<?xml version="1.0" standalone="yes"?>
<Paper uid="P84-1110">
  <Title>SEMANIIC PARSING AS GRAPH LANGUAGE TRANSFORMATION - A MULIIDIMENSIONAL APPROACH TO PARSING HIGHLY INFLECTIONAL LANGUAGES</Title>
  <Section position="8" start_page="518" end_page="519" type="concl">
    <SectionTitle>
MAZN
</SectionTitle>
    <Paragraph position="0"> a Finnish question. Node properties are not shown.</Paragraph>
    <Section position="1" start_page="518" end_page="519" type="sub_section">
      <SectionTitle>
2.3 Specification of the graph language
</SectionTitle>
      <Paragraph position="0"> transformation The transformation is specified by an agenda of prioritized c-graphs. Initially, the agenda consists of a set of sentence independent &amp;quot;transformational&amp;quot; c-graphs (that, for example, transform passive clauses into active ones) and  sentence dependent c-graphs corresponding to the syntactico-semantic categories of the individual words in the sentence. For example, the c-graph of fig. 1.4 corresponds to nouns belonging to category NOUN-HUMAN. It tries to identify semantic case constituents by the productions corresponding to the arcs. Fig. 1.2 illustrates the production ADJ-ATTR (adjective attribute) used in the c-graph of fig. 1.4. The interpretation of the production is: If there is an adjective preceeding a noun in the same case and number the words are in semantic KIND relation with each other. As a whole, the agenda constitutes a modular, sentence dependent c-graph.</Paragraph>
      <Paragraph position="1"> Parsing is performed by interpreting the agenda. Different strategies could be applied here; the structure of the c-graphs depend on the choice. In our experimental system parsing is performed by interpreting the first c-graph in the agenda. The c-graohs are defined in such way that they interpret each other and glue morphological representations of words into the derivation graph (arcs (READWORD) and (PUTLAST) in fig. 1.4) until a grammatical semantic representation (or in ambiguous cases several ones) is reached.</Paragraph>
    </Section>
    <Section position="2" start_page="519" end_page="519" type="sub_section">
      <SectionTitle>
2.4 Linguistic and computational
</SectionTitle>
      <Paragraph position="0"> motivations Most influential linguistic theories and ideas behind our parser are dependence grammar, semantic case grammar, and the notion of &amp;quot;word expert&amp;quot; parsing. The idea is that the c-graphs of word categories actively try to find the dependents of the main words and identify in what semantic roles they are (cf. the ADJ-ATTR-production of fig. 1.2). In some cases it it useful to assign active role to dependents. The c-graphs serve as illustrative linguistic descriptions of the syntactico-semantic features of word categories and other fenomena.</Paragraph>
      <Paragraph position="1"> Computationally, our formalism and parsing scheme gives high expressive power but its time complexity is not high. Only potentially relevant productions are tried to use during parsing. Graphs are illustrative and can be used to express both procedural and declarative knowledge. New word category models can be added to the parser rather independently from the other models.</Paragraph>
      <Paragraph position="2"> Our small experimental graph grammar parser for Finnish (Hyv6nen 1983) is still liguistically quite naive containing some 150 lexical entries, 50 productions, and 50 c-graphs. A larqer subset of Finnish needs to be modelled in order to evaluate the approach properly. We are currently developing the graph grammar approch further by generalizing the formalism into hierarchic graphs. By this way, for example, large graph structures could be manipulated more easily as single entities and identical structures could have different interpretations in different contexts. Also, a more elaborate coroutine based control structure for interpreting the c-graphs is under developement. We feel that the idea of seeing parsing as a multidimensional transformation of relational graphs in stead of as a delinearization process of a string into a parse tree is worth investicating further.</Paragraph>
    </Section>
  </Section>
class="xml-element"></Paper>