CA2125200C - Language translation apparatus and method using context-based translation models - Google Patents
Language translation apparatus and method using context-based translation modelsInfo
- Publication number
- CA2125200C CA2125200C CA002125200A CA2125200A CA2125200C CA 2125200 C CA2125200 C CA 2125200C CA 002125200 A CA002125200 A CA 002125200A CA 2125200 A CA2125200 A CA 2125200A CA 2125200 C CA2125200 C CA 2125200C
- Authority
- CA
- Canada
- Prior art keywords
- word
- target
- source
- hypothesis
- words
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Expired - Fee Related
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/40—Processing or translation of natural language
- G06F40/42—Data-driven translation
- G06F40/44—Statistical methods, e.g. probability models
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/40—Processing or translation of natural language
- G06F40/42—Data-driven translation
- G06F40/45—Example-based machine translation; Alignment
Abstract
An apparatus for translating a series of source words in a first language to a series of target words in a second language. For an input series of source words, at least two target hypotheses, each comprising a series of target words, are generated. Each target word has a context comprising at least one other word in the target hypothesis. For each target hypothesis, a language model match score comprises an estimate of the probability of occurrence of the series of words in the target hypothesis.
At least one alignment connecting each source word with at least one target word in the target hypothesis is identified. For each source word and each target hypothesis, a word match score comprises an estimate of the conditional probability of occurrence of the source word, given the target word in the target hypothesis which is connected to the source word and given the context in the target hypothesis of the target word which is connected to the source word. For each target hypothesis, a translation match score comprises a combination of the word match scores for the target hypothesis and the source words in the input series of source words. A target hypothesis match score comprises a combination of the language model match score for the target hypothesis and the translation match score for the target hypothesis The target hypothesis having the best target hypothesis match score is output.
At least one alignment connecting each source word with at least one target word in the target hypothesis is identified. For each source word and each target hypothesis, a word match score comprises an estimate of the conditional probability of occurrence of the source word, given the target word in the target hypothesis which is connected to the source word and given the context in the target hypothesis of the target word which is connected to the source word. For each target hypothesis, a translation match score comprises a combination of the word match scores for the target hypothesis and the source words in the input series of source words. A target hypothesis match score comprises a combination of the language model match score for the target hypothesis and the translation match score for the target hypothesis The target hypothesis having the best target hypothesis match score is output.
Description
212~200 LANGUAGE TRANSLATION APPARATUS ~ND METHOD USING
CONTEXT-BASED TRANSLATION MODELS
Background of the Invention The invention relates to computerized language translation, such as computerized translation of a French sentence into an English sentence.
In Canadian patent application serial number 2,068,780-1 filed May 15, 1992, (laid-open to the public on January 26, 1993), entitled "Method and System For Natural Language Translation" by Peter F. Brown et al there is described a computerized language trans]ation system for translating a text F in a source langua-3e to a text E in a target language. Tlle system described t}lerein evalua-tes, for each of a number of hyE)otllesizecl target- lancJuage texts E, the conditional probability P(E'F) of t.he target language text E given the source language text F. The hypothesized target language text ~ havi,llg tl~e hic3llest conditional probability P(EIF) is se]ected as t,he t,rans]atioll of t,}~e source ]anguage text F.
212~2~0 Using Bayes' theorem, the conditional probability P(EIF) of tlle target language text E given the source language text F Call be written as P(EIF)=~ p(-F~
Sil~ce tlle l~rob.~bility P(F) of tlle ~ource lan~uage text F in the onli llat:ol- of E(luatio~ 1 i s i ll(lel)ell(lellt~ of tlle tar~et lallguag~
t:ext: E, tlle ~ <l~t, 1<III(I~Ia(~ lexl: ~ llaVill(J t,ll~ llest ~on(litiollal l)rObabi]itY P(EIF) W;]] a]~;O llaVe tI1e }~ ]leSt I)rOdUCt P(FIE) P(E) .
