting the raw-data for the data collection and the annotation schema. We evaluate the results of the systems submitted to the shared task. The shared task homepage can be found at https://projects.cai. fbi.h-da.de/iggsa/
Overview of the IGGSA 2016 Shared Task on Source and Target Extraction from Political Speeches
(2016)
We present the second iteration of IGGSA’s Shared Task on Sentiment Analysis for German. It resumes the STEPS task of IGGSA’s 2014 evaluation campaign: Source, Subjective Expression and Target Extraction from Political Speeches. As before, the task is focused on fine-grained sentiment analysis, extracting sources and targets with their associated subjective expressions from a corpus of speeches given in the Swiss parliament. The second iteration exhibits some differences, however; mainly the use of an adjudicated gold standard and the availability of training data. The shared task had 2 participants submitting 7 runs for the full task and 3 runs for each of the subtasks. We evaluate the results and compare them to the baselines provided by the previous iteration. The shared task homepage can be found at http://iggsasharedtask2016.github.io/.
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).
In opinion mining, there has been only very little work investigating semi-supervised machine learning on document-level polarity classification. We show that semi-supervised learning performs significantly better than supervised learning when only few labelled data are available. Semi-supervised polarity classifiers rely on a predictive feature set. (Semi-)Manually built polarity lexicons are one option but they are expensive to obtain and do not necessarily work in an unknown domain. We show that extracting frequently occurring adjectives & adverbs of an unlabeled set of in-domain documents is an inexpensive alternative which works equally well throughout different domains.
In order to automatically extract opinion holders, we propose to harness the contexts of prototypical opinion holders, i.e. common nouns, such as experts or analysts, that describe particular groups of people whose profession or occupation is to form and express opinions towards specific items. We assess their effectiveness in supervised learning where these contexts are regarded as labelled training data and in rule-based classification which uses predicates that frequently co-occur with mentions of the prototypical opinion holders. Finally, we also examine in how far knowledge gained from these contexts can compensate the lack of large amounts of labeled training data in supervised learning by considering various amounts of actually labeled training sets.
We report on the two systems we built for Task 1 of the German Sentiment Analysis Shared Task, the task on Source, Subjective Expression and Target Extraction from Political Speeches (STEPS). The first system is a rule-based system relying on a predicate lexicon specifying extraction rules for verbs, nouns and adjectives, while the second is a translation-based system that has been obtained with the help of the (English) MPQA corpus.
We examine different features and classifiers for the categorization of opinion words into actor and speaker view. To our knowledge, this is the first comprehensive work to address sentiment views on the word level taking into consideration opinion verbs, nouns and adjectives. We consider many high-level features requiring only few labeled training data. A detailed feature analysis produces linguistic insights into the nature of sentiment views. We also examine how far global constraints between different opinion words help to increase classification performance. Finally, we show that our (prior) word-level annotation correlates with contextual sentiment views.
We examine the task of separating types from brands in the food domain. Framing the problem as a ranking task, we convert simple textual features extracted from a domain-specific corpus into a ranker without the need of labeled training data. Such method should rank brands (e.g. sprite) higher than types (e.g. lemonade). Apart from that, we also exploit knowledge induced by semi-supervised graph-based clustering for two different purposes. On the one hand, we produce an auxiliary categorization of food items according to the Food Guide Pyramid, and assume that a food item is a type when it belongs to a category unlikely to contain brands. On the other hand, we directly model the task of brand detection using seeds provided by the output of the textual ranking features. We also harness Wikipedia articles as an additional knowledge source.
Though polarity classification has been extensively explored at document level, there has been little work investigating feature design at sentence level. Due to the small number of words within a sentence, polarity classification at sentence level differs substantially from document-level classification in that resulting bag-of-words feature vectors tend to be very sparse resulting in a lower classification accuracy.
In this paper, we show that performance can be improved by adding features specifically designed for sentence-level polarity classification. We consider both explicit polarity information and various linguistic features. A great proportion of the improvement that can be obtained by using polarity information can also be achieved by using a set of simple domain-independent linguistic features.
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.
We investigate the task of detecting reliable statements about food-health relationships from natural language texts. For that purpose, we created a specially annotated web corpus from forum entries discussing the healthiness of certain food items. We examine a set of task-specific features (mostly) based on linguistic insights that are instrumental in finding utterances that are commonly perceived as reliable. These features are incorporated in a supervised classifier and compared against standard features that are widely used for various tasks in natural language processing, such as bag of words, part-of speech and syntactic parse information.