Idioms present many challenges to semantic annotation in a lexicalized framework, which leads to them being underrepresented or inadequately annotated in sembanks. In this work, we address this problem with respect to verbal idioms in the Parallel Meaning Bank (PMB), specifically in its German part, where only some idiomatic expressions have been annotated correctly. We first select candidate idiomatic expressions, then determine their idiomaticity status and whether they are decomposable or not, and then we annotate their semantics using WordNet senses and VerbNet semantic roles. Overall, inter-annotator agreement is very encouraging. A difficulty, however, is to choose the correct word sense. This is not surprising, given that English synsets are many and there is often no unique mapping from German idioms and words to them. Besides this, there are many subtle differences and interesting challenging cases. We discuss some of them in this paper.

Pre-trained Language Models (PLMs) are known to contain various kinds of knowledge.One method to infer relational knowledge is through the use of cloze-style prompts, where a model is tasked to predict missing subjects orobjects. Typically, designing these prompts is a tedious task because small differences in syntax or semantics can have a substantial impact on knowledge retrieval performance. Simultaneously, evaluating the impact of either prompt syntax or information is challenging due to their interdependence. We designed CONPARE-LAMA – a dedicated probe, consisting of 34 million distinct prompts that facilitate comparison across minimal paraphrases. These paraphrases follow a unified meta-template enabling the controlled variation of syntax and semantics across arbitrary relations.CONPARE-LAMA enables insights into the independent impact of either syntactical form or semantic information of paraphrases on the knowledge retrieval performance of PLMs. Extensive knowledge retrieval experiments using our probe reveal that prompts following clausal syntax have several desirable properties in comparison to appositive syntax: i) they are more useful when querying PLMs with a combination of supplementary information, ii) knowledge is more consistently recalled across different combinations of supplementary information, and iii) they decrease response uncertainty when retrieving known facts. In addition, range information can boost knowledge retrieval performance more than domain information, even though domain information is more reliably helpful across syntactic forms.

We introduce SPUD (Semantically Perturbed Universal Dependencies), a framework for creating nonce treebanks for the multilingual Universal Dependencies (UD) corpora. SPUD data satisfies syntactic argument structure, provides syntactic annotations, and ensures grammaticality via language-specific rules. We create nonce data in Arabic, English, French, German, and Russian, and demonstrate two use cases of SPUD treebanks. First, we investigate the effect of nonce data on word co-occurrence statistics, as measured by perplexity scores of autoregressive (ALM) and masked language models (MLM). We find that ALM scores are significantly more affected by nonce data than MLM scores. Second, we show how nonce data affects the performance of syntactic dependency probes. We replicate the findings of Müller-Eberstein et al. (2022) on nonce test data and show that the performance declines on both MLMs and ALMs wrt. original test data. However, a majority of the performance is kept, suggesting that the probe indeed learns syntax independently from semantics.

This paper addresses the problem of word sense induction (WSI) via clustering of word embeddings. It starts from the hypothesis that contextualized word representations obtained from pre-trained language models (LMs), while being a valuable source for WSI, encode more information than what is necessary for the identification of word senses and some of this information affect the performance negatively in unsupervised settings. We investigate whether using contextualized representations that are invariant to these ‘nuisance features’ can increase WSI performance. For this purpose, we propose an adaptation of the adversarial training framework proposed by Jaiswal et al. (2020) to erase specific information from the representations of LMs, thereby creating feature-invariant representations. We experiment with erasing (i) morphological and (ii) syntactic features. The results of subsequent clustering for WSI show that these features indeed act like noise: Using feature-invariant representations, compared to using the original representations, increases clustering-based WSI performance. Furthermore, we provide an in-depth analysis of how the information about the syntactic and morphological features of words relate to and affect WSI performance.

Text simplification is an intralingual translation task in which documents, or sentences of a complex source text are simplified for a target audience. The success of automatic text simplification systems is highly dependent on the quality of parallel data used for training and evaluation. To advance sentence simplification and document simplification in German, this paper presents DEplain, a new dataset of parallel, professionally written and manually aligned simplifications in plain German “plain DE” or in German: “Einfache Sprache”. DEplain consists of a news-domain (approx. 500 document pairs, approx. 13k sentence pairs) and a web-domain corpus (approx. 150 aligned documents, approx. 2k aligned sentence pairs). In addition, we are building a web harvester and experimenting with automatic alignment methods to facilitate the integration of non-aligned and to be-published parallel documents. Using this approach, we are dynamically increasing the web-domain corpus, so it is currently extended to approx. 750 document pairs and approx. 3.5k aligned sentence pairs. We show that using DEplain to train a transformer-based seq2seq text simplification model can achieve promising results. We make available the corpus, the adapted alignment methods for German, the web harvester and the trained models here: https://github.com/rstodden/DEPlain.

