Odin’s Runes: A Rule Language for Information Extraction

Odin’s Runes: A Rule Language for Information Extraction Marco A. Valenzuela-Esc´arcega, Gus Hahn-Powell, Mihai Surdeanu University of Arizona Tucson, AZ 85721, USA {marcov, hahnpowell, msurdeanu}@email.arizona.edu Abstract Odin is an information extraction framework that applies cascades of finite state automata over both surface text and syntactic dependency graphs. Support for syntactic patterns allow us to concisely define relations that are otherwise difficult to express in languages such as Common Pattern Specification Language (CPSL), which are currently limited to shallow linguistic features. The interaction of lexical and syntactic automata provides robustness and flexibility when writing extraction rules. This paper describes Odin’s declarative language for writing these cascaded automata. Keywords: rule-based, information extraction, cascade of finite state automata 1. Introduction We recently released Odin (Open Domain INformer), a novel rule-based information extraction (IE) frame- work (Valenzuela-Esc´arcega et al., 2015b). At the core of this framework is Odin’s Runes, our rule grammar lan- guage. The core feature of this language is supporting dif- ferent types of rules, e.g., operating over surface or syntac- tic structures, which can interact in the same grammar. At a high level, the design of this rule language follows the simplicity principles promoted by other natural language processing (NLP) toolkits, such as Stanford’s CoreNLP, which aim to “avoid over-design”, “do one thing well”, and have a user “up and running in ten minutes or less” (Man- ning et al., 2014). In particular, we aimed for the following desirable characteristics: Simplicity: The language extends familiar concepts from regular expressions and context free grammars. Expressivity: The rules capture complex constructs when necessary, such as: (a) nested structures, and (b) complex regular expressions over syntactic patterns for event argu- ments. Robustness: To recover from unavoidable syntactic errors, syntactic patterns can be used alongside token-based sur- face patterns that incorporate shallow linguistic features. Extensibility: The language is designed to be modular, i.e., new types of rules can be easily added to the language. We currently support rules based on syntactic and surface struc- tures, and we plan extensions over abstract meaning rep- resentation (AMR) (Banarescu et al., 2012) and semantic roles (Surdeanu et al., 2008). Importantly, all of these types of rules can operate within the same grammar. In this paper we summarize Odin’s Runes. However, given space limitations, this description is likely to be incom- plete. We recommend that interested readers examine the full manual (Valenzuela-Esc´arcega et al., 2015a), which has been made available as an arXiv document1. 1http://arxiv.org/abs/1509.07513v1 2. Related Work Since the advent of FASTUS (Appelt et al., 1993), most rule-based IE frameworks implement architectures rely- ing on a cascade of finite state automata (FSA). This ap- proach has proven capable of producing fast and robust parsers for unstructured text (Abney, 1996). The success of FSA cascades continues even today with systems such as GATE (Cunningham et al., 2002). FASTUS introduced the Common Pattern Specification Language (CPSL) as a formalism for specifying cascaded FSA grammars (Appelt and Onyshkevych, 1998). A gram- mar in CPSL is specified by defining a cascade of …

Content truncated to 500 words