<?xml version="1.0" standalone="yes"?>
<Paper uid="J89-1001">
  <Title>PARSING WITH FLEXIBILITY~ DYNAMIC STRATEGIES~ AND IDIOMS IN MIND</Title>
  <Section position="3" start_page="0" end_page="0" type="intro">
    <SectionTitle>
2 Computational Linguistics, Volume 15, Number 1, March 1989
</SectionTitle>
    <Paragraph position="0"> Oliviero Stock Parsing with Flexibility, Dynamic Strategies, and Idioms in Mind interactive environment has been built up to experiment with these ideas. Besides aspects present in other environments, e.g., D-PATR and the LFG workbench, our environment provides scope for the cognitive scientist, or the application-minded one dealing with a restricted sublanguage, to explore and define processing behaviour.</Paragraph>
    <Paragraph position="1"> In this paper, after an introductory example, we first outline the characteristics of linguistic knowledge for our parser and give an overview of the parser itself.</Paragraph>
    <Paragraph position="2"> Following that we describe the particular kind of unification used. The way non-determinism and disambiguation work is then discussed. This is followed by a brief description of our treatment of long-distance dependencies. After that comes a description of syntax-based idiom recognition, in our view one of the most relevant features of our work. A description of the environment built around the parser and of experiments carried on in defining dynamic strategies concludes the paper.</Paragraph>
    <Section position="1" start_page="0" end_page="0" type="sub_section">
      <SectionTitle>
1.1 AN INTRODUCTORY EXAMPLE
</SectionTitle>
      <Paragraph position="0"> We shall begin with an informal example to give the flavour of a basic part of our approach. Let us assume that the Italian sentence to be parsed is: Un itinerario molto noioso ai visitatori ha prescritto la guida, literally, &amp;quot;A very boring itinerary to the visitors has described the guide.&amp;quot; The sentence is topicalized, resulting in a quite common construction. The representations for the first noun phrase (NP) and the prepositional phrase (PP) are built up as parsing proceeds. The building includes both a functional and a semantic representation of the constituents, drawn from information explicited in the lexical entries. At the level of the subject (S) constituent being built, no commitment is taken for what will be the subject, even when the auxiliary ha is dealt with. The verb prescritto instead causes the merging of information available so far.</Paragraph>
      <Paragraph position="1"> Therefore some of the functions specified with the verb will find their values (notably the oblique object and the direct object functions). The constituent un itinerario molto noioso fails to be chosen as a subject, by virtue of the result of the interaction of the parser with a (separate, and not described in this paper) semantic component that is questioned at the moment of building a new semantic representation. In fact here the meaning of the proposed subject would not be compatible with the agent role for the verb; therefore the only possibility becomes the goal role. Note that interaction with semantics is possible because we have partial representations available at this point. After that, the analysis is guided by the context that has been built so far. There could be two possible constituents whose construction may begin with the word la (this word is ambiguous-one reading is an article, the other reading a clitic): an S and an NP. But these bottom-up proposals are checked and, as only an NP is expected, only that interpretation is carried out, and is ultimately unified with the previous representation to produce a complete representation of the sentence, with la guida as subject.</Paragraph>
      <Paragraph position="2"> With an algorithm that yields constituent structures at an initial stage and performs unification at a later stage, things would have been different. The dynamic interaction with semantics-based disambiguation, when the object is chosen, the selection of the correct interpretation of la guida later in the process would not have been straightforward; moreover, positional flexibility could have been achieved only at further cost and, finally, there would have been more redundancy. An approach based on strictly incremental unification would have needed instead either a lot of redundancy at the level of descriptive patterns or a useless tentative commitment to binding the first NP to information carried by the auxiliary.</Paragraph>
    </Section>
  </Section>
class="xml-element"></Paper>