Identifying semantic argument types in predication contexts is not a straightforward task for several reasons, such as inherent polysemy, coercion, and copredication phenomena. In this paper, we train monolingual and multilingual classifiers with a zero-shot cross-lingual approach to identify semantic argument types in predications using pre-trained language models as feature extractors. We train classifiers for different semantic argument types and for both verbal and adjectival predications. Furthermore, we propose a method to detect copredication using these classifiers through identifying the argument semantic type targeted in different predications over the same noun in a sentence. We evaluate the performance of the method on copredication test data with Food•Event nouns for 5 languages.

This paper introduces an approach which operationalizes the role of discourse connectives for detecting argument stance. Specifically, the study investigates the utility of masked language model probabilities of discourse connectives inserted between a claim and a premise that supports or attacks it. The research focuses on a range of connectives known to signal support or attack, such as because, but, so, or although. By employing a LightGBM classifier, the study reveals promising results in stance detection in English discourse. While the proposed system does not aim to outperform state-of-the-art architectures, the classification accuracy is surprisingly high, highlighting the potential of these features to enhance argument mining tasks, including stance detection.

This paper addresses the task of semantic frame induction based on pre-trained language models (LMs). The current state of the art is to directly use contextualized embeddings from models such as BERT and to cluster them in a two step clustering process (first lemma-internal, then over all verb tokens in the data set). We propose not to use the LM’s embeddings as such but rather to refine them via some transformer-based denoising autoencoder. The resulting embeddings allow to obtain competitive results while clustering them in a single pass. This shows clearly that the autoendocer allows to already concentrate on the information that is relevant for distinguishing event types.

We describe the first experimental results for data-driven semantic parsing with Tree Rewriting Grammars (TRGs) and semantic frames. While several theoretical papers previously discussed approaches for modeling frame semantics in the context of TRGs, this is the first data-driven implementation of such a parser. We experiment with Tree Wrapping Grammar (TWG), a grammar formalism closely related to Tree Adjoining Grammar (TAG), developed for formalizing the typologically inspired linguistic theory of Role and Reference Grammar (RRG). We use a transformer-based multi-task architecture to predict semantic supertags which are then decoded into RRG trees augmented with semantic feature structures. We present experiments for sentences in different genres for English data. We also discuss our compositional semantic analyses using TWG for several linguistic phenomena.

Treebanking for local languages is hampered by the lack of existing parsers to generate pre-annotations. However, it has been shown that reasonably accurate parsers can be bootstrapped with little initial training data when use is made of the information in interlinear glosses and translations that language documentation data for such treebanks typically comes with. In this paper, we improve upon such a bootstrapping model by representing glosses using a combination of morphological feature vectors and pre-trained lemma embeddings. We also contribute a mapping from glosses to Universal Dependencies morphological features.

We introduce TS-ANNO, an open-source web application for manual creation and for evaluation of parallel corpora for text simplification. TS-ANNO can be used for i) sentence–wise alignment, ii) rating alignment pairs (e.g., w.r.t. grammaticality, meaning preservation, ...), iii) annotating alignment pairs w.r.t. simplification transformations (e.g., lexical substitution, sentence splitting, ...), and iv) manual simplification of complex documents. For evaluation, TS-ANNO calculates inter-annotator agreement of alignments (i) and annotations (ii).

In this paper, we investigate to which extent contextual neural language models (LMs) implicitly learn syntactic structure. More concretely, we focus on constituent structure as represented in the Penn Treebank (PTB). Using standard probing techniques based on diagnostic classifiers, we assess the accuracy of representing constituents of different categories within the neuron activations of a LM such as RoBERTa. In order to make sure that our probe focuses on syntactic knowledge and not on implicit semantic generalizations, we also experiment on a PTB version that is obtained by randomly replacing constituents with each other while keeping syntactic structure, i.e., a semantically ill-formed but syntactically well-formed version of the PTB. We find that 4 pretrained transfomer LMs obtain high performance on our probing tasks even on manipulated data, suggesting that semantic and syntactic knowledge in their representations can be separated and that constituency information is in fact learned by the LM. Moreover, we show that a complete constituency tree can be linearly separated from LM representations.

