| { | |
| "paper_id": "1988", | |
| "header": { | |
| "generated_with": "S2ORC 1.0.0", | |
| "date_generated": "2023-01-19T07:44:11.797629Z" | |
| }, | |
| "title": "A Principle-based Korean/Japanese Machine Translation System: NARA", | |
| "authors": [ | |
| { | |
| "first": "Hee", | |
| "middle": [ | |
| "Sung" | |
| ], | |
| "last": "Chung", | |
| "suffix": "", | |
| "affiliation": { | |
| "laboratory": "", | |
| "institution": "Chung-Nam", | |
| "location": { | |
| "addrLine": "Dae-Dog Danji", | |
| "postBox": "P.O.Box 8", | |
| "country": "Republic of KOREA" | |
| } | |
| }, | |
| "email": "" | |
| } | |
| ], | |
| "year": "", | |
| "venue": null, | |
| "identifiers": {}, | |
| "abstract": "This paper presents methodological and theoretical principles for constructing a machine translation system between Korean and Japanese. We focus our discussion on the real time computing problem of the machine translation system. This problem is characterized in the time and space complexity during the machine translation. The NARA system has the real time computing algorithm which is based on a mathematical model integrating the linguistic competence and the linguistic performance of both languages, with consequence that the system NARA has also the functional characteristic: the two-way translation mechanism.", | |
| "pdf_parse": { | |
| "paper_id": "1988", | |
| "_pdf_hash": "", | |
| "abstract": [ | |
| { | |
| "text": "This paper presents methodological and theoretical principles for constructing a machine translation system between Korean and Japanese. We focus our discussion on the real time computing problem of the machine translation system. This problem is characterized in the time and space complexity during the machine translation. The NARA system has the real time computing algorithm which is based on a mathematical model integrating the linguistic competence and the linguistic performance of both languages, with consequence that the system NARA has also the functional characteristic: the two-way translation mechanism.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Abstract", | |
| "sec_num": null | |
| } | |
| ], | |
| "body_text": [ | |
| { | |
| "text": "We are developing a two-way (bidirectional) simultaneous Korean/Japanese machine translation: NARA [7] . The NARA system is designed by a specific computing model, which is a mathematical model based on the methodological and theoretical principles involving the formalization of the two-way simultaneity. The most significant characteristics of using the formal description for the NARA system is that the descriptive contents of representative algorithm do not depend upon the conventional approaches to machine translation. They have only methodological and theoretical arguments such as transfer method and pivot method, and adopt linguistic theory for language model with an ad hoc manner.", | |
| "cite_spans": [ | |
| { | |
| "start": 99, | |
| "end": 102, | |
| "text": "[7]", | |
| "ref_id": "BIBREF6" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Introduction", | |
| "sec_num": "1." | |
| }, | |
| { | |
| "text": "In other words, current approaches to machine translation are usually focusing on the engineering feasibility; therefore they explain only what kinds of data structure of the language they employ, how they are analyzed and how they are translated. They do not give the details of their capability and limitation for the methodology of the machine translation system. The aim to develop the machine translation system is to translate an enormous amount of information written in a foreign language. In this purpose, it is said that the translation is changing from an art to a technique. If we give a clear-cut answer to the argument that the translation is the technique, it is natural that the translation merely means the physical transference of the contents of the language. To realize the idea for mechanizing the translation process, we need to formalize the translation mechanism. In this paper, we propose a methodology needed for the improving the quality and quantity of a machine translation system.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Introduction", | |
| "sec_num": "1." | |
| }, | |
| { | |
| "text": "In the NARA system we take a methodological principle into consideration on the general and specific aspects for the machine translation system. The former is the hypothesis for constructing the machine translation system. The latter is the computational model that applies the general principles to the NARA system. The computational model for the NARA system is constructed by the principles of computing theory that produce a vital link between what and how. What means the linguistic knowledge for constructing machine translation system and How means the procedure that maps the collection of inputs to the desired outputs.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "The general principles of the NARA system", | |
| "sec_num": "2." | |
| }, | |
| { | |
| "text": "Our approach is intuitively motivated by Chomsky's hypothesis [4] : homogeneous communication by the same linguistic performance is possible among those who have the same linguistic competence. The linguistic performance means the real time processing of the language, and the linguistic competence means the knowledge of a language. The performance theory can not be developed without the competence theory.", | |
