NO790023L

NO790023L - PROCEDURE FOR USING A DIGITAL COMPUTER SYSTEM FOR TRANSLATION BETWEEN NATURAL LANGUAGES

Info

Publication number: NO790023L
Application number: NO790023A
Authority: NO
Inventors: Peter Toma
Original assignee: Systran Inst
Priority date: 1971-08-31
Filing date: 1979-01-04
Publication date: 1980-07-07
Also published as: ES476944A1; IL56396A0; IT7919669A0; PT69059A; FI790036A7; DD143667A5; DK5679A; BR7900395A

Description

Foreliggende oppfinnelse vedrorer en fremgangsmåte for bruk av The present invention relates to a method for using

en digital computer for oversettelse mellom naturlige språk. a digital computer for translation between natural languages.

Det er gjort forsok på å bruke digitale computere for oversettelse fra et språk til et annet, dvs. fra et kildespråk til et målspråko Oversettelses-systemene omfatter et programmerbart computersystem sammen med et program for gjennomforing av oversettelsen. De benyttede fremgangsmåter var teoretiske. Den teoretiske språkrnetode for syntaktisk analyse har ikke vært aksep-tabel, fordi den går ut fra lingvistiske antagelser, i stedet for å vurdere den programmerbare digitale computers muligheter og gå los på oversettelsen fra computeren som utgangspunkt. Attempts have been made to use digital computers for translation from one language to another, i.e. from a source language to a target language. The translation systems comprise a programmable computer system together with a program for carrying out the translation. The methods used were theoretical. The theoretical language method for syntactic analysis has not been acceptable, because it starts from linguistic assumptions, instead of assessing the possibilities of the programmable digital computer and starting with the translation from the computer as a starting point.

Ideen orn maskinell oversettelse ble unnfanget av Warren Weaver The idea of machine translation was conceived by Warren Weaver

og A.D. Booth i 1946. Det er siden gjort en rekke forsok på å finne frem til et maskinelt oversettelsessystem og sette det i drift. Projektene ble rettet mot utvikling av lingvistiske teo-rier som omslutter hele det naturlige språk, for så å gå til computeren med det. Denne fremgangsmåte har uunngåelig sviktet, fordi menneskesinnet ikke kan omslutte et språk i sin helhet. and A.D. Booth in 1946. Since then, a number of attempts have been made to find a machine translation system and put it into operation. The projects were aimed at developing linguistic theories that encompass the whole of natural language, and then go to the computer with it. This approach has inevitably failed, because the human mind cannot encompass a language in its entirety.

Her folger et kort resymé av metodene og teorien som ligger til grunn for kjente metoder: "General Analysis Technique" - GAT - (Institute of Languages and Linguistics Machine Translation Research Center, Washington DC, juni 1959, av Peter Torna, redigert av J„A. Mayne) brukte et lingvistisk orientert computerkodesystem. Ad hoc losninger ble innfiort og hele systemet ble begrenset i et hult, begrenset, s. mul at ed Linguistic Coinputcrsystetn" (SLC), .«skrevet i oktai-spiråk og var ikke åpent: for ytterligere forbedringer. Fulcrumteori-metoden ble utviklet fra 1959 til 1967 av Bunker-Ramo Corporation, USA, og rettet seg mot losning av problemene som forekom i en begrenset russisk tekst med en forholdsvis li-ten ordbok. Det ble ikke gjort forsok på å innfore opplosning av ord med flere betydninger; i stedet ble flere betydninger trykt i det utmatede resultat, atskilt med streker. Here follows a brief summary of the methods and theory underlying known methods: "General Analysis Technique" - GAT - (Institute of Languages and Linguistics Machine Translation Research Center, Washington DC, June 1959, by Peter Torna, edited by J„ A. Mayne) used a linguistically oriented computer coding system. Ad hoc solutions were introduced and the whole system was confined in a hollow, limited, s. mul at ed Linguistic Coinputcrsystetn" (SLC), .«written in octal spiracle and was not open: to further improvements. The Fulcrum theory method was developed from 1959 to 1967 by Bunker-Ramo Corporation, USA, and aimed at solving the problems encountered in a limited Russian text with a relatively small dictionary. No attempt was made to introduce the resolution of words with multiple meanings; instead multiple meanings were printed in the outputted result, separated by dashes.

Et syntaks-system ble utviklet av National Bureau of Standards og Massachusetts Institute of Technology i 1960-1964. Denne metode sviktet, fordi den bare tok hensyn til en begrenset vei til setningen. Dette system ble aldri realisert i storre skala, men bare brukt innenfor et begrenset, eksperimentelt felt. A syntax system was developed by the National Bureau of Standards and the Massachusetts Institute of Technology in 1960-1964. This method failed, because it only took into account a limited path to the sentence. This system was never realized on a large scale, but only used within a limited, experimental field.

