Spanish FrameNet and FrameSQL
Carlos Subirats Rüggeberg, H. Sato · 2004
In this paper, we introduce the Spanish FrameNet Project which is creating an online lexical resource for Spanish, based on Frame Semantics and supported by corpus evidence. Spanish and English lexical units in the Emotion and Motion frames are compared and differences between the lexicalization patterns and constructions in both languages are described. This paper also introduces FrameSQL which is a web-based application to search and view the Spanish and Berkeley’s FrameNet data on the web browser. The application handles both seamlessly, showing on the same window the Spanish and English lexical units belonging to the same frame. It makes it easier to compare semantic structures of the two lexicons. 1 Spanish FrameNet is a research project which is sponsored by the Department of Science and Technology of Spain (Grant No. TIC2002-01338), and is developed at the Autonomous University of Barcelona in cooperation with the FrameNet Project, administered at the International Computer Science Institute in Berkeley, CA. The authors would like to thank their colleagues, Rocio Dones, Jordi Duran, Mercedes Garcia, Lidia Moya, and Marc Ortega for their assistance. Background to Spanish FrameNet The Spanish FrameNet Project (http://gemini.uab.es/SFN) is creating an online lexical resource for Spanish, based on Frame Semantics (Fillmore, 1982, 1985) and supported by corpus evidence. The starter will be available to the public by January 2006, and will contain at least 1000 lexical items –predicative verbs, nouns and adjectives– representative of a wide range of semantic domains. The aim is to document the range of semantic and syntactic combinatory possibilities (valences) of predicates in specific senses, through: • human approved and automatic annotated example sentences and • automatic capture and organization of the annotation results. The Spanish FrameNet (SFN) database will be in a platform-independent format, and it will be able to be displayed and queried via the web and other interfaces. The SFN database will act both as a dictionary and a thesaurus. The dictionary features include: • definitions, tables showing how frame elements are syntactically expressed in sentences containing each word, • annotated examples from the corpus: human approved and automatically annotated, and an alphabetical index. Like a thesaurus, words are linked to the semantic frames in which they participate, and frames, in turn, are linked to wordlists and to related frames. The basic assumption of Frame Semantics is that each word evokes a particular frame and possibly profiles some element or aspect of that frame. Semantic frames are schematic representations of situations involving various participants, props, and other conceptual roles, each of which is called a frame element (FE). The semantic arguments of a predicating word correspond to the frame elements of the frame (or frames) associated to that word. A frame semantic description of a lexical unit identifies the frames which underlie a given meaning and specifies the ways in which frame elements are realized in structures headed by the word (See Johnson et al., 2002). For example, consider the Judgement_communication frame which deals with communicating a positive or negative judgment of an Evaluee to an Addressee, e.g. alabar (praise) or criticar (criticize). This frame minimally includes the FEs Communicator, Evaluee and Addressee. Sentence (1) below is a canonical example of a verb in the Judgement_communication frame. (1) Max elogio a Eva ante los Max praised to Eva in-front-of the directivos de la empresa. directors of the company Max praised Eva before the company directors. Here, Max fills the role of Communicator; Eva is the Evaluee; and los directivos de la empresa is the Addressee. Note that the Addressee is expressed in (1) above, but it may not be realized in other sentences, as shown in sentence (2) below, where la actuacion de la empresa is the Evaluee, and the Addressee is not instantiated. (2) Sara denuncio la actuacion de la empresa. Sara reported the performance of the company Sara reported the company performance. Each frame element tag is part of a set of three tags, consisting of the frame element (Communicator, Evaluee, etc.), the grammatical function and the phrase type of the annotated constituent. The mappings between the semantic and syntactic information constitutes its valence. This information is given in the triples of annotation for the set of sentence types in which a given lexical unit occurs. The goal of Spanish FrameNet is to annotate corpus citations and to discover