We t}lerefore arrive at E--argmaX P(F ¦ E)P(E) E (2) ~n E~uati.on 2, the pro~ability P(E) of tlle t:arqet language text, F, is a l.anguage model matcll score allcl may ~e estim~te(l from a t:ar~et language model. Wllile ally knowll l.allguage mo~lel may )~e used to estimate the probability P(E) of t~le target lallguage text E, Brow et al descxibe an n-gram language mo~e] ~oml)risin~J a 1-<~ralll mo~le], a 2-gram model, and a 3-gram model combined by parameters whose values are obtainea by interpolated estimation.
The conditional probability P(FIE) in Equation 2 is a translation match score. As described by Brown et al, the translation match Yo993-089 - 2 -~ ~ ~o~ ~Q ~
score P(F I E) for a source text F comprising a series of source words, given a target hypothesis E
co""" i~in~ a series of target words, may be estim~ted by finding all possible ~li&n~el~s connecting the source words in the source text F with the target words in the target text E, inclu-line alignments in which one or more source words are not connected to any target words, but not including llllr~ where a source word is com~e-;led to more than one target word. For each alignment and each target word e in the target text E connected to ~ source words in the source text F, there is estimated the fertility probability n(~ I e) that the target word e is connected to the ~ source words in the alignment. There is also estim~ted for each source word f in the source text F and each target word e in the target text E connected to the source word f by the ~lignm~nt, the lexical probability t(f I e) that the source word f would occur given the occurrence of the connected target word e.
For each alignment and each source word f, Brown et al further estim~te the distortion probability a(j I apm) that the source word f is located in position j of the source text F, given that the target word e conne~;led to the source word f is located in position aj in the target text E, and given that there ar m words in the source text F.
By co",bining the fertility probabilities for an alignment and for all target words e in the target text E, and multiplying the result 212~2~30 by tlle probability n~(~O~ ) of the number ~O of target words not i =l connected Wit]l any source words in the alignment, given the sum of the fertilities ~ of all of the target words in the target text E in the alignment, a fertility score for the target text E and the alignment is obtained.
~y combilling the lexical prol~abilities for an alignmellt and for <~ ource wor~ls in tlle source t-ext F, a lexical score for the ~ lmellt: is ol)t.~
By coml)illill~ the dis î:ortioll prol)al~ilities for ~n alignmellt and for all source words i.J~ t,he ~ource text E whic}l are connected to a t.~rq~t: wor(l ill t:)le ali(J~ ellt~ y m~llt,il)lyi.ll(J the resu]t ~y jl, (wlleLe ./", is 1,1~e nllnlbel: of ~,~lr~let wor(ll; ill the t,arget text E
tllat are not conllected wit)l ally source wor~ls), a distortion score for the ali~llmellt isi obtaille~
Finally, by com~)illin~ the ferli.lit:y, lexic~l, alld distortio scores for the alignmellt, and multiplyin~ the result by the combinatori~l f~ctor ~ j!), a t:rallslat-ioll m.lt:ch score for tlle J
alignment is obtailled. (See, Brown et al, Section 8.2.) The translation match score P(FIE~ for the source text F and the target hypothesis E may be the sum of the translation match scores for all permitted alignments between the source text F and the target hypothesis E. Preferably, the translation match score 21~200 P(FIE) for the source text F and the target hypotllesis E is the translation match score for the alignment estimated to be most probable.
Equation 2 may be used to directly estimate the target hypot}lesis matc}l score P(F¦E)P(E) for a hypot}lesi7ed target language text E
alld a source lallguage text F. Ilowever, in order to simply the lallg-lage model ~(E) and tlle trallslation model P(EIE), and in order l:o estimate tlle l)arallleters of tllese models from a manageable ~moUIIt of traillillg ~ata, srowll et al estimate tlle tar~et l~ypotllesis m<ltcll scoLe P(FIE)~'(E) for sim~)lifiecl illterme~iate forms E' ~nd F' of tlle tal-get ]allgu~ge text E and the source lc~ xl ~ iv~ly. Ec-l~ll illt~rnl~ lte tc~rget~
word e' represe~lts a class, of related target language words. Eacl illtermediate source language word f' represents ~ class of related ~o~lr~e ]~llgUage worlls. ~ s~ul-ce lallg~la(Je t r.lllSd~l~'eL ~'OIIVert~ le source language text F t(~ t)le illterme(li~lte form F'. Tlle )Iypothesi7.ed intermediate form target lallguage text E' llavillg tlle highest hypothesis matc}l score P(E'¦E')P(E') is estimated from Equation 2. ~ target language transducer converts the best matched intermediate target lallguage text E'to tlle target lallguage text E.
