<?xml version="1.0" standalone="yes"?>
<Paper uid="C86-1062">
  <Title>Domain Dependent Natural Language Understanding</Title>
  <Section position="3" start_page="0" end_page="0" type="metho">
    <SectionTitle>
2 Meaning Formation in Sentences
</SectionTitle>
    <Paragraph position="0"> Semantically, a text is regarded as a specification (denotation) of a series of propositions. In natural languages, propositions can be expressed not only by sentences, but also by other syntactic structures such as noun groups, infinitive phrases and embedded sentences. Thus a single sentence may express several propositions. The goal in understanding a text is to extract its propositions and specify them in a formal language.</Paragraph>
    <Paragraph position="1"> A sentence can be characterized as the basic independent structure in the language. Relating language to formal logic, the meaning of a sentence can be described by a predicate which is identified by the head verb of the sentence. The arguments of the predicate are denoted by the constituents of the sentence. Such a representation is the basis for both systemic (Winograd 1983), c&amp;se (Fillmore 1968) and lexical-functional (Bresnan 1981) grammars.</Paragraph>
  </Section>
  <Section position="4" start_page="0" end_page="260" type="metho">
    <SectionTitle>
3 Meaning Representation
</SectionTitle>
    <Paragraph position="0"> The meaning representation language is based on a case system (Bruce 1975) inspired by Filhnore's notion of deep cases.</Paragraph>
    <Paragraph position="1"> Basically, a text is represented by a list of propositions, each consisting of a proposition type corresponding to a predicate name, and a list of cases corresponding to the arguments of the predicate. Contradictory to Fillmore's notion, proposition types are not verbs, but abstract concepts defined in the case frames of a case grammar. Furthermore, cases show semantic relationships between proposition types and abstract concepts. The case system (set of cases) is chosen in a somewhat adhoc way. The cases, which are supposed to be necessary in order to describe the domain of thermodynamic exercises, are included. The cases and their use are explained below : object object being affected by an action or event, or being described.</Paragraph>
    <Paragraph position="2"> instr instrument for action.</Paragraph>
    <Paragraph position="3"> stuff materia force the thing or event forcing an action.</Paragraph>
    <Paragraph position="4"> action action being referred to.</Paragraph>
    <Paragraph position="5"> attr physical attribute.</Paragraph>
    <Paragraph position="6"> referent the object being referred to in a description.</Paragraph>
    <Paragraph position="7"> reason reason for event.</Paragraph>
    <Paragraph position="8"> direction direction of change.</Paragraph>
    <Paragraph position="9"> descr description of object.</Paragraph>
    <Paragraph position="10"> spatiaLloc &amp;quot;spatial location&amp;quot;, includes a object, which describes a physical location, a relation to the location and a direction (from/to/at).</Paragraph>
    <Paragraph position="11"> value_loe &amp;quot;value location&amp;quot;, as spatial_loc, but including a value.</Paragraph>
    <Paragraph position="12"> temporaLloc &amp;quot;temporal location&amp;quot;, includes an indication of time and a relation to this.</Paragraph>
    <Paragraph position="13">  Besides the case system, the meaning representation language includes elements which make it possible to introduce or define physical objects, to connect or refer to objects, actions or events, and to relate propositions temporally. The syntax of the meaning representation language is :</Paragraph>
    <Paragraph position="15"> Notice that it is possible to reference propositions and object definitions through their identifiers. 'Context ret' specifies whether an object is introduced in the text (introd) or being referred (refer).</Paragraph>
    <Paragraph position="16"> As an example of the representation language consider the sentence : &amp;quot;The calorimeter contains 100 g water with the temperature 50 C'. The corresponding representation is :</Paragraph>
    <Paragraph position="18"> Ilere the calorimeter and the water are defined as physical objects and denoted by the identifiers C and W respectively.</Paragraph>