the valence patterns for a large number of words showing how those valence patterns are instantiated in actual sentences. Each Spanish FrameNet entry will provide links to other lexical resources, including Spanish EuroWordNet synsets and syntactic subcategorization frames. The project's deliverables will consist of the SFN database itself: lexical entries for individual word senses, frame descriptions, and annotated subcorpora. The SFN project is based on the evidence offered by a 330 million-word corpus which includes both New World (60%) and European Spanish (40%). The corpus is POS tagged and lemmatized with a tool that uses an electronic dictionary of Spanish of 600,000 forms, both single (92%), and multi-word lexical units (8%), basically multi-word nouns (85%), like bomba atomica (atomic bomb), carga de profundidad (depth charge), and multiword adverbs (9%) like a ciegas (unknowingly), por ahora (by now), etc. Multi-word verbs like tener en cuenta (to take into account) and lexicalized prepositional phrases with support verbs like estar de moda (to be in fashion) are tagged and lemmatized with transducers. SFN uses the Corpus Workbench software from the Institut fur Maschinelle Sprachverarbeitung of the University of Stuttgart for searching the corpus. The semantic and syntactic annotation is carried out by using the FNDesktop, the system developed by the Berkeley FrameNet Project. The input of the FNDesktop is composed by sentences that have been automatically extracted from the corpus, and then POS tagged and lemmatized (Subirats and Ortega, 2000). The extraction of subcorpora where predicates appear in all their relevant constructions provide annotators with examples of each possible syntactic configuration in which a given lexical item can occur. Annotators then select sentences for annotation that best illustrate the ways in which frame elements are realized syntactically. Figure 1 shows an actual sentence from the database annotated with the FNDesktop. Figure 1: Annotation of a sentence in the Judgement_communication frame 2 http://www.ims.uni-stuttgart.de/ Different lexicalization patterns in English and Spanish emotion predicates SFN is studying areas of the lexicon that parallel existing English FrameNet desriptions. Most of the frames defined so far are valid cross-linguistically, because frames are meant to charachterize conceptual structures at a basic level of description. Valence descriptions provided by SFN and FN can be used to study different lexicalization patterns in English and Spanish. Thus, for instance, sorprender (to surprise) in (3) is a Cause_emotion verb characterizing an event, in which an agent seeks to cause an emotion on an Experiencer. The reflexive verb sorprenderse (to get surprised) in (4) and the adjectival past participle sorprendido (surprised) with the support verb estar (to be) in (5) are two Experiencer_objet predicates in which the Experiencer is the subject and the Stimulus is the object. (4) Maria se sorprendio de que Juan Maria REFL surprised of that Juan cantase. Sang Maria got surprised when Juan sang. (5) Maria esta sorprendida de que Juan cante. Maria is surprised of that Juan sang Maria is surprised that Juan sang Both sorprenderse in (4) and sorprendido in (5) express parts of the complex event characterized by sorprender in (3): sorprenderse is an inchoative verb which characterizes the beginning of an event and sorprendido expresses the ongoing state which occurs after the above mentioned beginning. Therefore, sorprenderse and sorprendido are simpler parts of the complex event sorprender (cf. Subirats and Petruck, 2003). This analysis allows us to study the lexicalization pattern differences among English and Spanish emotion predicates. Both Spanish and English lexicalize the causative meaning with two verbs, namely, surprise and sorprender. On the contrary, there exists a difference in the lexicalization of the inchoative meaning: Spanish uses the reflexive verb sorprenderse, whereas English uses the construction made by get and the adjectival past participle surprised. As a result, English only has the LU surprised in the Experiencer_subject frame and Spanish has two LUs, that is, the reflexive verb sorprenderse, and the adjective sorprendido (cf. Figure 2). These differences can be verified thanks to FrameSQL, an application which allows to compare predicates or predicate-related (3) Juan sorprendio a Maria al Juan surprised to Maria on contarle la verdad. explaining-her the truth Juan surprised Maria by telling her the truth. constructions in the frames which share the same name, and therefore the same characteristics in English and Spanish.