| "cite_spans": [ | |
| { | |
| "start": 62, | |
| "end": 65, | |
| "text": "[4]", | |
| "ref_id": "BIBREF3" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "The general principles of the NARA system", | |
| "sec_num": "2." | |
| }, | |
| { | |
| "text": "This hypothesis suggests that mutual communication is possible among different human language systems. Thus we may represent the above concept as follows: the description of mutual communication environment = the description of linguistic competence + the description of linguistic performance.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "The general principles of the NARA system", | |
| "sec_num": "2." | |
| }, | |
| { | |
| "text": "And we may analogously represent the concept of two way simultaneous translation as follows:", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "The general principles of the NARA system", | |
| "sec_num": "2." | |
| }, | |
| { | |
| "text": "the description of two way simultaneous translation = the description of knowledge of both languages + the description of performance knowledge of the both languages. This schema can be expanded a step further to be: a two way simultaneous translation algorithm = the model of the corresponding data structure of the source language and the target language + the model of real time processing.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "The general principles of the NARA system", | |
| "sec_num": "2." | |
| }, | |
| { | |
| "text": "A key point of contact between the theory of grammar and the translation control is the natural link between the theory of knowledge representation and the theory of knowledge processing for the machine translation system.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "The general principles of the NARA system", | |
| "sec_num": "2." | |
| }, | |
| { | |
| "text": "We define the knowledge representation and the knowledge processing for machine translation system as a competence model and as a performance model, respectively. The competence model consists of the various kind of linguistic knowledge: morphology, syntax and semantics for the NARA system, and the performance model consists of several subareas. The first is concerned with which knowledge representations are constructed during simultaneous translation; the second is concerned with how the representations are utilized during translation; the third is concerned with the measure of computational complexity during translation. We presume that these three components constitute a complete computational model for the machine translation mechanism: a knowledge representation, an algorithm and a complexity.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "The general principles of the NARA system", | |
| "sec_num": "2." | |
| }, | |
| { | |
| "text": "We summarize the following items as the subjects of the general principle for our computational approach of the NARA system.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "The general principles of the NARA system", | |
| "sec_num": "2." | |
| }, | |
| { | |
| "text": "(1) The theory of common grammar: we are requiring a common grammar to be suitable for the description of both languages. The common grammar is similar to the significance such that modern linguistic theory interprets the theory of universal grammar (UG) as part of a theory of language acquisition [3] . whereby we adopted a unification-based grammar formalism: K-J(J-K) grammar as a common grammar based on the correspondence existing between both languages.", | |
| "cite_spans": [ | |
| { | |
| "start": 299, | |
| "end": 302, | |
| "text": "[3]", | |
| "ref_id": "BIBREF2" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "The general principles of the NARA system", | |
| "sec_num": "2." | |
| }, | |
| { | |
| "text": "(2) The notion of direct realization of translation:", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "The general principles of the NARA system", | |
| "sec_num": "2." | |
| }, | |
| { | |
| "text": "as mentioned above, in order to guarantee two-way simultaneous translation, we identify the rules of the grammar with the manipulative units of translation in a one-to-one fashion. This notion is based on the grammatical covering, type transparency, grammar modification and invariants of formal language [10] .", | |
| "cite_spans": [ | |
| { | |
| "start": 305, | |
| "end": 309, | |
| "text": "[10]", | |
| "ref_id": "BIBREF9" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "The general principles of the NARA system", | |
| "sec_num": "2." | |
| }, | |
| { | |
| "text": "(3) The notion of complexity measure: the complexity of the algorithm, which is the direct association between the cost time and the sequential operation during translation, should be measured.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "The general principles of the NARA system", | |
| "sec_num": "2." | |
| }, | |
| { | |
| "text": "(4) The notion of translation results: we compare our translation results to Thorn's hypothesis: a principle of isomorphism [12] .", | |
| "cite_spans": [ | |