In their language translation system, Brown et al estimate the lexical probability of each source word f as the conditional probability t(fle) of each source word f given solely the target 21~2~0 words e connected to the source word in an alignmellt.
Consequently, the lexical probability provides only a coarse estimate of the probability of the source word f.
summary of the Invelltioll It is al~ object of t}le invellt:iollto provide an apparatus and metllod for translatillg a series of source words in a first langu~ge to a serie. of l-al-(3et wor~ls in ~ sec o~ c~u~ e (li~ferent f rom t lle first language, in wllich tlle al)p~ratus alld met-~lod provide improved estimates of tlle lexical pLo~al~ilities of tlle source words.
It is anotller o~ject of the invelltioll to l)rovide an apparatus an(l metllod for translating a series of soul-ce words in a first l~ngllage to a series of target wor~s in a secoll~ langu~qe different from the first language, in which the lexical E)robability of a source word is estim~ e(l as a conditiotlal probability given the target word conllected to t~le source word in an alignment, and given the context of tlle target- word conllectecl to the source word.
According to the invention, an apparatus for translating a series of source words in a first language to a series of target words in a second language different from the first language comprises means for inputting a series of source words. Means are also 2125~00 provided for generating at least two target hypot}leses. Each target hypothesis comprises a series of target words selected from a vocabulary of words in the second language. Each target word has a context comprising at least one other word in the target hypothesis.
language model match score generator generates, for each target ]lyI)otl~esis, a lallgu~ge model matcll score comprising an estimate Or t:lle pro})cll)ility of~ OCCUrrell<e of the series of words in tlle t,arget hyl)otlle~is. An aligllmellt identifier identifies at least c~lle aligllmellt ~etweell the inl~nt series of source words and each target }~yl)ot})esis. Tl~e alignmellt collnects each source word with at, }e~l~it, olle tar~1et, wor(l ill tlle t-arget, llyl)o~-lle.sis.
word mat(}l score generator is l~rovided for generatillg, for each source word all~l eacll target llypotllesis, a word matcll score comprising an estimate of the conditiol~al l~ro~ability of occurrence of t}le source word, givell 1he t,ar~et word in the target llypotllesis whicll is conllected to tlle so~lrce word and given the context of the target word in the target hypothesis whicll is connected to the source word. A translatioll match score generator generates, for each target hypot}lesis, a translatiotl match score comprising a combination of the word match scores for the target hypothesis and the source words in the input series of source words.
212~2~
A hypothesis matc}l score generator is provide-l for generatillg a target hypothesis match score for each target hypothesis. Each target hypothesis match score comprises a combination of the language model match score for the target hypothesis and the translation match score for the target hypothesis. The target hypothesis havillg the best target hypothesis match score is provided at an output.
Prefera~ly each target hypotllesis comprises a series of target words selecte~ from ~ vo(nll~nl~ry comprisillg words in tlle ~con l~ngnclqe al~cl a lull wor(l repre;elltillg tlle ~sence of a word.
The alignmellt i(lentifier may coml)ri-;e meall.s for identifyil-g two or more alignmellts ~etween the input series of source wor~s and eac}l target hypot}lesis. Each alignmellt conllects each source word Wit~l at least one targel worc~ in t~le tar(~et llypotllesis. The worl match score generator generates for e~ch source word and eac}
alignment and each target hypot}lesis a word matc}l score comprising an estim~te of the conditional pro~a}~ility of occurrence of the source word given the target word connected to the source word and given the context of tlle target word. The translation match score generator generates for each target hypothesis a translation match score comprising a ~ombinatioll of the word match scores for the target hypothesis and the alignments and the source words in the input series of source words.