Transformerende grammatikk var en annen metode. Men denne viste seg fullstendig uforenelig med de krav som stilles av computer-oversettelse. Det er bare utviklet små, eksperimentelle syste-mer på grunnlag av denne teori, og de måtte avbrytes, for noen oversettelse av betydning ble produsert. Transformative grammar was another method. But this turned out to be completely incompatible with the requirements of computer translation. Only small, experimental systems have been developed on the basis of this theory, and they had to be discontinued, for any translation of meaning was produced.

Sammenlignet med det som er kjent på området,går foreliggende oppfinnelse ut på å muliggjore noyaktig, hurtig, nesten omgåen-de oversettelse av store tekstmengder fra et kildespråk, d„vs. et naturlig språk, til et annet språk, det s.k. mål-språk„ Compared to what is known in the field, the present invention aims to enable accurate, fast, almost immediate translation of large amounts of text from a source language, i.e. a natural language, to another language, the so-called target language

I stedet for et naturlig språk kan et kunstig språk, som f.eks. Esperanto brukes som kilde- eller mål-språk. Instead of a natural language, an artificial language, such as Esperanto is used as a source or target language.

For oppnåelse av dette mål består oppfinnelsen av en fremgangsmåte ved bruk av et programmerbart, digitalt computersystem, hvor de involverte trinn omfatter To achieve this goal, the invention consists of a method using a programmable, digital computer system, where the steps involved include

a) omdanning av en kildetekst som skal oversettes fra synlige tegn til computer-forståelige, binærkodede signaler; b) lagring av den omdannede kildetekst som skal oversettes,i en hovedlagringsenhet for computersysternet; c) avsoking' og sammenligning av de omdannede kildetekstord med ordboker av kildespråkord, lagret i, en lagringsenhet og for a) transformation of a source text to be translated from visible characters into computer-understandable, binary-coded signals; b) storing the converted source text to be translated in a main storage unit for the computer system; c) searching' and comparing the transformed source text words with dictionaries of source language words, stored in, a storage device and for

hvert kildetekstord som det blir funnet; et sidestykke til, lagring i et kartotek i hovedlagringsenheten av hvert ord each source text word that is found; a side piece to, storage in a directory in the main storage unit of each word

og (sic) i tilknytning til hvert slikt ord, lagring av kodet informasjon, utledet fra ordboken, til bruk ved oversettelsen av ordet, hvorved den kodede informasjon omfatter lagringsenhets-forskyvningsaddresseledd til en lagringsenhet i computersystemet, hvor grammatikk og målspråk-oversettelser for ordet er lagret; and (sic) associated with each such word, storing encoded information, derived from the dictionary, for use in the translation of the word, wherein the encoded information comprises storage unit offset address entries of a storage unit in the computer system, where grammar and target language translations for the word are saved;

d) analysering av de omdannede kildetekstord i ordkatoteket, d) analyzing the transformed source text words in the dictionary,

en fullstendig setning av gangen, og omdanning av denne til one complete sentence at a time, and transformation of this into

en setning i målspråket ved bruk av den kodede informasjon og omfattende folgende trinn a sentence in the target language using the coded information and comprising following steps

1) bruk av lagringsenhets-forskyvningsaddresseledd for oppnåelse av målspråkoversettelsene for ord fra en lagringsenhet og 2) omordning av målspråkoversettelsen til korrekt mål-språksekvens, e) omdanning av målspråksekvensen fra computerforståelige, binærkodede signaler tilbake til synlige tegn, 1) use of storage unit offset address term to obtain the target language translations for words from a storage unit and 2) rearrangement of the target language translation into the correct target language sequence, e) conversion of the target language sequence from computer understandable, binary coded signals back to visible characters,

f) utskrivning på målspråket. f) printing in the target language.

Fremgangsmåten omfatter på en fordelaktig måte et trinn, hvor The method advantageously comprises a step, where

en spesiell underrutine for oppslagningsprosessen i ordbok avsoker en liste av ord (leksikal liste) for bestemmelse av.hvilke ord som krever innforing av en spesiell leksikal underrutine i ordbok-kartoteket og et separat leksikalt kontrollprogramm analyserer ordene i kildespråkets setning etter de tilfelle, hvor bare resultatene av den syntaktiske/semantiske analyse av setningen eller tilhorigheten av et ord i en grammatikalsk eller semantisk klasse kan bestemme om en leksikal under-rutine må a special subroutine for the look-up process in the dictionary searches a list of words (lexical list) to determine which words require the introduction of a special lexical subroutine in the dictionary index and a separate lexical control program analyzes the words in the source language sentence on a case-by-case basis, where only the results of the syntactic/semantic analysis of the sentence or the belonging of a word in a grammatical or semantic class can determine whether a lexical subroutine must

tas i bruk på tidspunktet for oversettelsen av et spesielt ord is used at the time of the translation of a particular word

eller uttrykk, for bestemmelse av meningen av dette ord eller uttrykk ved undersøkelse av de syntaktiske relasjoner som er blitt etablert for nevnte ord eller uttrykk og bruk av syntaktiske og/ eller semantiske regler som gjelder bare det ord eller den ord-klasse ordet horer til. or expression, for determining the meaning of this word or expression by examining the syntactic relations that have been established for said word or expression and using syntactic and/or semantic rules that apply only to the word or word-class the word belongs to.