This paper describes the first release of RRGparbank, a multilingual parallel treebank for Role and Reference Grammar (RRG) containing annotations of George Orwell’s novel 1984 and its translations. The release comprises the entire novel for English and a constructionally diverse and highly parallel sample (“seed”) for German, French and Russian. The paper gives an overview of annotation decisions that have been taken and describes the adopted treebanking methodology. Finally, as a possible application, a multilingual parser is trained on the treebank data. RRGparbank is one of the first resources to apply RRG to large amounts of real-world data. Furthermore, it enables comparative and typological corpus studies in RRG. And, finally, it creates new possibilities of data-driven NLP applications based on RRG.

The paper presents a frame-based model of inherently polysemous nouns (such as ‘book’, which denotes both a physical object and an informational content) in which the meaning facets are directly accessible via attributes and which also takes into account the semantic relations between the facets. Predication over meaning facets (as in ‘memorize the book’) is then modeled as targeting the value of the corresponding facet attribute while coercion (as in ‘finish the book’) is modeled via specific patterns that enrich the predication. We use a compositional framework whose basic components are lexicalized syntactic trees paired with semantic frames and in which frame unification is triggered by tree composition. The approach is applied to a variety of combinations of predications over meaning facets and coercions.

In this paper we examine a BiLSTM architecture for disambiguating verbal potentially idiomatic expressions (PIEs) as to whether they are used in a literal or an idiomatic reading with respect to explainability of its decisions. Concretely, we extend the BiLSTM with an additional attention mechanism and track the elements that get the highest attention. The goal is to better understand which parts of an input sentence are particularly discriminative for the classifier’s decision, based on the assumption that these elements receive a higher attention than others. In particular, we investigate POS tags and dependency relations to PIE verbs for the tokens with the maximal attention. It turns out that the elements with maximal attention are oftentimes nouns that are the subjects of the PIE verb. For longer sentences however (i.e., sentences containing, among others, more modifiers), the highest attention word often stands in a modifying relation to the PIE components. This is particularly frequent for PIEs classified as literal. Our study shows that an attention mechanism can contribute to the explainability of classification decisions that depend on specific cues in the sentential context, as it is the case for PIE disambiguation.

This paper considers the task of parsing low-resource languages in a scenario where parallel English data and also a limited seed of annotated sentences in the target language are available, as for example in bootstrapping parallel treebanks. We focus on constituency parsing using Role and Reference Grammar (RRG), a theory that has so far been understudied in computational linguistics but that is widely used in typological research, i.e., in particular in the context of low-resource languages. Starting from an existing RRG parser, we propose two strategies for low-resource parsing: first, we extend the parsing model into a cross-lingual parser, exploiting the parallel data in the high-resource language and unsupervised word alignments by providing internal states of the source-language parser to the target-language parser. Second, we adopt self-training, thereby iteratively expanding the training data, starting from the seed, by including the most confident new parses in each round. Both in simulated scenarios and with a real low-resource language (Daakaka), we find substantial and complementary improvements from both self-training and cross-lingual parsing. Moreover, we also experimented with using gloss embeddings in addition to token embeddings in the target language, and this also improves results. Finally, starting from what we have for Daakaka, we also consider parsing a related language (Dalkalaen) where glosses and English translations are available but no annotated trees at all, i.e., a no-resource scenario wrt. syntactic annotations. We start with cross-lingual parser trained on Daakaka with glosses and use self-training to adapt it to Dalkalaen. The results are surprisingly good.

This paper addresses the question to which extent neural contextual language models such as BERT implicitly represent complex semantic properties. More concretely, the paper shows that the neuron activations obtained from processing an English sentence provide discriminative features for predicting the (non-)causativity of the event denoted by the verb in a simple linear classifier. A layer-wise analysis reveals that the relevant properties are mostly learned in the higher layers. Moreover, further experiments show that appr. 10% of the neuron activations are enough to already predict causativity with a relatively high accuracy.