Yo993-089 - 8 -212~2~
Tlle source text il~put device may coml,rise mealls for tral~sformillq the input series of source words into a series of transformed source words. The alignment means identifies at least one alignment between the series of transformed source words and each target hypothesis. Each alignment connects each transformed source word with at least one target word in the target llypotllesis. The word match score generator generates for each transformed source wor~ nd each target hypot]lesis, a word matc~
score eomprisillg an estimate of tlle col~ditional pro~al~ility of occurrence of the trallsformed source word, givell the target wor~
in the target hypotllesis whic}l is conllected to the transforme~
wor~ and given the context of the target word in the target }~ypot}leiis whic}l is eolllleete~ 1o t}le trc~nsfornle~ souree word.
T}~e t l-c~llSl~tiOIl matcll score gellerator gellerates for each target llypot~lesis ~ t-rallsl~liol~ mat(ll ;cor~ coml)risil~g a coml~ atioll of the word match scores for the target hyl)ot-}lesis an(l t}le transformed source word. output means are provided for synthesi7ing a series of output words from the target }~ypot}lesis llaving the best target hypotllesis match score alld for OUtpUtti tl~e output wor~s.
The translation match score for a target hypothesis may comprise the product of the word match scores for the target hypotllesis and the source words in the input series of source words. The target hypothesis match score for a target hypothesis may comprise the Yo993-089 _ 9 -21252~
product of tlle language model match score for the target hypothesis multiplied by the translation match score for the target hypotllesis.
The context of the target word in tlle target hypothesis whicll is connected to a source word may be contained in at least one of two or more context classes. The estim~ted conditiollal probability of occurrence of the source word, ~iven tlle target word connected 1:O tll~ ';O~lrCe WC)Ld .Illd (1ivell t,he collt:ext, of t-lle target word, coml~rises ~t le~st one fnllcti,on h~villg ~ v~lue dependellt on t}le elass COIIt~illillg tlle cont:ext of the t~rget word.
Altel.ll~tiVe]y, the eStilll~lted ('Olldi t iOII~I] prol)ability of occurrence of t}le source word may ~onlprise a fun~t,ioll }l~vin9 a value depelldellt Oll the p~rt of speech in the target llypothesis of at le~st one word in t,}le context of tlle t,c~r~let, word W~liCll iS COIllleCt to tlle source word, or del)elldellt on tlle idellti,ty of at le~st one word in the context of tlle target word wllic~l is conllected to the ~ource word.
The me~ns for outputting tlle target hy~otllesi~ h~Vill9 the best match score may comprise a display. The input means may comprise a keyboard, a computer disk drive, or a computer tape drive.
By estimating the lexicaL probability of a source word as the conditional proba~ility of the source word given the target word 21252~û
conllected to tlle source word ill c~n aligllmellt, alld given tlle context of the target word connected to the source word, the present invention obtains translation match scores with improved accuracy.
Brief Description of the Drawinqs Figure 1 is a l~lock diagram of an example of an apparatus for trallslatillg words from a first language to a second language accordillg to tlle present inventioll.
Figure 2 scllematical.ly sl~ow.~ all ex.~ le of all aligllment betWeell a l~ypot)letical series of source words and a hyl)ot]~etical series of 1arget words.
Figure 3 scllematically s~ows an examl)le of a second alignmellt between t}le hypotlletical series of source words and the hypot}letical series of target words of Figure 2.
Figure 4 schematically sllows an example of a tllird aligllment between the hypothetical series of source words and the hypothetical series of target words of Figure 2.
Yo993-089 - 11 -~ 6 D
Desc~ ion ofthe Plerelled Embodiments Figure I is a block diagram of an eA~"p'e of an appalal~ls for tranQl~ting words from a first l~n~-~e to a second l~age accolding to the present invention. The apparatus co",p,ises a source text input device 10 for inputting a series of source words. The source text input device 10 may comprise, for example, a keyboard, a computer disk drive, or a computer tape drive.