| { | |
| "start": 124, | |
| "end": 128, | |
| "text": "[12]", | |
| "ref_id": "BIBREF11" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "The general principles of the NARA system", | |
| "sec_num": "2." | |
| }, | |
| { | |
| "text": "The NARA system adopts several specific and theoretical principles and they are described in the following.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "The specific principles of the NARA system", | |
| "sec_num": "3." | |
| }, | |
| { | |
| "text": "(1) Equivalence of grammar: only if two grammars generate the same sets of surface sentences, they are weakly equivalent. In addition, if the two grammars generate the same language by means of the same two structure (here, by a one-to-one correspondence of rule steps), the two grammars are strongly equivalent [10] . Paraphrasing it, grammar G 1 and G 2 are weakly equivalent if the string language generated by G 1 , L(G 1 ), is identical to that of G 2 , L(G 2 ). If G 1 and G 2 are strongly equivalent, G 1 and G 2 can assign the same structural description for each word in L(G l ) and L(G 2 ). We apply this notion to the correspondence between Korean and Japanese.", | |
| "cite_spans": [ | |
| { | |
| "start": 312, | |
| "end": 316, | |
| "text": "[10]", | |
| "ref_id": "BIBREF9" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "The specific principles of the NARA system", | |
| "sec_num": "3." | |
| }, | |
| { | |
| "text": "(2) Grammar covering and grammar modification:", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "The specific principles of the NARA system", | |
| "sec_num": "3." | |
| }, | |
| { | |
| "text": "intuitively, a grammar is said to cover another if the first grammar can be used to easily recover all the parse structure that the second grammar assigns to an input sentence. In other words, grammar covering means that the first grammar can be used instead of the second grammar to parse a sentence of the language generated by the second grammar. This grammatical covering relation is easy to understand from the mere fact that we use the first language to study the second language. More importantly, one of the two grammars can serve as the true competence grammar for a language because it generates the proper structural description. The reason for using this principle is that the covering grammar may be more suitable for the efficiency of the processing in terms of time and space, and if a grammar covers another, the semantic rule for translation between both languages can be used to pair exactly the same input string and its meaning.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "The specific principles of the NARA system", | |
| "sec_num": "3." | |
| }, | |
| { | |
| "text": "(3) Type transparency: in our view, the type transparency is the relationship between a covering grammar and the operation units of translation. From the usual linguistic claim that a more compact grammar is more easily processed, we impose the condition that the logical organization of rules and the structure incorporated in a grammar may be mirrored exactly in the mechanism of translation.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "The specific principles of the NARA system", | |
| "sec_num": "3." | |
| }, | |
| { | |
| "text": "According to our theoretical and specific principles, we can represent the structural description of translation processing, and then apply a simple mapping to the translation mechanism. This mapping is from a parse tree to a parse tree.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "The specific principles of the NARA system", | |
| "sec_num": "3." | |
| }, | |
| { | |
| "text": "In this section, we focus our attention on the concrete language knowledges such that what kinds of linguistic description are used. In order to investigate the correspondence between both languages, we partition a grammar into components: segmented word, word order, morphology, syntax and semantics. The hierarchical separation of a grammar constitutes an important step in the modularization of a translation subsystem.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "The competence model of the NARA system", | |
| "sec_num": "4." | |
| }, | |
| { | |
| "text": "The study of the structure of words, occupies an important place within the competence model, sandwiched as it is between phonology and syntax. Morphemes may also be partitioned into lexical and grammatical classes. In both languages, lexical morphemes are generally free, while many of the grammatical morphemes are bound.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Morphology", | |
| "sec_num": "4.1." | |
| }, | |
| { | |
| "text": "In a given Korean-Japanese/Japanese-Korean dictionary, let D k be the set of morphological words of Korean and D j be the set of morphological words of Japanese. Consider the cartesian product, D k x D j , of the two sets. A mapping between the sets may be defined as follows.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Morphology", | |
| "sec_num": "4.1." | |
| }, | |
| { | |
| "text": "I(D K ) = D J .", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Morphology", | |
| "sec_num": "4.1." | |
| }, | |
| { | |
| "text": "implying that the image of Dk is Dj; taking the inverse mapping", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Morphology", | |
