<?xml version="1.0" standalone="yes"?>
<Paper uid="C80-1067">
  <Title>ENGLISH-JAPANESE TRANSLATION THROUGH CASE-STRUCTURE CONVERSION</Title>
  <Section position="1" start_page="0" end_page="0" type="metho">
    <SectionTitle>
ENGLISH-JAPANESE TRANSLATION THROUGH CASE-STRUCTURE CONVERSION
</SectionTitle>
    <Paragraph position="0"> This paper reports some trials on mechanical translation from English to Japanese through a case structure constructed on Hornby~s verb patterns. Though the general theoryof case structures is still at the beginning of study, it provides partial sentential patterns with rough but resonable classification labels. After determination of schematic dependency relations, multi-vocal problems for choosing appropriate equivalents are dissolved using subcategories of terms and cases.</Paragraph>
    <Paragraph position="1"> Case structures of English are transformed into those of Japanese if necessary, and from those , Japanese sentences are generated by a Japanese grammar.</Paragraph>
    <Paragraph position="2"> i. Introduction English and Japanese are fairly different languages from each other. Accordingly, it will be necessary to make translation based on a kind of deep structures of sentences which determines sentential forms schematically in respective languages. The case structure used here is a conventional one which is closely related to surface structures. But it was found that the case structure is fairly useful by adding some new means in construction and usage of it.</Paragraph>
    <Paragraph position="3"> The parsing for obtaining the case structure from English sentences is done by category-matching on a fairly precise verb pattern given by A.S. Hornby. It dissolves most of multi-syntactic and semantic category problems of words and phrases, and determines global dependency relations among constituents of English sentences.</Paragraph>
    <Paragraph position="4"> Subsequently, it determines the appropriate Japanese equivalents and post-positions among several candidates by using subcategories of terms and individual prepositional words.</Paragraph>
    <Paragraph position="5"> The case structure thus obtained containes the modal part which represents various characteristics of the predicate part in almost independent form of respe- THINGS ctive languages. It also embeds idiomatic expressions in itself.</Paragraph>
    <Paragraph position="6"> At the level of the case structure, the difference between both language expressions is not so large. Transformations of sentential forms such as those of non-livlng subjects and those of principal clauses followed by long subordlnate ones for readability are processed at this stage without difficulty. From the case structures transformed into Japanese, translated sentences are generated under control of various involved</Paragraph>
  </Section>
  <Section position="2" start_page="0" end_page="0" type="metho">
    <SectionTitle>
-EVENT -
</SectionTitle>
    <Paragraph position="0"> parameters.</Paragraph>
  </Section>
  <Section position="3" start_page="0" end_page="0" type="metho">
    <SectionTitle>
2. Case structures
</SectionTitle>
    <Paragraph position="0"/>
    <Section position="1" start_page="0" end_page="0" type="sub_section">
      <SectionTitle>
2.1 Case structures of verbs
</SectionTitle>
      <Paragraph position="0"> Many English words have multi-syntactic and semantic categories. However, these multi-categories of a word in a sentence should be able to be dis -solved in a context of the sentence, namely, in syntactic and semantic relations to the neighboring words. A.S.Hornby has given comparatively precise verb patterns. 3The use of verb patterns at parsing is fairly effective to identify the syntactic categories of words and phrases as well as their syntactic roles such as complements and adverbial infinitives.</Paragraph>