Det er med fordel sorget for en fremgangsmåte for losning av semantiske flertydigheter, hvor hvert ord i stammeordboken (enkelt-ord) er tildelt et enestående begrenset semantisk (LS) tall, It is advantageous to provide a method for solving semantic ambiguities, where each word in the root dictionary (single word) is assigned a unique limited semantic (LS) number,

og de spesialiserte fler-ord uttrykk av LS sammensetningsordbo-ken (LS-uttrykk) som er sammensatt av LS tallhenvisningen for de enkelte ord i stammeordboken, deretter automatisk grupperes til en ordbok-opptegnelse etter uttrykkets hovedord, hvorved det foreligger en for hvert enestående hovedord, hvorpå opptegnelsen siden avsokes for hver gang hovedordet forekommer i den setning som skal oversettes, for bestemmelse om det eksisterer et sidestykke av umiddelbart foregående eller etterfolgende ord i teksten og noen av uttrykkene i opptegnelsen, hvorved den lengste form for sidestykke eller uttrykket ; med hoyest prioritet brukes for å bestemme den spesielle mening med denne ordgruppe, som skiller seg fra summen av meningen, av de enkelte ord; en det mengde av et LS uttrykk er det betingede begrensede semantiske (CLS) uytrykk, som tillater innbefatning av alle definerbare syntaktiske og semantiske regler, likesom enkle programmerings-instrukser som kan benyttes for å forandre den informasjon som er lagret i bits og bytes av setningsanalyseområdetp i selve LS uttrykket (dvs. i ordboken), slik at'ordboks-uttrykket bare and the specialized multi-word expressions of the LS compound dictionary (LS expressions) which are composed of the LS numerical reference for the individual words in the stem dictionary, are then automatically grouped into a dictionary record according to the main word of the expression, whereby there is one for each unique main word , after which the record is then searched for every time the main word appears in the sentence to be translated, to determine whether there is a side piece of immediately preceding or following words in the text and any of the expressions in the record, whereby the longest form of side piece or the expression ; with the highest priority is used to determine the special meaning of this group of words, which differs from the sum of the meaning of the individual words; One set of an LS expression is the conditional limited semantic (CLS) expression, which allows the inclusion of all definable syntactic and semantic rules, as well as simple programming instructions that can be used to change the information stored in the bits and bytes of the sentence analysis area. in the LS expression itself (ie in the dictionary), so that'the dictionary expression only

er gjengitt med sitt sidestykke, når de syntaktiske og/eller semantiske betingelser uttrykt i regelen/reglene tilsvares'av ordet/ordene i teksten, hvorved dette faktum bare kan bestemmes etter en fullstendig syntaktisk og semantisk analyse av setningen utfort av de lingvistiske analyseprogrammer. is reproduced with its counterpart, when the syntactic and/or semantic conditions expressed in the rule(s) are matched by the word(s) in the text, whereby this fact can only be determined after a complete syntactic and semantic analysis of the sentence carried out by the linguistic analysis programs.

Ved fremgangsmåten kan alle kildespråkord med fordel forsynes With the method, all source language words can be supplied with advantage

med semantiske kategorikoder i et variabelt lengdeformat, som deretter utsporres av kildespråk-analyseprogrammene, de leksikale rutiner og CLS ordbok-oppslagningsrutinene som hjelp for losning av semantiske flertydigheter; disse koder er uttrykt i en hierarkisk systematikk (sett av tre-strukturer) på en slik måte at kodingen av en kategori som eksisterer på et lavere nivå av et semantisk tre genererer automatisk koding av de passende koder på hoyere nivå' av selve systemet. with semantic category codes in a variable length format, which are then tracked by the source language analysis programs, the lexical routines and the CLS dictionary lookup routines to help resolve semantic ambiguities; these codes are expressed in a hierarchical system (set of tree structures) in such a way that the coding of a category that exists at a lower level of a semantic tree automatically generates the coding of the appropriate codes at a higher level' of the system itself.

