{ "paper_id": "1991", "header": { "generated_with": "S2ORC 1.0.0", "date_generated": "2023-01-19T07:35:14.731403Z" }, "title": "", "authors": [], "year": "", "venue": null, "identifiers": {}, "abstract": "", "pdf_parse": { "paper_id": "1991", "_pdf_hash": "", "abstract": [], "body_text": [ { "text": "The avalanche-like flow of documents in natural Languages (NL) calls for a reliable cybernetic. means to conduct its intel lectual processing and formalized catalogization and classifica tion . The most effective instrument helpina to achieve these tasks is Linguistic Automaton (LA) . LA is an all-round complex of hard-, soft-, lingua-, and partly tutorware . \u2022 During recent years , the linguistic research activity at Lenin grad Speech Statistics Group (SpStGr) on natural lanauaae processinl was . concentrated on the pursuit of two objectives : first , the lexico-semantical , morpholoaical and pragmatical problems of automatical dictionary (AD) and second , the construction of parsing programs . At the same time , it had long been asserted that semantic and pragmatic information contained in AD and in LDB must be used to resolve many of the lexical and arammatical ambiguiti\ufffds. that occur in the-text . The adequate resolution of ambiguities is often critical to the -MT process , since often amb tauities which occur . in eource lanauaae cannot be maintained in target language .", "cite_spans": [ { "start": 648, "end": 652, "text": "(AD)", "ref_id": null } ], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "The creation of such a complex needs , on one hand, extensi.ve theoretical investigations\u2022 in the field of systemic linauis tics and considerat ion of po ssible practical contributions \u2022in such diverse natural language processing (NLP ) areas as machine trans la tion information retrieval , indexing , automatic abetracting etc .On the other hand , all these systems need special parsina algorithms and special structure of automatic dictionary (AD) .", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "The conjugation of AD structure and parsing hierarchy is the focus of this paper. This conjugation is hindered by a series of antinomies, the principal of which are two paradoxes :", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "1.The linearization paradox consists of non-additivity of text undestanding while human text processina . The process of text undestandina is surnultaneous with text reception . When modelling this process on computer, mental sumultaneous-associative proceeses are success! v\u2022ely linearized during parsing .", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "2. The static and dynamic paradox consists of the necessity to model the dynamically and constantly enriching procese of text generation and reception durina the human intellectual activity with the help of previou\ufffdly fixed procedures on the basis of a static model of averaged professional competence , stated in LDB .", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "As a matter of fact , the creation of NLP system ie a process of gradual -overcoming these paradoxes . The success of such a-process is determined by : -the correctness of the elaborated models of professional competence ; -the database organization model and the professional competence model level ; -the level of the model of language competence , and correspondingly, -the leve l of linguistic alaorithms and program elaboration; -the optimum of parsing realization ; -the level of computer development .", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "Thus , when designing NLP system it is necessary to conjugate th\ufffd three previously established models in a united technological structure which allows to minimize the influence of the described paradoxes on the NLP rezult .", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "The basis of this conjugation is both the organisation of data processing (parsing ) and the organisat ion in LDB . It means that the source structures of each l\ufffdvel -are transformed into output structures which may be modified on a higher level in accordance whith the structur\ufffdl .features of thie higher level .", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "Thus the translation process is simul ated in the sy!