| "sec_num": "4.1." | |
| }, | |
| { | |
| "text": "I -1 (D j ) = D k", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Morphology", | |
| "sec_num": "4.1." | |
| }, | |
| { | |
| "text": "By generalizing the relation and the mapping between the two sets, we may consider the word set of the source language to be the domain, and the target language word set to be the range. Assuming the same cardinality for both the domain and the range, D k and D j may be partitioned as shown below. Here we suppose", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Morphology", | |
| "sec_num": "4.1." | |
| }, | |
| { | |
| "text": "(k 1 , k 2 , k 3 , .... k n ) \u2208 D k (j 1 ,j 2 ,j 3 ,....... j n ) \u2208 D j", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Morphology", | |
| "sec_num": "4.1." | |
| }, | |
| { | |
| "text": "(a) one-to-one (b) one-to-many (c) many-to-many Obviously, one-to-one correspondence is isomorphic. Thus our attention will be focused on one-to-many and many-to-many relations. The translation of these relations depends on various factors: allomorphs, synonyms and homonyms of both languages. As an elementary strategy for the translation of those correspondence, we adopted a normalization pro-cedure which ensures the decomposition of one-to-many and many-to-many correspondence into one-to-one correspondence. As for the translation which is dependent on synonyms or homonyms, we specify the canonical form and the semantic feature, respectively. In reality, there are some linguistic representation (words) which exists in Korean but do not exist in Japanese(and the converse in also true); therefore, the need to make new words.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Morphology", | |
| "sec_num": "4.1." | |
| }, | |
| { | |
| "text": "Between Korean and Japanese, some common properties are observed, such as an agglutinative language structure and same word order (SOV) [5] . In this subsection, we examine the word order in a segmented word. There are some corresponding properties in word order of the segmented words between both languages as follows:", | |
| "cite_spans": [ | |
| { | |
| "start": 136, | |
| "end": 139, | |
| "text": "[5]", | |
| "ref_id": "BIBREF4" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Word order in the segmented words", | |
| "sec_num": "4.2." | |
| }, | |
| { | |
| "text": "[property 1] correspondency.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Word order in the segmented words", | |
| "sec_num": "4.2." | |
| }, | |
| { | |
| "text": "[property 2] inversion.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Word order in the segmented words", | |
| "sec_num": "4.2." | |
| }, | |
| { | |
| "text": "[property 3] abbreviation.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Word order in the segmented words", | |
| "sec_num": "4.2." | |
| }, | |
| { | |
| "text": "Among the properties, property 3 depends upon Korean pragmatic information.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Word order in the segmented words", | |
| "sec_num": "4.2." | |
| }, | |
| { | |
| "text": "Korean and Japanese have a remarkable characteristics; namely, the structure of segmented words. The segmented words are the important language structure as a utterance unit, and play an important role in the analysis of both languages.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Word order in the segmented words", | |
| "sec_num": "4.2." | |
| }, | |
| { | |
| "text": "The production form of the segmented words can be described in the forms of a regular grammar: In other words, for each symbol a in the vocabulary of some regular set R, let R b be a particular regular set. Suppose that we replace each word a 1 , a 2 , a 3 ,... a n in R by the set of words of form w 1 , w 2 , w 3 ,... w n , where w i is an arbitrary word in Rb. Then the result is always a regular set. More formally, a substitution f is a mapping which is from vocabulary A to subsets (language family) of vocabulary B. Thus the mapping f is extended to strings as follows:", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Word order in the segmented words", | |
| "sec_num": "4.2." | |
| }, | |
| { | |
| "text": "S --> uB S, B \u2208 N (nonterminal symbols) S --> u u \u2208 T (terminal", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Word order in the segmented words", | |
| "sec_num": "4.2." | |
| }, | |
| { | |
| "text": "1) f(\u03be) = \u03be , 2 ) f(xa) = f(x)f(a).", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Word order in the segmented words", | |
| "sec_num": "4.2." | |
| }, | |
| { | |
| "text": "The mapping f is extended to languages by defining", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Word order in the segmented words", | |
| "sec_num": "4.2." | |
| }, | |
| { | |
| "text": "f(L) = \u222a f(x).", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Word order in the segmented words", | |
| "sec_num": "4.2." | |
| }, | |
| { | |
| "text": "x\u2208L A type of substitution that is of special interest is a homomorphism. A homomorphism h is a substitution such that h(a) contains a single string for each a. We generally take h(a) to be the string itself, rather than the set containing that string. It is useful to define the inverse homomorphic image of a language L to be", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Word order in the segmented words", | |