      <Paragraph position="1"> On the other hand, an efficient introduction of semantic information about words is very important in order to identify their semantic roles and appropriate Japanese equivalents as well as to dissolve the still remaining syntactic ambiguities after a syntactic analysis. For this purpose, there are several approaches based on the case grammar. 4 It is still in an incomplete state for general purpose of thorough comprehension of sentences. But if it is used on detailed syntactic rules of the conventional word usage with the aids of categories of an appropriate level at various stages of mechanical translation, these approaches based on the case grammar will serve fairly well to assign reasonable  semantic roles to various constituents in a source sentence as well as to choose their appropriate equivalents and sentential forms in the target language.</Paragraph>
      <Paragraph position="2"> From the above considerations, a case structure is assigned to a sentence based on Hornby's verb patterns.</Paragraph>
      <Paragraph position="3"> Events of discourse are classified by categories of predicates. Fig.l shows the upper part of the categorical tree used here. The categories used for describing case structures are, in many cases, those appearing at the rightmost in Fig.l.</Paragraph>
      <Paragraph position="4"> Semantic roles of constituents are described using case labels shown in Table i.</Paragraph>
      <Paragraph position="5"> Table i Case labels PREDicate, AGent, OBJect, EXPERiencer, RECIPient, SOurce, GOal, INSTRument, DEGree, COMPARison, LOCation, TIME, RANGE, MANNer, ROLE, CAUSE, RESULT, MEANS, PURPOSE, CONDition, MODal ( FORM, TENSE, ASPECT, AFfirmation, MOOD, VOICE ), deg.. By using several pairs of categories and case -labels chosen from the above the main parts of each event are schematically described and called case-frames of events. The first row of Table 2 is the main case-frame of the PTRANS event.</Paragraph>
      <Paragraph position="6"> The m~in part of a case structure based on a case-frame is transformed into a surface structure using a verb which has a sufficient number of  links to describe the main part and belongs to an appropriate verb pattern. Table 2 shows verb patterns used for English description of PTRANS. On the other hand, if a verb pattern is given, pairs of categories and case labels of dependants are almost determined every category of the predicate verb as shown in Table 3. Hence, if the verb patterns as well as the categories of a verb are recorded in a word dictionary, the possible case structure can be retrieved at parsing immediately. Table 4 shows a part of pairs of verb patterns and categories about ~take ~.</Paragraph>
      <Paragraph position="7"> Example i shows several examples of case structures obtained by parsing with the aids of Table  After determination of a case structure or a dependency relation of a sentence, there still remain several problems to be solved.</Paragraph>
      <Paragraph position="8"> One of them is the choice of an appropriate equivalent or a subcategory. The word rtake~ with the verb pattern VP6A and the case structure has fairly many Japanese equivalents. However, this kind of multi-vocal problems can be dissolved in many cases by looking up subcategories of a few designated dependants involved in the case structure and hence in advance by recording them for each equivalent in the word dictionary as shown in Table 4.</Paragraph>
      <Paragraph position="9"> The other is the choice of an appropriate Japanese postposition corresponding to a case label and a category of a dependant of a verb.</Paragraph>
      <Paragraph position="10"> Though a good correspondence generally holds between them, there are some exceptions. For these cases, the postposition is recorded at the item of the associated verb in the dictionary and picked up at parsing.</Paragraph>
    </Section>
  </Section>
  <Section position="4" start_page="0" end_page="0" type="metho">
    <SectionTitle>
DEG- PURPOSE MANN
QUANT -ACT -ACT
</SectionTitle>
    <Paragraph position="0"> VP6A BOD.ACT i. get into one's hand ( OBJ-PHYS-OBJ ) 2. eat ( OBJ-FOOD ) 3. get on ( OBJ-VEHICLE ) PTRANS I. go carrying ( OBJ-NON.LIVING ) 2. go leading ( OBJ-LIVING ) POSS'TRANS i. receive ( OBJ-COMMODITY ) MTRANS i. make a record of ( OBJ-MENT. OBJ : U EVENT ) VP22 PERCEP.ACT 1. suppose ( OBJ-EVENT ) (4) VP22 : PERCEPtual ACTion ~I took her intelligent. ~ (PRED-PERCEP.ACT:suppose, MOD:(TENSE:past, VOICE:active), EXPER-HUM:I, OBJ-EVENT: (PRED-ATTR:intelligent, OBJ-HUM:she))</Paragraph>
    <Section position="1" start_page="0" end_page="0" type="sub_section">
      <SectionTitle>
2.2 Optional cases
</SectionTitle>