$tem in ques t:ton as a composition of lexical . and semantic-syntactic translation prooess. During the lexical translation process the , identifiE;:ation of text and dictionary units and-the extraction of dictionary informa t . i'on from the lexicon blocks are carried out .. Durina eemantic-\u2022 s:vntact1o process the interlang. uage structure tran\ufffdfer which \u2022 uses the whole informat:l.on received on. the lexical t\ufffdan\ufffdlation phase and\u2022 J. 0-in$ qp grammar and semantic LDB blocks is carri\ufffdd out .. This tranafer procese ls sj,m\\;llated a!I an aggregate of verticaly conjuaated sub\ufffdyste\u2022\u2022 \u2022 the hierarchy of the components which are extraoted from the text .", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "The Linguist \u2022\ufffd aim. in this conception of translation process is to define\u2022 all the levels of tran$lation\u2022 and analysis., to formulate the set of charac-teristics which are necessary :for the source structure modification into the . taraet structure . of the definite level and to definite the specification of .the next: .. hiaher levels ..", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "Proceeding from the stated idea of the NLP system design l\ufffdt\u2022s analyse the structure of AD and the reciprocal correlation of arammar a. nd dictionary on each of the determined levels in the analysis and translat ion of the . Predicate of the sentence .", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "During the verb entry elaboration it is neoessary to choose t:he most important , key structural elements (which determines the dietionary volume) , and to state a set of rules for the singled out linguistic elements functioning (which determines the arammar volume and the principles of parsing).", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "For a multilanguage ATP system the choice of AD item is determined both by word-and formbuilding principles different in specific languagee as well as by the representation features of \u2022semantic text ltems . Besides that the choice of a basic dictionary item is determined b.}r the tasks of\u2022 NLP system and the LDB universality level.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "In the Soviet NLP systems the Russian lanauage is used as a metalan guage for souroe text definition as we ll ae the taraet language . The unity of the target language enables to unify its definition for all Nl,P systems from foreian languaaes into Russian and to unify the pt\u2022o\ufffdedures of morpholoaical synt.h\ufffdsis of a Rueeian wordf orms .", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "When we design MT system for translation from the Russian the procedures of the morpholoaical analysis are unified as w&ll. In any cas\ufffd machine morphology definition of the Russian language constitutes a separate module and is used in all versions of the system. SILOD-MULTIS AD inc ludes \u2022 source word dictionaries, which are . organized as dictionaries of word usages and d!;ctionaries of stems , source phrase dictionariee ,taraet stems def initions . and machine morphology f o\u2022 r different lanauages .", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "Any AD that characterizes a specific. language inc ludes a unive, rsal structure set of dictionary items and machine morpholoa . All the source language ADs have the same function and a united scheme oraanisation .", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "Thi\ufffd scheme allows to unify such procedures of the source languaae text processing as a selection of mini . mum text unite , th\ufffd morphological analysi\ufffd , the identification of the text with AD items\u2022, the organizat ion of the dictionary information file . Any lexical unit (LU) in AD acquires a desc\ufffdiption on the morphological , syntactic , semantic and functional levels as an appropriate characteristic set .", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "The basic version of the !SY!ltem includes dictionary items (DI), which consist of the following charact-eristics:", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "-the head LU as it is: a stem , a word or a phrase;", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "-the lexical and syntactic code (LSC) , which depends on the typological features of the source language , its grammar and parsing algorithms wpich are realized in the system in question ; -the tr\u2022 an slation , which is stored as references to the corresponding target language items (etems and lexical and grammatical characteristics ). For analytical languages the most expedient is the introduction of separate word. forms , as it allows to increase the speed of the system while the growth of the dictionary volume is negligible . For synthetical languages machine stems are the head LU in the DI and the input AD is filled up with machine morphology .", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "In order to reduce the memory volume for AD location we resort to the artificial morphology transformation , i.e. to the insertion of the agglutinative morphology . The essence of the latter consists in the process of the selection in any word usaKe a machine stem and an affix \"sticking\"to it .", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "The concept of the inserting of machine affix allows to elaborate the Russian machine grammar , formed as a set of paradigms -machine affix chains . Each typical paradigm correlates with the grammatical