Oppfinnelsen omfatter en fremgangsmåte for drift av en digital computer for oversettelse fra et naturlig kl.ldespråk, f...eks. russisk, til et naturlig målspråk, f.eks. engelsk. Fremgangsmåten omfatter tre faser. Odbok-oppslagningsfasen fastsetter målspråkets betydning av hvert ord eller uttrykk i kildeteksten. Denne ordbok-oppslagningsfase tilfbyer grammatikalske koder og målspråk-ekvivalenter til hvert ord og uttrykk i kildespråket. Den syntaktiske analysefase identifiserer syntaktisk informasjon på basis av den grammatikalske informasjon sorn er foyd til ordene og uttrykkene og utnytter også boyningen av ordet og ordet3 stil-ling i kildeteksten. Syntesefasen tar betydningen og den syntaktiske informasjon av alle ord i en setning i kildeteksten og former en setning i målspråket. The invention includes a method for operating a digital computer for translation from a natural language, e.g. Russian, to a natural target language, e.g. English. The procedure comprises three phases. The Odbok lookup phase determines the target language meaning of each word or phrase in the source text. This dictionary lookup phase provides grammatical codes and target language equivalents for each word and phrase in the source language. The syntactic analysis phase identifies syntactic information on the basis of the grammatical information added to the words and expressions and also utilizes the pronunciation of the word and word3 position in the source text. The synthesis phase takes the meaning and syntactic information of all words in a sentence in the source text and forms a sentence in the target language.

Spesielt begynner fremgangsmåten ved at kildeteksten mates inn In particular, the method begins by entering the source text

i lagringsenheten for en computer. Hvert kildetekstord blir deretter omdannet i et forvandlet kildetekstord. Det forvandlede kildetekstord består av kildetekstordet og kodet informasjon. in the storage device of a computer. Each source text word is then transformed into a transformed source text word. The transformed source text word consists of the source text word and encoded information.

Den kodede informasjon kan omfatte et lagringsenhets-addresse-ledd som gir atkomst til en lagringsenhets-plass som inneholder syntaktisk informasjon og oversettelse for kildetekstordet. De forvandlede kildetekstord som utledes av en kildetekst-setning syntetiseres deretter til en målspråkoversettelse av setningen. Syntesen etablerer korrekt både ordets betydning og ordets stil-ling i målspråkets setning. The coded information may comprise a storage unit address term which provides access to a storage unit location containing syntactic information and translation for the source text word. The transformed source text words derived from a source text sentence are then synthesized into a target language translation of the sentence. The synthesis correctly establishes both the word's meaning and the word's position in the target language sentence.

Et viktig aspekt av- oppfinnelsen er den separate behandling som gis hoyfrekvente ord i motsetning til lavfrekvente ord. For å maksimere den effektive kapasitet av kjernelagringsenheten for computeren, bærer de lavfrekvente ord sin oversettelsesinforma-sjon med seg, mens hvert hoyfrekvent ord bærer et lagringsenhets-forskyvningsaddresseledd, som gir enkel adgang til dets overset-telsesinf ormas jon som er lagret i kjernelagringsenheten. Slik blir oversettelsesinformasjonen for ofte brukte ord holdt på et lett. tilgjengelig sted i computeren, i stedet for sammen med hver forekomst av ordet, slik det gjores for lavfrekvente ord. An important aspect of the invention is the separate treatment given to high-frequency words in contrast to low-frequency words. To maximize the effective capacity of the core storage unit for the computer, the low-frequency words carry their translation information with them, while each high-frequency word carries a storage unit offset address term, which provides easy access to its translation information stored in the core storage unit. This way, the translation information for frequently used words is kept at a glance. available place in the computer, instead of with each occurrence of the word, as is done for low-frequency words.

Mens den ovenfor gitte beskrivelse avbilder en menneskelig ana-logi av hvordan oppfinnelsen funksjonerer, skal bemerkes at den faktiske drift av prosessen med computeren er helt forskjellig. Fra det oyeblikk kildeteksten orndarjnes til maskin-lesbare inngangsdata, inntil det tidspunkt, da de maskin-lesbare utgangs-data omdannes til en vanlig lesbar oversettelsestekst, gjennom-føres prosessen under styring av et computerprogram. Mens det er beleilig å beskrive programmets trinn, som om de ble utfort av en menneskelig oversetter, skjer i realiteten intet slikt. Computeren utforer tvert om en rekke automatiske, abstrakte, mate-matiske operasjoner med de abstrakte verdier som er lagret i computerens lagringsenhet. Programmet funksjonerer uavhengig av betydningen eller vekten av de data som det arbeider med. Det forhold at programmet er utformet i et programmeringsspråk på hoyt nivå, som tilsynelatende lar programmet gi signifikans til maskinoperasjonen, endrer ikke det faktum at maskinen faktisk gjennomforer en rekke abstrakte trinn, som intet har å gjore med oversettelse mellom naturlige språk. Hvis en annen slags informasjon ble matet inn i computeren, kunne programmet som benyttes .ifolge oppfinnelsen, muligens utfore en funksjon som er fullstendig annerledes enn oversettelse. While the description given above depicts a human analogy of how the invention functions, it should be noted that the actual operation of the process by the computer is completely different. From the moment the source text is arranged into machine-readable input data, until the time when the machine-readable output data is converted into a normal readable translation text, the process is carried out under the control of a computer program. While it is convenient to describe the steps of the program as if they were performed by a human translator, in reality no such thing happens. On the contrary, the computer performs a series of automatic, abstract, mathematical operations with the abstract values stored in the computer's storage unit. The program functions independently of the meaning or weight of the data it works with. The fact that the program is designed in a high-level programming language, which apparently allows the program to give significance to the machine operation, does not change the fact that the machine actually carries out a series of abstract steps, which have nothing to do with translation between natural languages. If some other kind of information were fed into the computer, the program used in accordance with the invention could possibly perform a function completely different from translation.