| "sec_num": "4.2." | |
| }, | |
| { | |
| "text": "h -1 (L) = {x|h(x) is in L}", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Word order in the segmented words", | |
| "sec_num": "4.2." | |
| }, | |
| { | |
| "text": "We also use, for string w;", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Word order in the segmented words", | |
| "sec_num": "4.2." | |
| }, | |
| { | |
| "text": "h -1 (w) = {x|h(x) is in w}", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Word order in the segmented words", | |
| "sec_num": "4.2." | |
| }, | |
| { | |
| "text": "Consequently, the translation between Korean and Japanese is closed in the substitution among the constituent which are called the segmented words.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Word order in the segmented words", | |
| "sec_num": "4.2." | |
| }, | |
| { | |
| "text": "It is seen intuitively from the correspondence in the segmented words and word order, that Korean and Japanese have the similar language structure [6] . Let us compare the two parse trees of the actual example sentences. It is obvious that the parse trees correspond to each other in a one-to-one fashion, but the lexical categories do not coincide with each other. This means that both languages do not generate the same set of sentential forms: S(G) = {w \u2208 (NUT) + | S -*-> w}. Furthermore, there is no algorithm for deciding whether or not two given context-free grammars generate the same sentential forms [9] . This proposition reveals the reason why we adopt the covering grammar and the grammar modification principle.", | |
| "cite_spans": [ | |
| { | |
| "start": 147, | |
| "end": 150, | |
| "text": "[6]", | |
| "ref_id": "BIBREF5" | |
| }, | |
| { | |
| "start": 610, | |
| "end": 613, | |
| "text": "[9]", | |
| "ref_id": "BIBREF8" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Syntax", | |
| "sec_num": "4.3." | |
| }, | |
| { | |
| "text": "If a sentence is syntactically ambiguous, it has more than one canonical derivation and is semantically ambiguous if, for a given canonical derivation, it has more than one translation. Derivations are not related directly to a language but to a grammar that generates it. In the translation between Korean and Japanese, there exist several kinds of inherently ambiguous sentences which are generated only by ambiguous grammar of both languages.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Semantics", | |
| "sec_num": "4.4." | |
| }, | |
| { | |
| "text": "In the NARA system, the semantic knowledge is used to eliminate the ambiguity in the syntactic-based translation. But, its role is a minimum essential in the NARA system. Because a semantic theory of natural language, for example situation semantic theory, being underdeveloped, and is not necessary and sufficient condition for the Korean and Japanese translation system. However, for the word that involves the ambiguity in the translation processing, we specify the lexical semantic features and introduce the individual semantic features into the syntactic feature system. In consequence, the lexical semantic features of the constituents are kept in the phrase structure and are applied to the semantic-based translation. That is, the constraints for the semantic sensitive translation are described in the partial phrase structure, and play a role of adjusting semantic sensitive translation.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Semantics", | |
| "sec_num": "4.4." | |
| }, | |
| { | |
| "text": "In this section, we design a K-J (or J-K) grammar which eliminates syntactic or semantic ambiguity of both languages. This grammar corresponds to the communicative competence model for the translation system between Korean and Japanese. The grammar is motivated by grammar modification and covering grammar; the original grammar is not often suitable for a particular parsing technique but can be modified into an equivalent form which is suitable. Where N and N' are nonterminals, T k , T j , T k , and T j , are terminals, sem-k is semantic features, P and P' are production rules, and S is the start symbol. The J-K grammar is designed by the method analogous to that of the K-J grammar. In the unification-based grammar framework, the semantic features are accepted by a special phrase structure rule, a linking rule (unification), which causes the relevant information about the phrase to be passed down the tree as a feature on the syntactic nodes. Therefore, translation procedure is constructed by a succinct algorithm founded on the K-J(J-K) grammar. Before describing the performance model of the NARA system, we briefly describe the NARA system below:", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "K-J Grammar", | |
| "sec_num": "4.5." | |
| }, | |
| { | |
| "text": "As the preprocessing for the morphological level translation, segmented word analysis is carried out on each word given by the lexicon information.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Analysis -Morphological analysis", | |
| "sec_num": "5.2." | |
| }, | |
| { | |