English verb alternations allow participating verbs to appear in a set of syntactically different constructions whose associated semantic frames are systematically related. We use ENCOW and VerbNet data to train classifiers to predict the instrument subject alternation and the causative-inchoative alternation, relying on count-based and vector-based features as well as perplexity-based language model features, which are intended to reflect each alternation’s felicity by simulating it. Beyond the prediction task, we use the classifier results as a source for a manual annotation step in order to identify new, unseen instances of each alternation. This is possible because existing alternation datasets contain positive, but no negative instances and are not comprehensive. Over several sequences of classification-annotation steps, we iteratively extend our sets of alternating verbs. Our hybrid approach to the identification of new alternating verbs reduces the required annotation effort by only presenting annotators with the highest-scoring candidates from the previous classification. Due to the success of semi-supervised and unsupervised features, our approach can easily be transferred to further alternations.

We describe an approach to statistical parsing with Tree-Wrapping Grammars (TWG). TWG is a tree-rewriting formalism which includes the tree-combination operations of substitution, sister-adjunction and tree-wrapping substitution. TWGs can be extracted from constituency treebanks and aim at representing long distance dependencies (LDDs) in a linguistically adequate way. We present a parsing algorithm for TWGs based on neural supertagging and A* parsing. We extract a TWG for English from the treebanks for Role and Reference Grammar and discuss first parsing results with this grammar.

In this system demonstration paper, we present an open-source web-based application with a responsive design for modular semantic frame annotation (SFA). Besides letting experienced and inexperienced users do suggestion-based and slightly-controlled annotations, the system keeps track of the time and changes during the annotation process and stores the users’ confidence with the current annotation. This collected metadata can be used to get insights regarding the difficulty of an annotation with the same type or frame or can be used as an input of an annotation cost measurement for an active learning algorithm. The tool was already used to build a manually annotated corpus with semantic frames and its arguments for task 2 of SemEval 2019 regarding unsupervised lexical frame induction (QasemiZadeh et al., 2019). Although English sentences from the Wall Street Journal corpus of the Penn Treebank were annotated for this task, it is also possible to use the proposed tool for the annotation of sentences in other languages.

Supervised disambiguation of verbal idioms (VID) poses special demands on the quality and quantity of the annotated data used for learning and evaluation. In this paper, we present a new VID corpus for German and perform a series of VID disambiguation experiments on it. Our best classifier, based on a neural architecture, yields an error reduction across VIDs of 57% in terms of accuracy compared to a simple majority baseline.

In text simplification and readability research, several features have been proposed to estimate or simplify a complex text, e.g., readability scores, sentence length, or proportion of POS tags. These features are however mainly developed for English. In this paper, we investigate their relevance for Czech, German, English, Spanish, and Italian text simplification corpora. Our multi-lingual and multi-domain corpus analysis shows that the relevance of different features for text simplification is different per corpora, language, and domain. For example, the relevance of the lexical complexity is different across all languages, the BLEU score across all domains, and 14 features within the web domain corpora. Overall, the negative statistical tests regarding the other features across and within domains and languages lead to the assumption that text simplification models may be transferable between different domains or different languages.

This paper presents Unsupervised Lexical Frame Induction, Task 2 of the International Workshop on Semantic Evaluation in 2019. Given a set of prespecified syntactic forms in context, the task requires that verbs and their arguments be clustered to resemble semantic frame structures. Results are useful in identifying polysemous words, i.e., those whose frame structures are not easily distinguished, as well as discerning semantic relations of the arguments. Evaluation of unsupervised frame induction methods fell into two tracks: Task A) Verb Clustering based on FrameNet 1.7; and B) Argument Clustering, with B.1) based on FrameNet’s core frame elements, and B.2) on VerbNet 3.2 semantic roles. The shared task attracted nine teams, of whom three reported promising results. This paper describes the task and its data, reports on methods and resources that these systems used, and offers a comparison to human annotation.