The source text input device 10 may further comprise means for tran~ro"~ g the input series of source words into a series of ~,~nsro""ed source words. Each llansrolllled source word may ,~,~se,lt a set of related input source words. For example, each ofthe input source words in the set {etre, etant, ete, suis, es, est, sommes, etes, sont, fus, fumes, serai, serons, sois, soit, soyons, soyez, soient} are forms of the verb infinitive "etre", and may be ~ sr~""led to "etre" with a tag I epl ese"ling the tense of the original input source word.
The means for l~nsl~",ling the input series of source words into a series of llansr~Jlllled source words may col~p,ise the source tr~n~lucers desclil)ed in Sections 3, 4, and 11 of Brown et al, above.
In essence, these tr~nsd~lcers check and correct the spelling of the input source words, check and correct the case of the input source words, detect titles of documPnt~ in the input series of source , .
..
CONTEXT-BASED TRANSLATION MODELS
Background of the Invention The invention relates to computerized language translation, such as computerized translation of a French sentence into an English sentence.
In Canadian patent application serial number 2,068,780-1 filed May 15, 1992, (laid-open to the public on January 26, 1993), entitled "Method and System For Natural Language Translation" by Peter F. Brown et al there is described a computerized language trans]ation system for translating a text F in a source langua-3e to a text E in a target language. Tlle system described t}lerein evalua-tes, for each of a number of hyE)otllesizecl target- lancJuage texts E, the conditional probability P(E'F) of t.he target language text E given the source language text F. The hypothesized target language text ~ havi,llg tl~e hic3llest conditional probability P(EIF) is se]ected as t,he t,rans]atioll of t,}~e source ]anguage text F.
212~2~0 Using Bayes' theorem, the conditional probability P(EIF) of tlle target language text E given the source language text F Call be written as P(EIF)=~ p(-F~
Sil~ce tlle l~rob.~bility P(F) of tlle ~ource lan~uage text F in the onli llat:ol- of E(luatio~ 1 i s i ll(lel)ell(lellt~ of tlle tar~et lallguag~
t:ext: E, tlle ~ <l~t, 1<III(I~Ia(~ lexl: ~ llaVill(J t,ll~ llest ~on(litiollal l)rObabi]itY P(EIF) W;]] a]~;O llaVe tI1e }~ ]leSt I)rOdUCt P(FIE) P(E) .
We t}lerefore arrive at E--argmaX P(F ¦ E)P(E) E (2) ~n E~uati.on 2, the pro~ability P(E) of tlle t:arqet language text, F, is a l.anguage model matcll score allcl may ~e estim~te(l from a t:ar~et language model. Wllile ally knowll l.allguage mo~lel may )~e used to estimate the probability P(E) of t~le target lallguage text E, Brow et al descxibe an n-gram language mo~e] ~oml)risin~J a 1-<~ralll mo~le], a 2-gram model, and a 3-gram model combined by parameters whose values are obtainea by interpolated estimation.
The conditional probability P(FIE) in Equation 2 is a translation match score. As described by Brown et al, the translation match Yo993-089 - 2 -~ ~ ~o~ ~Q ~
score P(F I E) for a source text F comprising a series of source words, given a target hypothesis E
co""" i~in~ a series of target words, may be estim~ted by finding all possible ~li&n~el~s connecting the source words in the source text F with the target words in the target text E, inclu-line alignments in which one or more source words are not connected to any target words, but not including llllr~ where a source word is com~e-;led to more than one target word. For each alignment and each target word e in the target text E connected to ~ source words in the source text F, there is estimated the fertility probability n(~ I e) that the target word e is connected to the ~ source words in the alignment. There is also estim~ted for each source word f in the source text F and each target word e in the target text E connected to the source word f by the ~lignm~nt, the lexical probability t(f I e) that the source word f would occur given the occurrence of the connected target word e.
For each alignment and each source word f, Brown et al further estim~te the distortion probability a(j I apm) that the source word f is located in position j of the source text F, given that the target word e conne~;led to the source word f is located in position aj in the target text E, and given that there ar m words in the source text F.
