A Novel FrameNet-based Resource for the Semantic Web Volha Bryl, Sara Tonelli, Claudio Giuliano, Luciano Serafini Fondazione Bruno Kessler via Sommarive 18, 38123 Trento, Italy {bryl,satonelli,giuliano,serafini}@fbk.eu ABSTRACT FrameNet is a large-scale lexical resource encoding informa- tion about semantic frames (situations) and semantic roles. The aim of the paper is to enrich FrameNet by mapping the lexical fillers of semantic roles to WordNet using a Wikipedia- based detour. The applied methodology relies on a word sense disambiguation step, in which a Wikipedia page is as- signed to a role filler, and then BabelNet and YAGO are used to acquire WordNet synsets for a filler. We show how to represent the acquired resource in OWL, linking it to the existing RDF/OWL representations of FrameNet and WordNet. Part of the resource is evaluated by matching it with the WordNet synsets manually assigned by FrameNet lexicographers to a subset of semantic roles. Categories and Subject Descriptors I.2.4 [Computing Methodologies]: Artificial Intelligence— Knowledge Representation Formalisms and Methods ; I.2.7 [Computing Methodologies]: Artificial Intelligence—Nat- ural Language Processing Keywords Semantic web, FrameNet, WordNet, word sense disambigua- tion, OWL 1. INTRODUCTION FrameNet [13] has become one of the most important semantic resources encoding information about situations, the frames, and participants, the semantic roles, also called frame elements (FEs). It has been widely used in natu- ral language processing tasks, from textual entailment [1] to question answering [17]. However, some major issues have emerged regarding coverage and ontological consistency [12]. One crucial aspect that would require improvement is the in- formation about possible lexical fillers for FEs, which may impact on the performance of semantic frame parsers. In the last FrameNet version (1.5), about 40 semantic types Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. To copy otherwise, to republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee. SAC’12 March 25-29, 2012, Riva del Garda, Italy. Copyright 2012 ACM 978-1-4503-0857-1/12/03 ...$10.00. were defined to provide semantic constraints on FE fillers, e.g. the semantic type Sentient assigned to the Agent FE 1 . However, this information does not cover the whole FE set (only 54% of the FEs have a semantic type), is often very high-level (consider e.g. Physical entity type coupled with Instrument FE), and only some of the semantic types are mapped to a WordNet synset. Therefore, it could hardly be used during large-scale semantic analysis. In this work, we aim at enriching FE information by map- ping their lexical fillers to WordNet [5] using a Wikipedia- based detour. Our methodology relies first on a word sense disambiguation (WSD) step, in which a Wikipedia page is automatically assigned to nominal FE fillers. The direct mapping to WordNet is not possible as no WordNet-based corpus of considerable dimensions is available to train a WSD system. Then, we apply BabelNet [10], an existing resource that maps Wikipedia and WordNet, to create a repository of synsets for each FE. In order to improve Babel- Net coverage, we further link the fillers having no WordNet information but linked to Wikipedia with the corresponding concepts in YAGO [19] ontology, which is, in turn, linked to WordNet. The above methodology allows us to create a new com- plementary FrameNet-based resource: a repository of senses for semantic roles. The repository, on the one hand, is built by making use of the existing semantic web resources and techniques. On the other hand, the acquired resource is beneficial for a number of text processing tasks, primarily, for semantic frame annotation, and thus, it contributes to enriching and linking tasks, which are of high importance for the semantic web. In the paper, we also show how to represent the acquired resource in OWL language linking it to RDF/OWL representations of FrameNet and WordNet, thus making the repository available to the semantic web. The paper is structured as follows: in Section 2 we de- scribe previous approaches aimed at translating FrameNet into markups for the semantic web and at enriching it with additional semantic information. In Section 3 the method- ology for creating the synset repository is detailed and some relevant statistics are given. In Section 4 we present the word sense disambiguation system used for assigning a Wikipedia page to FE fillers. In Sections 5 and 6 we present the OWL representation of the sense repository and evalu- ate, illustrate and discuss the resource. 2. RELATED WORK 1 73 semantic types are listed in FrameNet 1.5. However, some of them refer to lexical units and not to FEs.