The MITRE logical form generation system

The MITRE Logical Form Generation System Samuel Bayer and John Burger and Warren Greiff and Ben Wellner The MITRE Corporation Bedford, MA 01730 � sam,john,greiff,wellner � @mitre.org Abstract In this paper, we describe MITRE’s contribution to the logical form generation track of Senseval-3. We begin with a description of the context of MITRE’s work, followed by a description of the MITRE sys- tem and its results. We conclude with a commentary on the form and structure of this evaluation track. 1 Introduction The logic form identification track of the 2004 Sen- seval evaluation requires its participants to produce a version of each input sentence with each input word in citation form, annotated with both a scope- free Davidsonian logic and lexical category infor- mation for major categories. The output ignores el- ements like determiners and negation, and features such as plurals and verb tenses. This evaluation is of interest to the MITRE Cor- poration because it has a long-standing interest in text processing and understanding, in all its various dimensions. In our current internally funded Read- ing Comprehension (RC) project, we focus on the detailed understanding of individual stories, using the ability to answer comprehension questions as- sociated with these stories as our evaluation metric. At the moment, we are interested in getting a sense of how much inference is routinely needed in order to answer RC questions; so generation of sentence meanings is not currently our research focus. How- ever, in the context of our exploration, we continue to maintain an automated system for producing sen- tence meanings from text. 2 The MITRE logic generation system The system which MITRE employed for the Senseval-3 logical form evaluation consists of the following components: � the Humphreys/Carroll/Minnen morphological analyzer (Minnen et al., 2001) � the CMU Link Grammar parser (Sleator and Temperley, 1991) � a link interpretation language which is used to produce a dependency graph � additional lexical knowledge sources � an argument canonicalizer based partially on the principles of Relational Grammar (Perl- mutter, 1983) � a task-specific logical form generator The morphological analyzer is straightforward, and we will not say more about it. We discuss the remaining components below. 2.1 The CMU Link Grammar parser The Link Grammar formalism consists of labeled, undirected links among pairs of words. Each word in the Link Grammar dictionary is mapped to a com- plex logical expression of the link ends the word can participate in. These link ends have a major compo- nent (indicated by uppercase letters), a minor com- ponent (indicated by lowercase letters), and a re- quired direction (looking leftward (-) or rightward (+)). Two words can be joined by a link if their link ends are compatible. The Link Parser provides rea- sonable performance achieving 75% labeled con- stituent accuracy on the TreeBank data. There are a large number of link types some of which pro- vide very detailed distinctions beyond those found in phrase structure grammars. For further details, see (Sleator and Temperley, 1991). Figure 1 shows the …

Content truncated to 500 words