Oppfinnelsen omfatter også den prosess ved hvilken informasjon trekkes ut fra computeren, inklusive utskrivning av informasjonen, dvs. det trinn som omdanner målspråksekvensen fra computerles-bare binærkodede signaler tilbake til synlige tegn. The invention also encompasses the process by which information is extracted from the computer, including printing the information, i.e. the step which converts the target language sequence from computer-readable binary coded signals back into visible characters.

Det er en stor fordel at ethvert kildespråk ved hjelp av foreliggende oppfinnelse kan oversettes til et målspråk ved bruk av den foreslåtte fremgangsmåte og ved bruk av samme utstyr, ganske enkelt ved at de språkavhengige deler av utstyret, som ordboker og lingvistiske programmer som inneholder alle syntaktiske og semantiske regler for begge språk skiftes ut. It is a great advantage that any source language with the help of the present invention can be translated into a target language using the proposed method and using the same equipment, simply by the language-dependent parts of the equipment, such as dictionaries and linguistic programs containing all syntactic and semantic rules for both languages are exchanged.

En annen fordel ved fremgangsmåten ifolge oppfinnelsen er den konsekvente oversettelse, dvs. et bestemt uttrykk oversettes fra kildespråket til målspråket med samme kontekst; skifter konteks - ten, vil oppfinnelsen folgelig fore til at uttrykkets betydning skifter med ovennevnte konsekvens. Det er også en fordel at alle syntaktiske og semantiske trekk ved kildespråket gjenkjennes og reflekteres i målspråket. Another advantage of the method according to the invention is the consistent translation, i.e. a specific expression is translated from the source language to the target language with the same context; if the context changes, the invention will consequently cause the meaning of the expression to change with the above-mentioned consequence. It is also an advantage that all syntactic and semantic features of the source language are recognized and reflected in the target language.

Spesielle utforelseseksempler av oppfinne!sen skal nå "beskrives under henvisning til tegningene,hvor Special embodiments of the invention will now be described with reference to the drawings, where

fig. 1 illustrerer oversettelsesbehandlingen, fig. 1 illustrates the translation processing,

fig. 2 viser et skjema for ordbokoppslagningen og fig. 3 viser et skjematisk sammendrag av programmer. fig. 2 shows a diagram for the dictionary lookup and fig. 3 shows a schematic summary of programs.

Som vist i fig. 1 mates tekster inn i oversettelsessystemet enten on-line, via en terminal eller off-line, via tast-til pla-te, eller tast-til-båndanordninger. (Innmating via punchkort er også mulig.) Den innmatede tekst og oversettelses-ordbokene og programmene fins på platekartotek som er tilgjengelige for computeren som utforer oversettelsen. Resultatene skrives deretter ut i det onskede format på skriveren etter As shown in fig. 1, texts are fed into the translation system either on-line, via a terminal or off-line, via key-to-record or key-to-tape devices. (Input via punch cards is also possible.) The entered text and the translation dictionaries and programs are on record cards that are accessible to the computer that performs the translation. The results are then printed in the desired format on the printer

at skriveren har tatt hensyn til eventuelle formatkoder i den innmatede tekst. that the printer has taken into account any format codes in the entered text.

Etter-redigering av den utmatede tekst og eventuelt resulteren-de ordbok-ajourfbring kan, om onsket, finne sted via on-line gjensidig påvirkbar terminal (eller kan utfores off-line av etter-redaktorer og ordbok-koder). Post-editing of the outputted text and any resulting dictionary updating can, if desired, take place via an on-line interoperable terminal (or can be carried out off-line by post-editors and dictionary codes).

Diagrammer og tabeller som må inngå i samme form i det oversatte utgangsmateriale, kan opptas i teksten via en fotokompo-sisjonsanordning. Diagrams and tables that must be included in the same form in the translated source material can be included in the text via a photocomposition device.