    <Paragraph position="19"> The cMorimeter is in definite form, it is referring, and the referent cannot be found. The proposition type &amp;quot;obj_attr_val&amp;quot; relates an attribute of an object to a value or quantity. Finally, the proposition type &amp;quot;contain&amp;quot; relates an object, which contains, to an object which is contained.</Paragraph>
  </Section>
  <Section position="5" start_page="260" end_page="261" type="metho">
    <SectionTitle>
4 Relating Syntax to Cases
</SectionTitle>
    <Paragraph position="0"> The transformation from syntactic structures to the meaning representation language is controlled by a case grammar.</Paragraph>
    <Paragraph position="1"> The case grammar specifies the correspondence between syntactic representations, based on sm'face grammatical roles of t)hrases and sentences, and case representations.</Paragraph>
    <Paragraph position="2"> The semanl;i(:al analysis of a sentence is based on its head verb, while the analysis of a noun group is be based on the head noun and also on adjective descriptors, genitive determiners and prepositional phrases. For each head verb, head noun, etc,, tile case granlmat' contains a case frame. A case frame consists of the following parts : cases, selection, constraints, extract and presence. The &amp;quot;cases-part&amp;quot; states what a phrase shall be transformed into by means of proposition types and cases. The &amp;quot;selections&amp;quot; relate elements in the case frame to syntactic constituents. &amp;quot;ConstrMnts&amp;quot; contain semantic constraints on elements of the case frame. &amp;quot;Extract&amp;quot; makes it possible to extract elements fi'om compound, or complex, semantic elements, and finally, &amp;quot;presence&amp;quot; specifies whether eonstiLuents are mandatory, elliptic or optional.</Paragraph>
    <Paragraph position="3"> As an example of a ease frame consider the verb to &amp;quot;rise&amp;quot; in combination with &amp;quot;temperature&amp;quot; or any other physical attribute. Some examples of sentences contannng rise are &amp;quot;The temperature rises 5 degrees&amp;quot; &amp;quot;The temperature of the liquid rises from 50 to 55 degrees&amp;quot; &amp;quot;The temperature rises&amp;quot; Observe that in the first sentence the object with the mentioned temperature is denoted by an elliptic reference. In the analysis of the sentence, it has to be found using the context, i.e. theprevious sentences. A case frame for &amp;quot;rise&amp;quot; is shown below. Here the selections &amp;quot;subject&amp;quot;, &amp;quot;sdir&amp;quot; and &amp;quot;prep&amp;quot; refer respectively to the subject, the direct object and prepositional phrases in a sentence. The constraint 'is_a(x,y)' means that x is of type y according to the taxonomy. 'has_attr(o,at)' means that the object o has the attribute at.</Paragraph>
    <Paragraph position="4"> rise : proposition etxange cases : object(O), attr(At), valueAoc(to,equal,Rv), value_lee (from,equal,Sv), valueAoc (relative,equal,Gv).</Paragraph>
    <Paragraph position="5"> selection : subjeet(Subj), sdir(nv), prep(from,Sv), prep(to,Cv), prep(with,Rv).</Paragraph>
    <Paragraph position="7"> Notice that the subject, which besides being a compound structure consisting of an attribute and an object, may alternatively take form of an attribute only (because O is ellipsed). The constituents of the subject are extracted by the ex attr and ex_obj predicates.</Paragraph>
    <Paragraph position="8"> The semantical analysis of a syntactic structure is carried out in a mixed bottom up - top down way. The formation of the meaning of a phrase progresses bottom up, while the control of its constituents (selectkm of them and constraints on them) progresses top down. Cenerally, when a case fl'ame is applied in the analysis of a phrase, the elements specified in the selection-part are matched with the constituents of the phrase. If an element has the same syntactical role as a constituent, the constituent is analysed, while possible constraints are imposed on it. The result of the analysis is a list of propositions derived fl'oln the phrase as well as the semantic element which the phrase denotes.</Paragraph>
    <Paragraph position="9"> To illustrate the semantical analysis consider the sentence : &amp;quot;the liquid in the container is transferred to the calorimeter&amp;quot;. Suppose the sentence is analysed in isolation, so that the definite descriptions cannot be solved. The case frames needed to analyse the sentence are : transfer : proposition transfer cases : object(O), spatial_lee(goal,in,G) selection sdir(O), prep(to,G).</Paragraph>