characteristics of stems and the word formation mode . The link between a machine stem and a paradigm is realized with the help of a special code , which characterizes all the word forms which can be generated from the stem in question .", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "The use of this machine morphology allows to realize the wordform generation proceduress in accordance with the lexical and grammatical characteristics which are formed in the course of MT , and to make. this procedure a universal one for any language pair .", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "Accordingly, the elaborated Russian stem dictionary perniits to ident ify automatically the text words with dictionary items and to ascribe their morphological characteristics accurately to case homonyms . The result of morphological analysis, which is received with the help of LDB and special lexical and morphological analysis algorithms , is a \ufffdource for parsing and transferring alaorithms for Ru ssian-English MT.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "A two-layer system of lexical and semantic coding is reali zed in the LOB of SILOD-MULTIS system . The upper level of this coding is constituted by 30-element LSC which is formed in DI immediately. LSC formation. is created in accordance with the -coding tables elaborated for every source system languages . This information can be formed on-line .", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "The levels discussed above specify the lexical and grammatical descript ion of LU in LDB . The syntactic definition covers-the functional LU ch\ufffdracteristics wh ich determine their potential capac ities to accompl ish a specific role in syntactical sentence $tructure . The semantic definition which constitutes in a distinct , internal level is concerned with the transfer from the linguistic phenomena proper to the extralinauistic ones. The formation . of this definition is based on the structural investigation of the domain, that is to be manifested .", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "Let's consider the structure of information on the example of verb entry of French-Russian MT system , which is the base for parsing system .", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "On the lexical level of the analysis the predicat. e equal to the morphological verb-form is development . In\u2022 the French-Russian MT system the verb is presented in two ways : as word-forms for the irregular and suppletive verbs ( avoir, etre , aller, vouloir ) and as machine-stems with their standard paradigm .", "cite_spans": [ { "start": 229, "end": 261, "text": "( avoir, etre , aller, vouloir )", "ref_id": null } ], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "Each source standard paradigm inc ludes information sufficient to establish a link with \u2022 a def inite stem and a corresponding word entry (item) . The analysis procedure is performed according to the morphological tree .", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "On the functional unit level a verb and nominal seaments are tdentified . The structures of this level include verb segments \ufffdqual to the complex verb, -tense of the pronominal verbs and of the vArbs in active and passive forms . The procedure is performed on the information contained. in various positione of theverb entry-:", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "--the information of the _ verbs belonging to the auxiliary class are contained in the LSC . This information is necessary for the discrimination of the _ complex verb tenses . Position Six of the LSC of the verbs .. aller\" , \"venir\" contains the information necessary for \"Immediate\"_ tenses identification ; -the passive form identification footholds on Position Eiaht (transitivity notes ), but for its translation the corresponding rules 9f Position Eleven are to be used. This position contains the information of the possibility of the shortened passive participle form usage (\"est ouvert\" -opened ), the pronounal form usage (\"est prepare .. --is prepaired ), the active form usaae (\"est 9uivi \" -follows ). The pronounal form is tran9lated according to the information of Position Twelve of the verb entry . The compound nominal \u2022 p\ufffderlicate identification and tran9lation is performed on the basis ol : Position Fourteen .", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "\u2022 As to\u2022 the designing of the grammar rules which\u2022 direct the analysis and translation of-. impereonal construction it is pre\ufffdcribed by_ the information of Position Fift\ufffden .", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": ". The inner verb class relations are of fixed character . Thi\ufffd makes it po!Ssible to present a \u2022 verb segment as a frame i.ncludina all verb-connected elements \u2022 (the obj ective pronouns , the. negative and limiting partycles ) and verb elements ( the auxiliary verbs and the participles of a conjugated \u2022 . verb ) . During the analysie on the . functional segment level the procedure of homonymy elimination is \u2022realized.