      <Paragraph position="0"> Adverbial phrases and clauses outside verb patterns constitute optional cases. Most of idiomatic phrase prepositions such as rwith respect to p and many subordinate conjunctions such as Cwhen' and Calthough * determine the case labels and the Japanese postpositions of the following phrases and clauses. On the other hand, many prepositional phrases which respectively have only one prepositions such as c with ~ and Cfor' require the information about frame of discourse for deter i mination of the case label and the appropriate Japanese postposition. These information is the categories of the object term of the phrase and those of main verb of the governor and furthermore, in some cases, other information about case structure of the verb pattern.</Paragraph>
      <Paragraph position="1"> For example, an adverbial phrase consisting of a prefixed preposition ~ in' and an object term with category~PHYSical LOCation' can be assigned a case label'LOC' or CGOal', but the appropriate postposition and case label are determined by referring to the semantic category of the main verb.</Paragraph>
      <Paragraph position="2"> The necessary information for determining them are recorded each preposition and subordinate conjunction in the word dictionary.</Paragraph>
    </Section>
    <Section position="2" start_page="0" end_page="0" type="sub_section">
      <SectionTitle>
2.3 Case structures of noun phrases and clauses
</SectionTitle>
      <Paragraph position="0"> Case structures of noun phrases and clauses have similar forms to those of sentences as follows; t (Kr-Cl :t~ ,.. * ,K6-CL:t,&amp;quot; * * ,K~-Cn:t~ ) (i) where t is the main term, ~ s are the dependants and K~-~ is a pair of a case label and a category.</Paragraph>
      <Paragraph position="1"> The case structure only consisting of several individual objects and prepositions are determined by referring to the categories of the objects and their syntactic patterns linked with the individual prepositions.</Paragraph>
      <Paragraph position="2"> If a noun phrase contains some verbal nouns, the case structure is constructed based on the verbalized words of some of them.</Paragraph>
      <Paragraph position="3">  Example 2 (i) Creceipt of his letter ~ C-EVENT: receive(PRED-POSS. TRANS: *, OBJ-PHYS.OBJ:Ietter(OBJ: *, POSS-HUM:his)) (2) t punctual arrival of trains'  C-EVENT: arrive(PRED-PTRANS: *, AG-VEHICLE: trains, MANN-MODAL:punctual) where C denotes a certain case for the outside governor and the symbol * denotes the prefixed term to the case frame containing it.</Paragraph>
    </Section>
  </Section>
  <Section position="5" start_page="0" end_page="0" type="metho">
    <SectionTitle>
3. Parsing
</SectionTitle>
    <Paragraph position="0"> First, processing of negative expressions and idioms are described. They are processed in a somewhat special manner.</Paragraph>
    <Section position="1" start_page="0" end_page="0" type="sub_section">
      <SectionTitle>
3.1 Negative expressions
</SectionTitle>
      <Paragraph position="0"> Negative expressions of English are fairly different from those of Japanese at the points where negative adjectives or adverbs of English correspond to some negative modifiers and a negative auxiliary verb in Japanese. Hence, it  will be better to express the negative expression of an English sentence in a language-free form at the modal part of the case structure and to generate the corresponding Japanese sentence from that anew. The types of NEGation are classified to INTensified TOTal,TOTal and PARtial ones.</Paragraph>
      <Paragraph position="1"> At parsing, the value of type is put in the modal part of the case structure together with the terms directly negated except the case when the terms belong to the predicate part of the sentence.</Paragraph>
      <Paragraph position="2"> Example 3  (i) ~No student can solve the problem~ (PRED-THINK.ACT:solve, MOD:(AF:( INT. TOT.NEG, student), MOOD:can), AG-HUM:student, OBJ -MENT. OBJ:problem(OBJ:*, DET:def)) (2) ~Converses are not always true.' (PRED-ATTR:true, MOD:(AF:(PAR'NEG, always)), OBJ-MENT.OBJ:converses, FREQuency: always) Most of terms of modal cases are omitted here and hereafter except important terms for simplicity. null</Paragraph>
    </Section>
    <Section position="2" start_page="0" end_page="0" type="sub_section">