| "text": "The structure of a sentence is analyzed on the phrase structure level. A tree structure which serves as an intermediate structure for translation is constructed.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "-Syntactic analysis", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "We formulate the internal interface for the translation. This interface corresponds to the transducer of translation. We can define the K-J(J-K) system as a 3-tuple grammar G=(w j , w k , k(or j) ), where w k and w j are Korean words and Japanese words, respectively, k(j): w j -> w k (w k -> w j ) is the homomorphism. The K-J(J-K) system ul G defines the following sequence preserving the word order:", | |
| "cite_spans": [], | |
| "ref_spans": [ | |
| { | |
| "start": 173, | |
| "end": 195, | |
| "text": "G=(w j , w k , k(or j)", | |
| "ref_id": null | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "K-J(J-K) system", | |
| "sec_num": "5.3." | |
| }, | |
| { | |
| "text": "w k l = k(w j l ), w k 1 w k 2 = k(w j 1 ) k(w j 2 ),....", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "K-J(J-K) system", | |
| "sec_num": "5.3." | |
| }, | |
| { | |
| "text": "It also defines the following language", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "K-J(J-K) system", | |
| "sec_num": "5.3." | |
| }, | |
| { | |
| "text": "L(G) = {k i (w j ) | i > 0 } .", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "K-J(J-K) system", | |
| "sec_num": "5.3." | |
| }, | |
| { | |
| "text": "As mentioned above, the K-J system constitutes a simple device for the translation. A language defined by the K-J(J-K) system corresponds to the target language. Inversely, the mapping j of w k into w j is such that the inverse homomorphism j(w j ) = {w j | k(w j ) = w k }, j = k -1 exists. Thus, we define the two-way simultaneous translation system NARA by:", | |
| "cite_spans": [ | |
| { | |
| "start": 281, | |
| "end": 283, | |
| "text": "-1", | |
| "ref_id": null | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "K-J(J-K) system", | |
| "sec_num": "5.3." | |
| }, | |
| { | |
| "text": "j(L k ) = k -1 (L k ) = {w j |k (w j ) \u2208 L k }.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "K-J(J-K) system", | |
| "sec_num": "5.3." | |
| }, | |
| { | |
| "text": "We can define the NARA system using the extended notion; the inverse homomorphism can be replaced by the direct operation of a finite substitution as follows. Consider a grammar (e.g. Korean) G j = (N j , T j , P j , S j , ) and let j be a finite substitution, defined on the vocabulary (N k \u222a T k )* such that j(a) is a finite (possibly empty) set of word for each word a. We denote j(N k ) = N j , j(T k ) = T j , P j \u2283 j(P k ), S j \u2283 j(S k ).", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "K-J(J-K) system", | |
| "sec_num": "5.3." | |
| }, | |
| { | |
| "text": "Then, the grammar (e.g. Japanese)", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "K-J(J-K) system", | |
| "sec_num": "5.3." | |
| }, | |
| { | |
| "text": "G j = (N j ,T j , P j , S j ) is the translation of G k . If I(G k ), I(G j ) are the sets of all translation of G k and G j , respectively, then I(G k ) = I(G j ), and I is an invariant for G k and G j .", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "K-J(J-K) system", | |
| "sec_num": "5.3." | |
| }, | |
| { | |
| "text": "From the morphology dependent on the target language is generated with the aid the phonological form file, the correct phonological form of the target language, which is subsequently output.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Synthesis", | |
| "sec_num": "5.4." | |
| }, | |
| { | |
| "text": "The dictionary consists of the K-J(J-K) grammar and the lexicon. A dictionary compiler is used to transform a visual dictionary into a system dictionary implemented in the form of a B-tree. The modularity of the grammar and the ease way of operation which to update the dictionary serve as major factor in the system.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Dictionary", | |
| "sec_num": "5.5." | |
| }, | |
| { | |
| "text": "In this section, we present how we can predict the time or memory space or sequential operation that will be needed to perform the computing model of the NARA system, and how the translation process can be specified clearly and unambiguously. The complexity of the algorithm in usually measured by the growth rate of its time or space requirements, as a function of the size of the input (or the length of input string) to which the algorithm is applied. We shall now define the time characteristics of translation process. There are some kinds of syntactical relations such that structural distance is naturally involved in the simultaneous translation. Consider the following example: Such ambiguity arises in translation due to one-to-many relation on morphological level, the sentence 2) is the well formed translation. The reason is co-occurrence relation; namely, a Japanese verb a-u (meet) co-occurs with a postposition ni (dative case), and a Korean verb mana-da co-occurs with a postposition lul (accusative case). If the postposition proceeds the verb, then simultaneous translation is impossible. In this case, delay time p > 1 for complete translation is required before two words bind, and one more operation is required to rescan the translated sentence. We refer to this case as quasi-real time translation. We formalize the time complexity of translation. An utterance string of the source language L is the sequence string S t (L). S t (L) = (k 1 , k 2 .... k t ) is a partial utterance string up to time t, and K-J(S t (t)) is a translation sequence string up to time t. Also T t (L) = (j 1 , j 2 ,...j t ) is a target language which is generated by the K-J system. The translation operates in real time so that delay time is 0. Therefore, K-J(S t (L)) = (T t (L)) where S t = T t . The translation operates in the quasi-real time so that delay time p > 1. Therefore, K-J(S t (L)) = (T t (L)) where T t -S t > 1. However, the nature of on-line translation is unchangeable.