Fig. 2 illustrerer det program som mater inngangsteksten i computerens lagringsenhet (LØADTXT), identifiserer de høyfrekvente ord og uttrykkene i teksten og gir hvert ord i teksten et eget serienummer. Det tilfoyer umiddelbart en oversettelse til uttrykkene og sikrer at informasjonen for alle hoyfrekvente ord er umiddelbart tilgjengelig i computerens kjernelagringsenhet under oversettelsesprosessen. De gjenstående (lavfrekvente) ord sorteres deretter i alfabetisk orden for å lette hoved-ordbokoppslagningen (MDL). Hoved-ordbokoppslagningen gir hvert lav-frekvent ord den nodvendige grammatikalske informasjon likesom en kryss-henvisning til opptegnelsen av begrensede fler-ord semantiske (LS ) sammensetninger som inneholder det ordet. Etter hoved-ordbok-oppslagning gjen-sorteres alle. ord i teksten i deres opprinnelige rekkefolge. Fig. 2 illustrates the program that feeds the input text into the computer's storage unit (LØADTXT), identifies the high-frequency words and expressions in the text and gives each word in the text a separate serial number. It immediately adds a translation to the expressions and ensures that the information for all high-frequency words is immediately available in the computer's core storage unit during the translation process. The remaining (low-frequency) words are then sorted alphabetically to facilitate the main dictionary lookup (MDL). The main dictionary lookup provides each low-frequency word with the necessary grammatical information as well as a cross-reference to the record of limited multi-word semantic (LS ) compounds containing that word. After the main dictionary lookup, everyone is re-sorted. words in the text in their original order.

Ifolge fig. 3 styrer programmet INITCALL den faktiske overset-telsesprosess. Det anroper forst programmet GETSENTN, som etablerer et setningsanalyseområde i computerens lagringsenhet, bestående av 160 bytes informasjon med fast lengde for hvert ord, med kryssreferanser til ytterligere informasjonsområder med informasjon av variabel lengde» According to fig. 3, the program INITCALL controls the actual translation process. It first calls the program GETSENTN, which establishes a sentence analysis area in the computer's storage unit, consisting of 160 bytes of fixed-length information for each word, with cross-references to additional information areas with information of variable length.

Forst inneholder setningsanalyseområdet bare ordbok-informasjonen for hvert ord. Under oversettelsesprosessen, tilfoyes ytterligere informasjon av kildespråk-analyseprogrammene (betegnet som "tilpasninger"), som loser syntaktiske og semantiske flertydigheter, etablerer setningsgrenser, beskriver grunnleggende, syntaktiske relasjoner mellom ord, identifiserer subjekt/predikat-forhold og analyserer eventuelle preposisjoners funksjon. First, the sentence analysis area only contains the dictionary information for each word. During the translation process, additional information is added by the source language analysis programs (referred to as "adaptations"), which resolve syntactic and semantic ambiguities, establish sentence boundaries, describe basic syntactic relations between words, identify subject/predicate relationships and analyze the function of any prepositions.

Begrenset semantisk (LS) og betinget begrenset semantisk (CLS) ordbok-oppslagning finner sted på passende steder i analysen, hvorved betinget begrenset semantisk oppslagning bare er mulig etter at hele setningen er blitt analysert. Limited Semantic (LS) and Conditional Limited Semantic (CLS) dictionary lookup takes place at appropriate points in the analysis, whereby conditional limited semantic lookup is only possible after the entire sentence has been analyzed.

De gjenstående fire trinn, som syntetiserer målspråket på grunnlag av den totale informasjon i setningsanalyseområdet, er: oversettelse av preposisjoner, losning av ord-spesifikke pro-blemer ved hjelp av leksikale rutiner, den egentlige syntese av målspråkordene og omredigeringen av setningen til korrekt ord-rekkefolge i målspråket. The remaining four steps, which synthesize the target language on the basis of the total information in the sentence analysis area, are: translation of prepositions, solving word-specific problems using lexical routines, the actual synthesis of the target language words and the re-editing of the sentence into the correct word- sequence in the target language.

Siste trinn i oversettelsesprosessen er utskriving av det oversatte utgangsmateriale ved program TRPRINT. The last step in the translation process is printing the translated source material using the program TRPRINT.