    <Paragraph position="10"> constraints : is_a(O,physical object), is_a(G,contai~ter).</Paragraph>
    <Paragraph position="11"> presence obligatory(O), elliptic(G).</Paragraph>
    <Paragraph position="12"> calorimeter : object(calorimeter).</Paragraph>
    <Paragraph position="13"> container object(container).</Paragraph>
    <Paragraph position="14"> liquid stuff(liquid).</Paragraph>
    <Paragraph position="15"> in : proposition contain cases : object(O), referent0~ ) selection head(R), prep(in,O) constraints : is a(O,container), is_.a(l~,physical object) presence : obligatory(O), obligatory(R).</Paragraph>
    <Paragraph position="16"> The syntactic structure of the sentence can be depicted as a tree where each node is numbered : sentence, i  the container The sentence analysis selects the case frame for the verb &amp;quot;transfer&amp;quot;. The case frame claims the mandatory occurenee of a direct object O which must be a physical object. Thus O  matches by i{,s syntactical role the constituent identified by node 3. When analysing the corresponding noun group, the case frame for the head noun &amp;quot;liquid&amp;quot; is used at first. The constraint of the noun group (being a physical object) is fullfilled, thus the analysis proceeds. The determiner combined with the head noun determines the context dependency of the liquid as referring. Since the referent cammt be found, an object definition is generated, though marking the object as referring.</Paragraph>
    <Paragraph position="17"> The prepositional phrase of the noun group is analysed by first selecting the case frame for the preposition &amp;quot;in&amp;quot;. In this frame, it is claimed that the head noun must be a physical object. This is already known to be true. Furthermore, the head of the prepositional phrase must specify an object in the class &amp;quot;container&amp;quot;. Thereafter the container is defined as an object and a proposition of type &amp;quot;contain&amp;quot; is generated. The result of the analysis of the noun group is the object identifier denoting the liquid and the &amp;quot;contain&amp;quot; - proposition. The rest of the analysis will not be commented. The representation of the sentence is :</Paragraph>
  </Section>
  <Section position="6" start_page="261" end_page="261" type="metho">
    <SectionTitle>
5 Context-Dependent Analysis
</SectionTitle>
    <Paragraph position="0"> The context-dependent analysis covers resolution of the most important types of anaphoric references. The system resolves the following types of references in a text : identic, synonymous, pronominal, adverbial, modifying and some of the elliptic references. Examples of these references are : Identic a calorimeter contains helium, and the calorimeter ...</Paragraph>
    <Paragraph position="1"> Synonymous a calorimeter contains helium, and the gas Pronomial a' calorimeter contains helium, and it ...</Paragraph>
    <Paragraph position="2"> Adverbial in the calorimeter is gas, and there is also ...</Paragraph>
    <Paragraph position="3"> Modifying the calorimeter is heated to 50 C, and the heating ...</Paragraph>
    <Paragraph position="4"> Elliptic the calorimeter contains gas and the bucket (contains) water.</Paragraph>
    <Paragraph position="5"> Elliptic a calorimeter contains water. The temperature (in the calorimeter or of the water) is 50 C.</Paragraph>
    <Paragraph position="6"> During the semantical analysis, the references are resolved as soon as they are met. In order to be able to do this, the leftmost context of a text must be reachable when analysing a phrase. The leftmost context is all propositions derived from the text so far.</Paragraph>
    <Paragraph position="7"> The system uses no special features for delimiting the scope of referred objects. When a reference is to be solved, the objects and events specified in the leftmost context are examined. An object or event, which fullfills the constraints specified in the case frame and which matches possible syntactic features (gender and number), is claimed to be the token referred to. The resolution of synonymous references (for instance of gas in : &amp;quot;A container contains llelium, and the gas ... &amp;quot;) uses the is-a hierarchy.</Paragraph>
  </Section>
class="xml-element"></Paper>