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Complexity of system NARA", | |
| "sec_num": "5.6." | |
| }, | |
| { | |
| "text": "We compare our translation results to Rene Thorn's hypothesis: the principle of isomorphism concerning linguistic universality. Let T 1 be a text of language L 1 , and T 2 be a text to be translated from T 1 into language L 2 . Suppose {Q 1 i } and {Q 2 j } are phrase elements of decompositions of T 1 and T 2 , respectively, then the following principle of isomorphism holds:", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Complexity of system NARA", | |
| "sec_num": "5.6." | |
| }, | |
| { | |
| "text": "[Principle of Isomorphism] A one to one correspondence exists between {Q 1 i } and {Q 2 j } which conserve each signification. Moreover, this correspondence nearly preserves the order of phrase elements; in other words, if the ith element Q 1 i of T 1 corresponds to the jth element Q 2 j of T 2 , then |j-i| < 4.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Complexity of system NARA", | |
| "sec_num": "5.6." | |
| }, | |
| { | |
| "text": "We consider that Thom's hypothesis provides the index for the measurement of the translation complexity between some two languages.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Complexity of system NARA", | |
| "sec_num": "5.6." | |
| }, | |
| { | |
| "text": "Our approach for constructing the NARA system included logical study and experimental study; the former was given by the mathematical formalization, the latter by the correspondence of two languages. In the view of computational linguistics, we separated the mechanism of two way simultaneous translation system into the levels of abstract theory, algorithm, and implementation to carve out the results at each level in more independent fashion. In order to do so, we specified four important levels of the description: the lowest level is the morphology, the second level is the segmented words, the third levels are the syntax and the semantic, and the top level controls the computing model of each level. Hence, we could determine the range of correspondence between internal representations of both grammars, and the basic architecture of the machinery actually instantiates the algorithm. Consequently, our model produces the extra power by the proposed theory with multiple levels of representation and systematic mapping between the corresponding levels of two languages, because translation efficiency requires both a functional and a mathematical argument. In the view of software engineering, going through each level of abstraction we expect to make an elegant program which satisfies the requirements of the machine translation system such as simplicity, reliability, adoptability and modularity. Nevertheless, the complete pragmatic translation remains quite obscure.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Concluding remarks", | |
| "sec_num": "6." | |
| } | |
| ], | |
| "back_matter": [ | |
| { | |
| "text": "The author is deeply grateful to Dr. Gil Rok Oh and Dr. Min Ho Kang for their encouragement.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Acknowledgement", | |
| "sec_num": null | |
| } | |
| ], | |
| "bib_entries": { | |
| "BIBREF1": { | |
| "ref_id": "b1", | |
| "title": "The Grammatical Basis of Linguistic Performance: Language Use and Acquisition", | |
| "authors": [ | |
| { | |
| "first": "R", | |
| "middle": [], | |
| "last": "Berwick", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "A", | |
| "middle": [], | |
| "last": "Weiberg", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1984, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Berwick, R., and Weiberg, A., The Grammatical Basis of Linguistic Performance: Language Use and Acquisition, M.I.T. press, Reading, 1984.", | |
| "links": null | |
| }, | |
| "BIBREF2": { | |
| "ref_id": "b2", | |
| "title": "The Acquisition of Syntactic Knowledge", | |
| "authors": [ | |
| { | |
| "first": "R", | |
| "middle": [], | |
| "last": "Berwick", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1985, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Berwick, R., The Acquisition of Syntactic Knowledge, M.I.T. press, Reading, 1985.", | |
| "links": null | |
| }, | |
| "BIBREF3": { | |
| "ref_id": "b3", | |
| "title": "Aspects of the Theory of Syntax", | |
| "authors": [ | |
| { | |
| "first": "N", | |
| "middle": [], | |