By co",bining the fertility probabilities for an alignment and for all target words e in the target text E, and multiplying the result 212~2~30 by tlle probability n~(~O~ ) of the number ~O of target words not i =l connected Wit]l any source words in the alignment, given the sum of the fertilities ~ of all of the target words in the target text E in the alignment, a fertility score for the target text E and the alignment is obtained.
~y combilling the lexical prol~abilities for an alignmellt and for <~ ource wor~ls in tlle source t-ext F, a lexical score for the ~ lmellt: is ol)t.~
By coml)illill~ the dis î:ortioll prol)al~ilities for ~n alignmellt and for all source words i.J~ t,he ~ource text E whic}l are connected to a t.~rq~t: wor(l ill t:)le ali(J~ ellt~ y m~llt,il)lyi.ll(J the resu]t ~y jl, (wlleLe ./", is 1,1~e nllnlbel: of ~,~lr~let wor(ll; ill the t,arget text E
tllat are not conllected wit)l ally source wor~ls), a distortion score for the ali~llmellt isi obtaille~
Finally, by com~)illin~ the ferli.lit:y, lexic~l, alld distortio scores for the alignmellt, and multiplyin~ the result by the combinatori~l f~ctor ~ j!), a t:rallslat-ioll m.lt:ch score for tlle J
alignment is obtailled. (See, Brown et al, Section 8.2.) The translation match score P(FIE~ for the source text F and the target hypothesis E may be the sum of the translation match scores for all permitted alignments between the source text F and the target hypothesis E. Preferably, the translation match score 21~200 P(FIE) for the source text F and the target hypotllesis E is the translation match score for the alignment estimated to be most probable.
Equation 2 may be used to directly estimate the target hypot}lesis matc}l score P(F¦E)P(E) for a hypot}lesi7ed target language text E
alld a source lallguage text F. Ilowever, in order to simply the lallg-lage model ~(E) and tlle trallslation model P(EIE), and in order l:o estimate tlle l)arallleters of tllese models from a manageable ~moUIIt of traillillg ~ata, srowll et al estimate tlle tar~et l~ypotllesis m<ltcll scoLe P(FIE)~'(E) for sim~)lifiecl illterme~iate forms E' ~nd F' of tlle tal-get ]allgu~ge text E and the source lc~ xl ~ iv~ly. Ec-l~ll illt~rnl~ lte tc~rget~
word e' represe~lts a class, of related target language words. Eacl illtermediate source language word f' represents ~ class of related ~o~lr~e ]~llgUage worlls. ~ s~ul-ce lallg~la(Je t r.lllSd~l~'eL ~'OIIVert~ le source language text F t(~ t)le illterme(li~lte form F'. Tlle )Iypothesi7.ed intermediate form target lallguage text E' llavillg tlle highest hypothesis matc}l score P(E'¦E')P(E') is estimated from Equation 2. ~ target language transducer converts the best matched intermediate target lallguage text E'to tlle target lallguage text E.
In their language translation system, Brown et al estimate the lexical probability of each source word f as the conditional probability t(fle) of each source word f given solely the target 21~2~0 words e connected to the source word in an alignmellt.
Consequently, the lexical probability provides only a coarse estimate of the probability of the source word f.
summary of the Invelltioll It is al~ object of t}le invellt:iollto provide an apparatus and metllod for translatillg a series of source words in a first langu~ge to a serie. of l-al-(3et wor~ls in ~ sec o~ c~u~ e (li~ferent f rom t lle first language, in wllich tlle al)p~ratus alld met-~lod provide improved estimates of tlle lexical pLo~al~ilities of tlle source words.
It is anotller o~ject of the invelltioll to l)rovide an apparatus an(l metllod for translating a series of soul-ce words in a first l~ngllage to a series of target wor~s in a secoll~ langu~qe different from the first language, in which the lexical E)robability of a source word is estim~ e(l as a conditiotlal probability given the target word conllected to t~le source word in an alignment, and given the context of tlle target- word conllectecl to the source word.