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "'fhe re!lult ' of the procedure on thie \u2022 level is a chain of source and target functional segment . Together with this the taraet fun,jtional eepent ( a verb group ) gets a certain set \u2022 of indications necessary for -the next level -the sentence level analysis.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "'rhe peculiarity of verb elemente analysis i9 their immediate functioning on the sentence level, as to the nominal groups , they hav\u00ab3 an add! tional staae -the stage of functioning components formation .\u2022 This is explained by the divers ity in the interrelations of the nominal group elements.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "Thus up to the beaining of the eentence level analy!Sis the structure of the v\ufffdrb functional sepent is known , the waye of the given verb structure presentation are defined ; the verb elements homonymy is eliminated. The designed output structure gets the total set of indi cations necessary for it9 analysis on th\ufffd sentence level .", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "This set is compiled of the activ\ufffd form\u2022 verb entry inf-ormation: -the indication of the \ufffd obli gatory direct obj ect according to This set is . also compiled of the information a9cribed to the pronominal verbs (the -type of government ); according . to Position Thi rteen, .and to the passive form verbs according to Poeition Ten ; and the\u2022 indications f.ormed in the translation process on the preceding levels of the analysis (tense , number , per, 9on and others ) of the compound verb conetructions in all mentioned forms . By the sentence leve l analysis stage a number of \"refusals\" , g0t on the previous levels, are piled because of various cause\ufffd (ambiguity of the structure in a bil ingual situation , unel iminated homonymy , impossibility of the analysis on the preceding stages of n number of constructions (infinitive, passive , impersonal, pronominal ) requiring the subject-ob ject 'transfoormations for a r\ufffdorrect translation ). Thus it is pos!Sible to pass over to the choice of the translational structure of the whole sentence only 8ft9r the functional of the nominal and verb groups as sentence m\ufffdmhere is defined . While choosing the translational equivalent on the sentence l\ufffdvel some difficulties arise in the case of the input and output ::;tructures inadequacy .", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "Then it is possible to resort to the subject\ufffdobject t.ransformations . The subject-object transformations may be realized r\ufffd lther with the help of the sentence members rearrangement or by I.he case forms of the target structure change or by the convers ives s\ufffdarch .", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "The conversives search practically leads to the inorease of I.he number of the verb translational equivalents . Hore pr_ oductive is the way of subject-object transformations , connected not with the sentence members rearrangement but with the case relations r\ufffd hange in the output structure . The resul te of the sentence level 0. l aboration is the obtaining of the output sentence structure .", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "On the phrase level the translation of the whole\u2022 complex s\ufffdntence is performed . Here the subordinate clause translation is nor\ufffdected. In particular the testing of the cor\ufffdecit choice of the (\ufffd :on.iunctione and relative pronouns, introducing the subordinate clauses . Thus for a correct choice of the translational. equivalent ,yf a. . n homonymous form \"que\" (what , so that , which) it is necessary Lo resort to Position Eight of the word entry inf ormation . The i.nformation contained in it gives an opportunity to (' hoose the core et form (indicative or subjunctive ) for the \ufffdrn bordinate clause verb translation . The same process takes place \u2022;1hen translatin,: the subordinate clause with \"clout\" . The correct 1\ufffd! hoice of the translational equivalentfor the whole subordinate clause is realized only with the orientation to te indication of Position Nine of the main clause verb .", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "thus the chosen point of view on the MT system elaboration makes it possible to realize the whole volume of the research goals. In this circumstance that is an indipeneable facil ity for the designing of the interaction of grammar and dictionary on \ufffdach of the system levels.