Complex predicates formed of a semantically ‘light’ verbal head and a noun or verb which contributes the major part of the meaning are frequently referred to as ‘light verb constructions’ (LVCs). In the paper, we present a case study of LVCs with the German posture verb stehen ‘stand’. In our account, we model the syntactic as well as semantic composition of such LVCs by combining Lexicalized Tree Adjoining Grammar (LTAG) with frames. Starting from the analysis of the literal uses of posture verbs, we show how the meaning components of the literal uses are systematically exploited in the interpretation of stehen-LVCs. The paper constitutes an important step towards a compositional and computational analysis of LVCs. We show that LTAG allows us to separate constructional from lexical meaning components and that frames enable elegant generalizations over event types and related constraints.

We propose to tackle the problem of verbal multiword expression (VMWE) identification using a neural graph parsing-based approach. Our solution involves encoding VMWE annotations as labellings of dependency trees and, subsequently, applying a neural network to model the probabilities of different labellings. This strategy can be particularly effective when applied to discontinuous VMWEs and, thanks to dense, pre-trained word vector representations, VMWEs unseen during training. Evaluation of our approach on three PARSEME datasets (German, French, and Polish) shows that it allows to achieve performance on par with the previous state-of-the-art (Al Saied et al., 2018).

We present ongoing work on data-driven parsing of German and French with Lexicalized Tree Adjoining Grammars. We use a supertagging approach combined with deep learning. We show the challenges of extracting LTAG supertags from the French Treebank, introduce the use of left- and right-sister-adjunction, present a neural architecture for the supertagger, and report experiments of n-best supertagging for French and German.

We describe the TRAPACC system and its variant TRAPACCS that participated in the closed track of the PARSEME Shared Task 2018 on labeling verbal multiword expressions (VMWEs). TRAPACC is a modified arc-standard transition system based on Constant and Nivre’s (2016) model of joint syntactic and lexical analysis in which the oracle is approximated using a classifier. For TRAPACC, the classifier consists of a data-independent dimension reduction and a convolutional neural network (CNN) for learning and labelling transitions. TRAPACCS extends TRAPACC by replacing the softmax layer of the CNN with a support vector machine (SVM). We report the results obtained for 19 languages, for 8 of which our system yields the best results compared to other participating systems in the closed-track of the shared task.

We present a method for unsupervised lexical frame acquisition at the syntax–semantics interface. Given a set of input strings derived from dependency parses, our method generates a set of clusters that resemble lexical frame structures. Our work is motivated not only by its practical applications (e.g., to build, or expand the coverage of lexical frame databases), but also to gain linguistic insight into frame structures with respect to lexical distributions in relation to grammatical structures. We model our task using a hierarchical Bayesian network and employ tools and methods from latent variable probabilistic context free grammars (L-PCFGs) for statistical inference and parameter fitting, for which we propose a new split and merge procedure. We show that our model outperforms several baselines on a portion of the Wall Street Journal sentences that we have newly annotated for evaluation purposes.

Arabic dialects do not just share a common koiné, but there are shared pan-dialectal linguistic phenomena that allow computational models for dialects to learn from each other. In this paper we build a unified segmentation model where the training data for different dialects are combined and a single model is trained. The model yields higher accuracies than dialect-specific models, eliminating the need for dialect identification before segmentation. We also measure the degree of relatedness between four major Arabic dialects by testing how a segmentation model trained on one dialect performs on the other dialects. We found that linguistic relatedness is contingent with geographical proximity. In our experiments we use SVM-based ranking and bi-LSTM-CRF sequence labeling.

This paper describes the HHU system that participated in Task 2 of SemEval 2017, Multilingual and Cross-lingual Semantic Word Similarity. We introduce our unsupervised embedding learning technique and describe how it was employed and configured to address the problems of monolingual and multilingual word similarity measurement. This paper reports from empirical evaluations on the benchmark provided by the task’s organizers.

The automated processing of Arabic Dialects is challenging due to the lack of spelling standards and to the scarcity of annotated data and resources in general. Segmentation of words into its constituent parts is an important processing building block. In this paper, we show how a segmenter can be trained using only 350 annotated tweets using neural networks without any normalization or use of lexical features or lexical resources. We deal with segmentation as a sequence labeling problem at the character level. We show experimentally that our model can rival state-of-the-art methods that rely on additional resources.