Claims

1. Procedure for translation between natural source and target languages using a programmable, digital computer system, characterized by the following steps: a) a source text to be translated is converted from visible characters into computer-understandable, binary code signals; b) storing the converted source text to be translated; in the main storage unit of the computer system; c) scanning and comparing converted source text words with dictionaries of source text words, stored in a storage unit and storing each source text word for which a counterpart is found, in an index in the main storage unit and (sic) in association with each such word, coded information derived from the dictionary for use in the translation of each such word, whereby the encoded information comprises storage unit offset address links to a storage unit in the computer system, where grammar and target language translations for the word are stored; d) analyzing the transformed source text words in the dictionary, one complete sentence at a time, and transforming the same into a sentence in the target language using the coded information and including the following steps: 1) utilizing the storage unit offset address links to obtain the target language translations of words from the storage unit and 2) re-editing the target language translation into the correct word order in the target language; f) printing in the target language. 2„ Method as stated in claim 1, characterized by the following steps: storage of a dictionary for high-frequency source words and assigned displacement address links, whereby the displacement address links identify the storage location for grammar and meaning information about the source words; storing a dictionary of low frequency source words associated with grammar code meanings for each word; a stem word from the source language together with a displacement clause thereby identifies the location of target language grammar codes^and meanings; comparison of the words of the source language text with the high-frequency dictionary words and, after registration of equivalence with a word, storing the word and assigned displacement address term together in a high-frequency index; and comparing source text words with the low-frequency dictionary words and by recording similarity, storing the word and assigned grammar code meanings in a low-frequency index.

3. Method as set forth in claim 2, characterized in that second and subsequent expression words, if such exist, in a source language expression are stored in at least one dictionary in the storage unit; The high-frequency dictionary includes the first words in expressions in the source language and, in connection with these, address links to other and subsequent words in the same expression, which are placed in the expression dictionary, whereby the first word and any subsequent words in an expression have an assigned target language meaning stored in connection with them; and during comparison with the high-frequency dictionary a) registration of the words that are similar and are the first word in an expression; b) utilization of the offset address terms for locating the additional word or words in the same expression, which are placed in the same expression dictionary; c) comparing the located additional words in the phrase with the words in the source text following the first phrase word to find similarity; and d) after registration of such similarity, storage of the first expression word together with the meaning of the target language in the high-frequency index; e) storing each of the second and subsequent expression words together with an expression mark in the high-frequency index, whereby the expression mark indicates that the word is not the first word in the expression; f) storing the first word in an expression together with a mark in the high-frequency index, whereby the mark indicates that the word is the first word in an expression; g) during analysis, selecting the meaning of the target language in connection with a first expression word as the translation of the first and subsequent expression words. in

4. Method as set forth in claim 1, which includes storing an indication in connection with the words in the low-frequency index whether the words are main words and/or words that form part of limited semantic connections; whereby such connections, which are stored in separate index cards, are stored containing limited semantic numbers for each main word, the limited semantic numbers indicating all participating words, whereby the limited semantic numbers for participating words are associated with the main words in the same order as the participating words form expressive connections with the main word; characterized in that information during search and comparison with the dictionaries is attached to the main word in relation to all compound expressions in which the main word can act as a key word in a connection, whereby the added information contains either rows of limited semantic numbers or an address to an auxiliary storage unit, where such information is located, and including the additional steps that the limited semantic numbers indicating a subject-field code. to the source text is brought into a main storage unit from the auxiliary storage unit during the parsing of the word index, if the same compound expression in the source language, indicated by the limited semantic numbers, has different target language meanings, then selection of the correct meaning of the expression, depending on the subject field code, where the translation takes place.

5. Method as stated in claim 1, characterized in that the analysis step includes analyzing the source text and placing the source language sentence in a specific area, reserved for this purpose in the computer storage unit, whereby the analysis moves from the beginning to the end and from the end to the beginning of the source language sentence , and indicate marks for special relations of the sentence context, and that the marks are set by routines in the computer program which are provided in response to different parts of the language among the words; during the analysis, information is stored for use for recollection from one source text sentence to the next, to enable the translation of the words that require control from the previous sentence; during the analysis, semantic and syntactic adjustments are also made by adding coded information to each source sentence, whereby the adjustments are made from beginning to end or from end to beginning, depending on the codes that have been added to the individual words, and on the nature of the source language; and during the scanning, comparison and storage, parts of language codes are added to source text words, so that the added parts of language codes are used during the analysis for distribution to a desired routine by adding the value of the language part to a fixed address for carrying out an analysis; that the computer, while analyzing the words of the source language, keeps abreast of previously encountered situations by setting a number of storage device switches; each time a new language part occurs, a number of previously set switches are selectively turned off, while a. or several additional switches are selectively turned on, whereby there is a routine for derailing the switches in a particular order; during analysis, each word in the sentence is also provided with specific, coded information about whether it belongs to a main or subordinate clause, a routine is added to search the source words in such a sentence for predicate(s) and subject(s), later, during the analysis, predicate(s) and subject(s) in the source language are searched by routines and correspondingly marked with codes, whereby the predicate(s) and subject(s) are searched by the routines for interjected sentences, which are thereby taken into account, as the routines determine whether there is a clue whether the function of a comma or a conjunction is to begin a new sentence, and performs a separate search for predicate(s) and subject(s) and determines the clue as a result of this analysis, each source word in the sentence is provided with marks expressing the independence relation between all words in the sentence; at the same time, the syntactic function is added to each word in a sentence, the syntactic function being expressed in compressed form using coded bits ("bis"), whereby the syntactic analysis continues until every word in the sentence has been assigned at least one function.