      <SectionTitle>
3.2 Idiomatic expressions
</SectionTitle>
      <Paragraph position="0"> For an efficient parsing, idiomatic expressions are processed with some priority. They are recorded near the heading of the leftmost wnrd among multivocal constituents of the idiom in the word dictionary. If the idiom has a separated form tl~t z as shown in Eq.(2),</Paragraph>
      <Paragraph position="2"> the syntactic and semantic conditions on the strings of words or phrases sl, * ..,s~ lying between tl and t 2 are recorded together with the partial case structure and the Japanese equivalent of the idiom in the word dictionary as follows; so ~ that ~ (DEG-QUALity:so\[ADVerb\] very*, PRED-ATTR: ~f \[ADJective\], RESULT(that\[SUBORDInate CONJunction\])-EVENT: ~\[SENTence\]), where ~very*~denotes the Japanese word equivalent to the English word ~so ~.</Paragraph>
      <Paragraph position="3"> When a constituent word. of an idiom is found in a sentence at parsing, it is examined whether~ or not there exist the rest parts of the idiom in the sentence by reading the sentence ahead or by examining the partial case structure already constructed. If there is, and if the intermediate part sl --. s~ satisfies specified conditions by parsing, the partial case structure and the Japanese equivalent of the idiom are added to those already constructed by referring to the word dictionary. Otherwise, ordinary mode of parsing restarts from term t I</Paragraph>
    </Section>
    <Section position="3" start_page="0" end_page="0" type="sub_section">
      <SectionTitle>
3.3 Construction of case structures
</SectionTitle>
      <Paragraph position="0"> The rewriting rules used for parsing are described in a Chomsky-like form as follows~</Paragraph>
      <Paragraph position="2"> symbol, ~- ~ : ~ a sequent of several pai~s of a case label and a category followed by a term, and VPL a label of a verb pattern.</Paragraph>
      <Paragraph position="3"> Eq.(3) represents the type of rewriting rules of a subject and a predicate part, and Eq.(4) that of a predicate part and a right dependent.</Paragraph>
      <Paragraph position="4"> The parsing is done from left to right in a bottom-up manner and case structures are constructed in the order of the predicate part, the right part of a verb pattern of the predicate part, the subject part and the optional adverbial parts. This order is considered to be efficient for restricting possible construction of case structures of many sentences of English because the central structure is generally constructed in this order.</Paragraph>
      <Paragraph position="5"> At first, the partial case structure of the predicate part is constructed. If the main verb is a constituent of an idiom, parsing is done in a manner described in the preceding section.</Paragraph>
      <Paragraph position="6"> Otherwise, all the case structures for the verb patterns are constructed using the word dictionary and a table of case frames.</Paragraph>
      <Paragraph position="7"> When a main term t6 with a category CL comes to be included in a partial case structure S} (~ =1,2, ...,m) as shown in Fig.2, syntactic and categorical consistency is checked.</Paragraph>
      <Paragraph position="8"> If the consistency holds, the term with the category C~L is added to the partial case structure S~ with a designated case label K~L.</Paragraph>
      <Paragraph position="9"> And if the case KiL is designated for dissolving multi-Japanese equivalents of the main verb in the case structure S~ , appropriate Japanese equivalents are extracted by subcategory matching. null Otherwise, if the category C~&amp; of the main term t i does not match one of any partial case structure S~ ( ~=1,2, ..-,n), the category C~L is deleted. Similarly, if a partial case structure requires a main term with a designated semantic category and a designated syntactic category or a position and cannot find the term, the partial case structure is deleted.</Paragraph>
      <Paragraph position="11"> Fig. 2 Construction of case structures In such a manner, the number of partial case structures decreases or increases as a term is added to partial case structures and, in many cases, it reduces to unity at the end of parsing.</Paragraph>
    </Section>
  </Section>
class="xml-element"></Paper>