| "last": "Chomsky", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1963, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Chomsky, N., Aspects of the Theory of Syntax, M.I.T. Press, Reading, 1963.", | |
| "links": null | |
| }, | |
| "BIBREF4": { | |
| "ref_id": "b4", | |
| "title": "Current Korean: Elementary Sentence Patterns and Structures", | |
| "authors": [ | |
| { | |
| "first": "H", | |
| "middle": [], | |
| "last": "Chung", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1982, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Chung, H., Current Korean: Elementary Sentence Patterns and Structures, Komasholin, Reading, 1982 (in Japanese).", | |
| "links": null | |
| }, | |
| "BIBREF5": { | |
| "ref_id": "b5", | |
| "title": "Korean language Information Processing", | |
| "authors": [ | |
| { | |
| "first": "H", | |
| "middle": [], | |
| "last": "Chung", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1986, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Chung, H., Korean language Information Processing, Ph.D. dissertation, Tokyo Univ., 1986.", | |
| "links": null | |
| }, | |
| "BIBREF6": { | |
| "ref_id": "b6", | |
| "title": "A Two-way Simultaneous Interpretation System between Korean and Japanese: NARA, Proceeding of COLING'86", | |
| "authors": [ | |
| { | |
| "first": "", | |
| "middle": [], | |
| "last": "Chung", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "T", | |
| "middle": [], | |
| "last": "Kunii", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1986, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Chung, H and Kunii, T., A Two-way Simultaneous Interpretation System between Kore- an and Japanese: NARA, Proceeding of COLING'86, 1986.", | |
| "links": null | |
| }, | |
| "BIBREF7": { | |
| "ref_id": "b7", | |
| "title": "On the decidability of homomorphism equivalence for language", | |
| "authors": [ | |
| { | |
| "first": "L", | |
| "middle": [], | |
| "last": "Culik", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "A", | |
| "middle": [], | |
| "last": "Salomaa", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1978, | |
| "venue": "Journal of Computer and System Science", | |
| "volume": "18", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Culik, L. and Salomaa, A., On the decidability of homomorphism equivalence for language, Journal of Computer and System Science, 18, 1978.", | |
| "links": null | |
| }, | |
| "BIBREF8": { | |
| "ref_id": "b8", | |
| "title": "Introduction to Formal Language Theory, Reading", | |
| "authors": [ | |
| { | |
| "first": "M", | |
| "middle": [], | |
| "last": "Harrison", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1978, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Harrison, M., Introduction to Formal Language Theory, Reading, Addison-Wesley, 1978.", | |
| "links": null | |
| }, | |
| "BIBREF9": { | |
| "ref_id": "b9", | |
| "title": "Context-free Grammar: Cover, Normal Forms and Parsing", | |
| "authors": [ | |
| { | |
| "first": "A", | |
| "middle": [], | |
| "last": "Niholt", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1980, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Niholt, A., Context-free Grammar: Cover, Normal Forms and Parsing, Springer, Read- ing, 1980.", | |
| "links": null | |
| }, | |
| "BIBREF10": { | |
| "ref_id": "b10", | |
| "title": "Jewels of Formal Language Theory", | |
| "authors": [ | |
| { | |
| "first": "A", | |
| "middle": [], | |
| "last": "Salomaa", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1981, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Salomaa, A., Jewels of Formal Language Theory, Computer Science Press, Reading, 1981.", | |
| "links": null | |
| }, | |
| "BIBREF11": { | |
| "ref_id": "b11", | |
| "title": "Essays on Topology and Related Topics", | |
| "authors": [ | |
| { | |
| "first": "R", | |
| "middle": [], | |
| "last": "Thom", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "", | |
| "middle": [], | |
| "last": "Topologie", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "", | |
| "middle": [], | |
| "last": "Linguistique", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1970, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Thom, R., Topologie et Linguistique, in Essays on Topology and Related Topics, Springer, 1970.", | |
| "links": null | |
| } | |
| }, | |
| "ref_entries": { | |
| "FIGREF0": { | |
| "text": "symbols) They are both right linear. Denoting the language defined by such a regular grammar by L = L(G) lead to the existence of a finite state automaton M such that L(G) = T(M) ={ w | M accepts w}. And, if L(G) = L(G'), there is a sequence equivalence such as S(G) = S(G').", | |
| "num": null, | |
| "uris": null, | |
| "type_str": "figure" | |
| }, | |
| "FIGREF1": { | |
| "text": "ALGORITHM: irregularity categories removal or adjustment and semantic features insertion Input: a 5 tuple phrase structure grammar G = (N, T k , T j , P, S) for the translation. Output: an equivalent 5 tuple phrase structure grammar G' = (N', T k[sem-k] , T j , P', S).Method: empirical and heuristic method.", | |
| "num": null, | |
| "uris": null, | |
| "type_str": "figure" | |
| }, | |
| "FIGREF2": { | |
| "text": "The performance model of the NARA system 5.1. System architecture of the NARA system", | |
| "num": null, | |
| "uris": null, | |
| "type_str": "figure" | |
| } | |
| } | |
| } | |
| } |