Non-Negative Randomized Word Embedding We propose a word embedding method which is based on a novel random projection technique. We show that weighting methods such as positive pointwise mutual information (PPMI) can be applied to our models after their construction and at a reduced dimensionality. Hence, the proposed technique can efficiently transfer words onto semantically discriminative spaces while demonstrating high computational performance, besides benefits such as ease of update and a simple mechanism for interoperability. We report the performance of our method on several tasks and show that it yields competitive results compared to neural embedding methods in monolingual corpus-based setups.

This paper describes our system submission to the CogALex-2016 Shared Task on Corpus-Based Identification of Semantic Relations. Our system won first place for Task-1 and second place for Task-2. The evaluation results of our system on the test set is 88.1% (79.0% for TRUE only) f-measure for Task-1 on detecting semantic similarity, and 76.0% (42.3% when excluding RANDOM) for Task-2 on identifying finer-grained semantic relations. In our experiments, we try word analogy, linear regression, and multi-task Convolutional Neural Networks (CNNs) with word embeddings from publicly available word vectors. We found that linear regression performs better in the binary classification (Task-1), while CNNs have better performance in the multi-class semantic classification (Task-2). We assume that word analogy is more suited for deterministic answers rather than handling the ambiguity of one-to-many and many-to-many relationships. We also show that classifier performance could benefit from balancing the distribution of labels in the training data.

Nous présentons ici différents algorithmes d’analyse pour grammaires à concaténation d’intervalles (Range Concatenation Grammar, RCG), dont un nouvel algorithme de type Earley, dans le paradigme de l’analyse déductive. Notre travail est motivé par l’intérêt porté récemment à ce type de grammaire, et comble un manque dans la littérature existante.

Cet article étudie la relation entre les grammaires d’arbres adjoints à composantes multiples avec tuples d’arbres (TT-MCTAG), un formalisme utilisé en linguistique informatique, et les grammaires à concaténation d’intervalles (RCG). Les RCGs sont connues pour décrire exactement la classe PTIME, il a en outre été démontré que les RCGs « simples » sont même équivalentes aux systèmes de réécriture hors-contextes linéaires (LCFRS), en d’autres termes, elles sont légèrement sensibles au contexte. TT-MCTAG a été proposé pour modéliser les langages à ordre des mots libre. En général ces langages sont NP-complets. Dans cet article, nous définissons une contrainte additionnelle sur les dérivations autorisées par le formalisme TT-MCTAG. Nous montrons ensuite comment cette forme restreinte de TT-MCTAG peut être convertie en une RCG simple équivalente. Le résultat est intéressant pour des raisons théoriques (puisqu’il montre que la forme restreinte de TT-MCTAG est légèrement sensible au contexte), mais également pour des raisons pratiques (la transformation proposée ici a été utilisée pour implanter un analyseur pour TT-MCTAG).

Developing linguistic resources, in particular grammars, is known to be a complex task in itself, because of (amongst others) redundancy and consistency issues. Furthermore some languages can reveal themselves hard to describe because of specific characteristics, e.g. the free word order in German. In this context, we present (i) a framework allowing to describe tree-based grammars, and (ii) an actual fragment of a core multicomponent tree-adjoining grammar with tree tuples (TT-MCTAG) for German developed using this framework. This framework combines a metagrammar compiler and a parser based on range concatenation grammar (RCG) to respectively check the consistency and the correction of the grammar. The German grammar being developed within this framework already deals with a wide range of scrambling and extraction phenomena.

Multicomponent Tree Adjoining Grammars (MCTAG) is a formalism that has been shown to be useful for many natural language applications. The definition of MCTAG however is problematic since it refers to the process of the derivation itself: a simultaneity constraint must be respected concerning the way the members of the elementary tree sets are added. Looking only at the result of a derivation (i.e., the derived tree and the derivation tree), this simultaneity is no longer visible and therefore cannot be checked. I.e., this way of characterizing MCTAG does not allow to abstract away from the concrete order of derivation. Therefore, in this paper, we propose an alternative definition of MCTAG that characterizes the trees in the tree language of an MCTAG via the properties of the derivation trees the MCTAG licences.

Tree Adjoining Grammars (TAG) are known not to be powerful enough to deal with scrambling in free word order languages. The TAG-variants proposed so far in order to account for scrambling are not entirely satisfying. Therefore, an alternative extension of TAG is introduced based on the notion of node sharing. Considering data from German and Korean, it is shown that this TAG-extension can adequately analyse scrambling data, also in combination with extraposition and topicalization.