Lasha Abzianidze, Johannes Bjerva, Kilian Evang, Hessel Haagsma, Rik van Noord, Pierre Ludmann, Duc-Duy Nguyen, Johan Bos. Proceedings of the 15th Conference European Chapter Association for Computational Linguistics: Volume 2, Short Papers. 2017.

Hitomi Yanaka, Koji Mineshima, Daisuke Bekki, Kentaro Inui, Satoshi Sekine, Lasha Abzianidze, Johan Bos. Proceedings of the 2019 ACL Workshop BlackboxNLP: Analyzing and Interpreting Neural Networks for NLP. 2019.

Stephan Oepen, Omri Abend, Lasha Abzianidze, Johan Bos, Jan Hajic, Daniel Hershcovich, Bin Li, Tim O'Gorman, Nianwen Xue, Zeman. Proceedings of the CoNLL 2020 Shared Task: Cross-Framework Meaning Representation Parsing. 2020.

Hitomi Yanaka, Koji Mineshima, Daisuke Bekki, Kentaro Inui, Satoshi Sekine, Lasha Abzianidze, Johan Bos. Proceedings of the Eighth Joint Conference on Lexical and Computational Semantics (*SEM 2019). 2019.

Modeling the entailment relation over sentences is one of generic problems natural language understanding.In order to account for this problem, we design a theorem prover Natural Logic, logic whose terms resemble expressions.The based on an analytic tableau method and employs syntactically semantically motivated schematic rules.Pairing with preprocessor, which generates formulas Logic from linguistic expressions, results in proof system language.It shown that obtains comparable accuracy (≈...

Neural methods have had several recent successes in semantic parsing, though they yet to face the challenge of producing meaning representations based on formal semantics. We present a sequence-to-sequence neural parser that is able produce Discourse Representation Structures (DRSs) for English sentences with high accuracy, outperforming traditional DRS parsers. To facilitate learning output, we represent DRSs as sequence flat clauses and introduce method verify produced are well-formed...

Semantic parsing offers many opportunities to improve natural language understanding. We present a semantically annotated parallel corpus for English, German, Italian, and Dutch where sentences are aligned with scoped meaning representations in order capture the semantics of negation, modals, quantification, presupposition triggers. The semantic formalism is based on Discourse Representation Theory, but concepts represented by WordNet synsets thematic roles VerbNet relations. Translating...

The paper proposes the task of universal semantic tagging---tagging word tokens with language-neutral, semantically informative tags. We argue that task, its independent nature, contributes to better analysis for wide-coverage multilingual text. present initial version tagset and show (a) tags provide fine-grained information, (b) they are suitable cross-lingual parsing. An application tagging in Parallel Meaning Bank supports both these points as contribute formal lexical semantics their...

LangPro is an automated theorem prover for natural language. Given a set of premises and hypothesis, it able to prove semantic relations between them. The based on version analytic tableau method specially designed logic. proof procedure operates logical forms that preserve linguistic expressions large extent. %This property makes the easily obtainable from syntactic trees. %, in particular, Combinatory Categorial Grammar derivation nature proofs deductive transparent. On FraCaS SICK textual...

Reasoning over several premises is not a common feature of RTE systems as it usually requires deep semantic analysis.On the other hand, FraCaS collection entailment problems consisting multiple and covering semantically challenging phenomena.We employ tableau theorem prover for natural language to solve in way.The expressiveness type theory, transparency logic schematic nature inference rules make easy model phenomena.The efficiency proving also becomes when reasoning premises.After adapting...

The paper presents the IWCS 2019 shared task on semantic parsing where goal is to produce Discourse Representation Structures (DRSs) for English sentences. DRSs originate from Theory and represent scoped meaning representations that capture semantics of negation, modals, quantification, presupposition triggers. Additionally, concepts event-participants in are described with WordNet synsets thematic roles VerbNet. To measure similarity between two DRSs, they represented a clausal form, i.e....

Shared tasks are indisputably drivers of progress and interest for problems in NLP. This is reflected by their increasing popularity, as well the fact that new shared regularly emerge under-researched under-resourced topics, especially at workshops smaller conferences.The general procedures conventions organizing a task have arisen organically over time (Paroubek, Chaudiron, Hirschman, 2007, Section 7). There no consistent framework describes how should be organized. not harmful thing per...

Meaning banking--creating a semantically annotated corpus for the purpose of semantic parsing or generation--is challenging task. It is quite simple to come up with complex meaning representation, but it hard design representation that captures many nuances meaning. This paper lists some lessons learned in nearly ten years annotation during development Groningen Bank (Bos et al., 2017) and Parallel (Abzianidze 2017). The paper's format rather unconventional: there no explicit related work,...

Tackling Natural Language Inference with a logic-based method is becoming less and common. While this might have been counterintuitive several decades ago, nowadays it seems pretty obvious. The main reasons for such conception are that (a) methods usually brittle when comes to processing wide-coverage texts, (b) instead of automatically learning from data, they require much manual effort development. We make step towards overcome shortcomings by modeling data as abduction: reversing...

We investigate the effects of multi-task learning using recently introduced task semantic tagging. employ tagging as an auxiliary for three different NLP tasks: part-of-speech tagging, Universal Dependency parsing, and Natural Language Inference. compare full neural network sharing, partial what we term to share setting where negative transfer between tasks is less likely. Our findings show considerable improvements all tasks, particularly in which shows consistent gains across tasks.

Monotonicity reasoning is one of the important skills for any intelligent natural language inference (NLI) model in that it requires ability to capture interaction between lexical and syntactic structures. Since no test set has been developed monotonicity with wide coverage, still unclear whether neural models can perform a proper way. To investigate this issue, we introduce Entailment Dataset (MED). Performance by state-of-the-art NLI on new substantially worse, under 55%, especially...

Discourse Representation Theory (DRT) is a formal account for representing the meaning of natural language discourse. Meaning in DRT modeled via Structure (DRS), representation with model-theoretic interpretation, which usually depicted as nested boxes. In contrast, directed labeled graph common data structure used to encode semantics texts. The paper describes procedure dressing up DRSs graphs include new framework 2020 shared task on Cross-Framework and Cross-Lingual Parsing. Since one...