================================================================================= VerbAtlas 1.0 http://verbatlas.org Andrea Di Fabio, Simone Conia and Roberto Navigli Sapienza NLP Group, Sapienza University of Rome http://nlp.uniroma1.it Web site and resource maintenance by Babelscape http://babelscape.com ================================================================================= VerbAtlas is a novel large-scale manually-crafted semantic resource for wide-coverage, intelligible and scalable Semantic Role Labeling. The goal of VerbAtlas is to manually cluster WordNet synsets that share similar semantics into a set of semantically-coherent frames. VerbAtlas is licensed under the CC BY-NC-SA 4.0 License. ================================================================================= PACKAGE CONTENTS ================================================================================= * README.txt (this file); * LICENSES.txt (terms and conditions for the files provided in this package) * VERBATLAS_LICENSE.txt (terms and conditions of the CC BY-NC-SA 4.0 License); * WORDNET_LICENSE.txt (original license from WordNet 3.0); * BABELNET_LICENSE.txt (original license from BabelNet 4.0); * VerbAtlas-1.0/ (contains the files for VerbAtlas 1.0) * VA_frame_ids.tsv (name and id of each VerbAtlas frame) * VA_va2pas.tsv (the argument structure of each VerbAtlas frame) * VA_bn2va.tsv (the synsets in each VerbAtlas frame) * VA_preference_ids.tsv (name, id, reference synset for each preference) * VA_bn2sp.tsv (preferences for each synset) * VA_bn2shadow.tsv (shadow arguments for each synset) * VA_bn2implicit.tsv (implicit arguments for each synset) * pb2va.tsv (mapping from PropBank to VerbAtlas) * bn2wn.tsv (one-to-one mapping from BabelNet to WordNet) * wn2lemma.tsv (mapping from WordNet synset to lemmas) * wn2sense.tsv (mapping from WordNet synset to WordNet sense key) Every file in this package is licensed under the CC BY-NC-SA 4.0 License but wn2lemma.tsv and wn2sense.tsv, which are licensed under the WordNet 3.0 License, and bn2wn.tsv, which is licensed under the BabelNet 4.0 license. ================================================================================= RESOURCES ================================================================================= VerbAtlas relies on: * WordNet 3.0 (https://wordnet.princeton.edu/download/current-version) * BabelNet 4.0.1 (http://babelnet.org) * Proposition Bank I (PropBank - http://propbank.github.io/) ================================================================================= FORMAT ================================================================================= 1. VA_frame_ids.tsv Each line contains the ID and the name of a VerbAtlas frame, separated by a tab. For example: va:0001f TOLERATE va:0002f OBEY va:0003f CARRY_TRANSPORT ... 2. VA_va2pas.tsv Each line contains the ID of a VerbAtlas frame and its argument structure, where each element is separated by a tab. For example: va:0001f Agent Theme Beneficiary Attribute va:0002f Agent Theme va:0003f Agent Theme Destination Source Instrument ... ... 3. VA_bn2va.tsv Each line contains a BabelNet synset ID and its corresponding VerbAtlas frame ID, separated by a tab. For example: bn:00082138v va:0001f bn:00082200v va:0001f bn:00082203v va:0001f ... 4. VA_preference_ids.tsv Each line contains the ID of a VerbAtlas selectional preference, its reference BabelNet synset ID, and its name. For example: va:0001p bn:00000467n absorbent va:0002p bn:00000492n abstraction va:0003p bn:00000902n acid ... 5. VA_bn2sp.tsv Each line contains the selectional preferences for each BabelNet synset, separated by a tab. If a role supports multiple selectional preferences, they are separated by a pipe (|). For example: bn:00015081n Patient va:0084p|va:0083p Instrument va:0084p|va:0083p bn:00082116v Agent va:0081p|va:0103p Topic va:0106p|va:0026p ... bn:00082117v Agent va:0084p|va:0083p ... ... 6. VA_bn2shadow.tsv Each line contains the shadow arguments, if any, for each BabelNet synset ID. A synset may support multiple shadow arguments in different roles. For example: bn:00082230v Theme bn:00070772n bn:00082124v Stimulus bn:00030466n bn:00082266v Result bn:00098546a Instrument bn:00000902n ... 7. VA_bn2implicit.tsv Each line contains the implicit arguments, if any, for each BabelNet synset ID. A synset may support multiple implicit arguments in different roles. A role may have multiple implicit arguments, separated by a pipe (|). For example: bn:00082144v Patient bn:00011769n bn:00082205v Theme bn:00067181n Source bn:00021015n bn:00082162v Patient bn:00035378n|bn:00035596n|bn:00036686n ... 8. pb2va.tsv A mapping from PropBank to VerbAtlas. Each line maps a PropBank predicate sense and its corresponding argument structure to a VerbAtlas frame and its argument structure as follows: ... For example: abandon.01 va:0255f A0 Agent A1 Theme A2 Attribute abandon.02 va:0016f A0 Agent A1 Patient A2 Result abandon.03 va:0253f A0 Agent A1 Patient A2 Recipient ... 9. bn2wn.tsv A mapping from BabelNet synset to WordNet synset for verbs. Each line contains a BabelNet synset ID and its corresponding WordNet synset ID, separated by a tab. For example: bn:00082116v wn:00865776v bn:00082117v wn:02168378v bn:00082118v wn:02228031v ... 10. wn2lemma.tsv A mapping from a WordNet synset to a lemma. Each line contains a WordNet synset ID and a lemma, separated by a tab. NOTE: the same WordNet synset ID may appear in multiple lines. NOTE: the same lemma may appear in multiple lines. For example: wn:00001740v breathe wn:00001740v take-a-breath wn:00002325v respire wn:00002573v respire ... 11. wn2sense.tsv A mapping from a WordNet synset to a WordNet sense key. Each line contains a WordNet synset ID and a WordNet sense key, separated by a tab. NOTE: the same WordNet synset ID may appear in multiple lines. For example: wn:00001740v breathe%2:29:00:: wn:00001740v take_a_breath%2:29:00:: wn:00002325v respire%2:29:01:: wn:00002573v respire%2:29:02:: ... ================================================================================= REFERENCE PAPER ================================================================================= When using this resource, please refer to the following paper: Andrea Di Fabio, Simone Conia and Roberto Navigli "VerbAtlas: a Novel Large-Scale Verbal Semantic Resource and Its Application to Semantic Role Labeling" In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP), Hong Kong, China, November 3-7, 2019, pages 627-637. ================================================================================= CONTACTS ================================================================================= If you have any enquiries, please contact: Andrea Di Fabio - Sapienza Università di Roma (difabio [at] di [dot] uniroma1 [dot] it) Simone Conia - Sapienza Università di Roma (conia [at] di [dot] uniroma1 [dot] it) Roberto Navigli - Sapienza Università di Roma (navigli [at] di [dot] uniroma1 [dot] it)