<?xml version="1.0" standalone="yes"?>
<Paper uid="W04-0914">
  <Title>Semantic Forensics: An Application of Ontological Semantics to Information Assurance</Title>
  <Section position="3" start_page="0" end_page="0" type="intro">
    <SectionTitle>
2 Ontological Semantics in Brief
</SectionTitle>
    <Paragraph position="0"> ONSE contains several modules, with an ontology at the center; the other important modules are lexicons of languages and a fact repository, in which information about the world is stored, and, of course, the analyzer and generator. The analytical goal of ONSE is to produce a TMR for NL input as well as NL or other output for each TMR (see figure 1).</Paragraph>
    <Paragraph position="1">  The ontology is a tangled hierarchy (lattice) of concepts, beginning at the root ALL, branching into OBJECT, EVENT, and PROPERTY, and so forth. Each node of the hierarchy is a concept with a set of properties, many of which are inherited from its ancestors, and at least one property other than the IS-A property is distinguished from its parent node as well as from its sibling nodes. The ontological concept for PAY might therefore look like figure 2 (cf. Nirenburg and Raskin 2004: 196ff.).</Paragraph>
    <Paragraph position="2"> As we see, the IS-A and SUBCLASSES slots are filled with other ontological concepts, as are AGENT, THEME, and PATIENT, the case-role slots. VALUE, SEM, RELAXABLE-TO and DEFAULT are all facets of their slots.</Paragraph>
    <Paragraph position="3">  Lexicons contain the actual words of a language, in contrast to the ontology's universal, language-independent concepts. The entry for each word in the lexicon contains all possible senses of that word, labeled with a part of speech and a sense number. The lexical entry for the English word pay contains three senses, respectively pay-n1, pay-n2, and pay-v1. Each of the senses is then assigned, most importantly, the information about the acceptable syntactic environments for the sense, or SYN-STRUC, and information about the word's meaning, or SEM-STRUC. It is in SEM-STRUC that each lexical item is linked to one or more ontological concepts, or to literals. The lexical entry for the English adjective good looks something like figure 3.</Paragraph>
    <Paragraph position="4"> When a text is fed into the ONSE system, its lexical items are identified, as well as several TMR parameters, such as discourse relations including modalities, aspect, information about ordering and duration in time, style, and sets of concepts working together. The first step in building a TMR is finding meanings for heads of clauses in the syntactic representation of input, which are most commonly verbs. The TMR, however, will typically end up containing more event instances than there are verbs in the original text. After identifying these events, building the TMR is a (non-trivial) matter of fitting all the other information of the text into the filler slots of the events and the additional parameters. In figure 4 are the much-simplified TMRs for three related sentences, which demonstrate how small changes in texts affect TMRs.</Paragraph>
    <Paragraph position="5"> Who won the match?  The next section reviews the NL IAS applications discovered and explored--from initial steps to pilot implementations--in the ongoing effort to export NLP into computer and information security.</Paragraph>
  </Section>
class="xml-element"></Paper>