6. Method as stated in claim 1, characterized in that a dictionary of subject glossary information is stored and contains different target language meanings for a source word in accordance with different technical subjects, including a program routine in the step for examining each source sentence during the analysis and for selecting meanings from subject words - the book that corresponds to the technical subject the sentence refers to.

7. Procedure as stated in claim 1, characterized by the fact that the source language sentences are analyzed in hierarchical order, "whereby a first adaptation resolves basic ambiguities, a second, subsequent adaptation establishes primary, syntactic mutual connection and a third, subsequent adaptation expands these mutual connections and delimits phrases and sentences, a fourth, subsequent predicate-subject matching searches in hierarchical order for predicates and subjects and resolves the remaining sentence separation ambiguities, and a fifth, subsequent matching ensures that the function of each word in the sentence is correctly registered* 8„ Procedure as stated in claim 1, characterized in that the main syntactic adaptations call generalized subroutines for determining and coding syntactic relations, and the subroutines in turn selectively call smaller subroutines for carrying out basic tasks.

9. Method as stated in claim 1, characterized in that a separate program determines and analyzes the prepositions in the source language and determines their inherent meaning, whereby the meaning of a preposition is either determined by words that control the preposition or by words that function as the preposition, whereby coded information for translation of one or more prepositions is added to the words from the dictionaries.

10. Method as stated in claim 1, characterized by the step that each word is analyzed during several adaptations of each sentence, whereby codes are added, evaluation of all codes that previous adaptations added to a word and addition of target language synthesis codes to the meaning that the code function has in the sentence, placement of the word in a form that corresponds to the target language depending on the analysis and assessment of all relevant codes that have been added to the word, whereby the translated words are formed after analysis in a printing area in a sequence determined by the relationships between the source and target languages, as previously set up by the numbers given after each word and transfer of the translation, when the printing area is filled, either directly or to a printer or to a temporary a auxiliary storage unit.

11. Method as stated in claim 1, characterized in that it comprises a step, where a special sub-routine of the dictionary look-up process searches a word list (lexical list) to determine which words require the introduction of a special lexical sub-routine in the dictionary index and where a separate lexical control program analyzes the words in the source language sentence in those cases, where only the results of the syntactic/semantic analysis of the sentence or the occurrence of a word in a grammatical or semantic category can determine whether a lexical subroutine must be called at the time of the translation of a specific word or expression for determining the meaning of the word or expression by examining the syntactic relations that have been established for the word/expression and utilizing syntactic and/or semantic rules that only apply to this word or the word category to which it belongs.

12. Procedure as stated in claim 1 for solving semantic ambiguities, characterized by the fact that each word - in the stem (single word) dictionary is added a separate, limited, semantic (LS) number, after which the specialized multi-word expressions in the LS compound dictionary ( LS expression), sorn is composed of LS numbers that represent the individual words in the stem dictionary, are automatically grouped into a dictionary entry according to the expression's main word, of which there is one for each individual main word, after which the entries are separated every time the main word occurs in the sentence that must be translated, to determine whether there is similarity between a group of consecutive words in the text and one of the expressions in the entries, whereby the longest match or the expression with the highest priority is used to determine the specialized meaning of said word group, which differs from the sum of the meanings of the individual words and a subset of an LS expression is the conditional, limited, semantic (CLS) expression kk which allows the inclusion of all definable syntactic and semantic rules, as well as simple programming instructions, which can be used to change the information stored in bits and bytes in the sentence analysis area, in the LS expression itself (i.e. in the dictionary), so that the dictionary expression is only matched, when the syntactic and/or semantic conditions that are expressed, in the rule(s) are satisfied by the word(s) in the text, whereby this relationship can only be determined after a complete syntactic and semantic analysis of the sentence, carried out by the linguistic analysis programs.

13"Procedure as specified in claim 1, characterized in that all source language words can be supplied with semantic category codes in a variable length format, which are then traced by the source language analysis programs, the lexical routines and the CLS dictionary lookup routines to help resolve semantic ambiguities; these codes are expressed in a hierarchical systematic (set of tree structures) in such a way that the coding of a category that exists at a lower level of a semantic tree automatically generates the coding of the appropriate codes of higher levels of the system itself.

14. Method as stated in claim 1, characterized in that it includes analyzing a series of words in sequence in the source language within phrases and sentences in relation to the target language word sequence, whereby the target language word sequence is expressed symbolically by added numbers and that the translation is printed taking into account each source word.