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "Hence this conception creates the necessary faciolities for the deve lopment of the eystems forecasting the analysis of newly \ufffdrising situations on the basis of the once elaborated situations .", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null } ], "back_matter": [], "bib_entries": { "BIBREF0": { "ref_id": "b0", "title": "Machine Translation : Past , Present , Future . Chichester : Ellis Horwood", "authors": [ { "first": "W", "middle": [ "J" ], "last": "Hutchins", "suffix": "" } ], "year": 1986, "venue": "", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "W.J.Hutchins . Machine Translation : Past , Present , Future . Chichester : Ellis Horwood , 1986; 382p .", "links": null }, "BIBREF1": { "ref_id": "b1", "title": "Organization of the Russian-English MT-al8orithm", "authors": [ { "first": "", "middle": [], "last": "Ju", "suffix": "" }, { "first": "R", "middle": [], "last": "Kondratieva", "suffix": "" }, { "first": "S", "middle": [], "last": "Piotrowski", "suffix": "" }, { "first": "; N", "middle": [], "last": "Sokolova", "suffix": "" }, { "first": "; H", "middle": [], "last": "Maruyama", "suffix": "" }, { "first": "S", "middle": [], "last": "Horohashi", "suffix": "" }, { "first": "E", "middle": [], "last": "Umeda", "suffix": "" }, { "first": "", "middle": [], "last": "Sumita", "suffix": "" } ], "year": 1987, "venue": "Machine Traslation : Theoret ical and Methodological Issues", "volume": "32", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Ju . Kondratieva , R.Piotrowski , S.Sokolova . Organization of the Russian-English MT-al8orithm . SCCAC Newsletter, No 4, Bowling Green , 1988, p.2 1. N.Maruyama; H.Horohashi , S.Umeda , E.Sumita . A Japanese sentence ana.lyzer. -In: IBM Journal of Re search and Development , Vol . 32 , No 2, March 1988 , pp . 238-250 . - S.Nirenburg(ed) . Machine Traslation : Theoret ical and Methodological Issues . Cambridge : Cambridge University Press , 1987 . -350p. SILOD . A Russian-English Translation Support . New-Delhi : Elorg Computronics India, 1988 . -16p .", "links": null } }, "ref_entries": { "FIGREF0": { "type_str": "figure", "text": "LDB organization: \u2022 must answer to the next requirement), : , 1) \u2022the data , which are '\u2022inserted in \u2022the LDB , and t\ufffde data _ de. scriptions must\u2022 be : st\u2022ructurized in accordance with \u2022 th\ufffd procedures , wh ich are realized in a epecif.ic NLP system ; um l.DB -oraanizati-' on requi\ufffdes a modular design -.,rhich --_ constets \u2022ill' realiz\ufffdt:'ion' of 'LOB as \u2022 a set of \u2022 nonrilidly-linked modules . .. Th-i s\u2022 modularity-allow-s to arranlie \ufffda LDB as . modules\u2022 are ready and elimfna:tes the , data duplicati ' on . Beside!i. it allowi s \u2022 the'\u2022 etE ? of terminology and syntax _ of a special \u2022ciomaln . The organi !sation of LNP !lYf!te\u2022 m implie!l the systemic . princfple\u2022 of source arid target languaaes syntaxis; -the interface between LOB and software . \u2022 In accordance with this we can establish the main principles of HT system deshrn . They are a!l follows : 1.The principle of modular and hierarchical organisation .' \u2022 -2. The principle of '!leparation of baste and problem-or.iented modules of linaua-and software . 3.The principle of -the transfer as the translation process basis . The main,-feature of our LA de!lip -approach is a tend to separate the aroup!l of interconnected processes in a compl icated ATP process as a whole. This separation is to be done eo that -their interaction both g\u2022 ive certain system stability for different input data and allow to preeserve-open modular etructure . _ At -the same time these principal points in NLP 9ystem development inevitably lead to dimention crisis . That' s why in the eleborated system the hierarchy of translation levels is clearly defined . The developme\u2022 nt of \u2022the hierarchy etructure _ of the sy!!t\u2022 em is realized in a descending line,\"from top to bottom\" .Thi!I point -of view implies the :' acctirdance \"' with this\u2022 \u2022 the procedure: of translat!\u00b0qn is_ -devtd\ufffd\ufffd __ _ into subprocesses-flevel:s) each having its own \u2022f\\i ' nctlonal' . value . \u2022 The . i :re eults of developine. nt \u2022 of each level form the \u2022 basis for \u2022 pro\u2022 cess: h1g on a higher level . Thus a phrase level , asentence level , a\u2022-fun\ufffdtional component level , \u2022 a functional unit' level, a lexical unit ley\ufffd( -:\ufffdre \u2022 separated . Each \u2022 lever \u2022 1s connected with the tran. elation proce!i\ufffdL Translation is regarded here as a multi-level process , each of ' i tfr procedures translates a component of the speci' al level:", "uris": null, "num": null }, "FIGREF1": { "type_str": "figure", "text": "of the. pqssible information distribution according \u2022 to Position Six ; -the indication of the \ufffdossible object or adverbi\ufffdl modifier according to Position Nine .", "uris": null, "num": null } } } }