According to the invention, an apparatus for translating a series of source words in a first language to a series of target words in a second language different from the first language comprises means for inputting a series of source words. Means are also 2125~00 provided for generating at least two target hypot}leses. Each target hypothesis comprises a series of target words selected from a vocabulary of words in the second language. Each target word has a context comprising at least one other word in the target hypothesis.
language model match score generator generates, for each target ]lyI)otl~esis, a lallgu~ge model matcll score comprising an estimate Or t:lle pro})cll)ility of~ OCCUrrell<e of the series of words in tlle t,arget hyl)otlle~is. An aligllmellt identifier identifies at least c~lle aligllmellt ~etweell the inl~nt series of source words and each target }~yl)ot})esis. Tl~e alignmellt collnects each source word with at, }e~l~it, olle tar~1et, wor(l ill tlle t-arget, llyl)o~-lle.sis.
word mat(}l score generator is l~rovided for generatillg, for each source word all~l eacll target llypotllesis, a word matcll score comprising an estimate of the conditiol~al l~ro~ability of occurrence of t}le source word, givell 1he t,ar~et word in the target llypotllesis whicll is conllected to tlle so~lrce word and given the context of the target word in the target hypothesis whicll is connected to the source word. A translatioll match score generator generates, for each target hypot}lesis, a translatiotl match score comprising a combination of the word match scores for the target hypothesis and the source words in the input series of source words.
212~2~
A hypothesis matc}l score generator is provide-l for generatillg a target hypothesis match score for each target hypothesis. Each target hypothesis match score comprises a combination of the language model match score for the target hypothesis and the translation match score for the target hypothesis. The target hypothesis havillg the best target hypothesis match score is provided at an output.
Prefera~ly each target hypotllesis comprises a series of target words selecte~ from ~ vo(nll~nl~ry comprisillg words in tlle ~con l~ngnclqe al~cl a lull wor(l repre;elltillg tlle ~sence of a word.
The alignmellt i(lentifier may coml)ri-;e meall.s for identifyil-g two or more alignmellts ~etween the input series of source wor~s and eac}l target hypot}lesis. Each alignmellt conllects each source word Wit~l at least one targel worc~ in t~le tar(~et llypotllesis. The worl match score generator generates for e~ch source word and eac}
alignment and each target hypot}lesis a word matc}l score comprising an estim~te of the conditional pro~a}~ility of occurrence of the source word given the target word connected to the source word and given the context of tlle target word. The translation match score generator generates for each target hypothesis a translation match score comprising a ~ombinatioll of the word match scores for the target hypothesis and the alignments and the source words in the input series of source words.
Yo993-089 - 8 -212~2~
Tlle source text il~put device may coml,rise mealls for tral~sformillq the input series of source words into a series of transformed source words. The alignment means identifies at least one alignment between the series of transformed source words and each target hypothesis. Each alignment connects each transformed source word with at least one target word in the target llypotllesis. The word match score generator generates for each transformed source wor~ nd each target hypot]lesis, a word matc~
score eomprisillg an estimate of tlle col~ditional pro~al~ility of occurrence of the trallsformed source word, givell the target wor~
in the target hypotllesis whic}l is conllected to the transforme~
wor~ and given the context of the target word in the target }~ypot}leiis whic}l is eolllleete~ 1o t}le trc~nsfornle~ souree word.
T}~e t l-c~llSl~tiOIl matcll score gellerator gellerates for each target llypot~lesis ~ t-rallsl~liol~ mat(ll ;cor~ coml)risil~g a coml~ atioll of the word match scores for the target hyl)ot-}lesis an(l t}le transformed source word. output means are provided for synthesi7ing a series of output words from the target }~ypot}lesis llaving the best target hypotllesis match score alld for OUtpUtti tl~e output wor~s.
The translation match score for a target hypothesis may comprise the product of the word match scores for the target hypotllesis and the source words in the input series of source words. The target hypothesis match score for a target hypothesis may comprise the Yo993-089 _ 9 -21252~
product of tlle language model match score for the target hypothesis multiplied by the translation match score for the target hypotllesis.
The context of the target word in tlle target hypothesis whicll is connected to a source word may be contained in at least one of two or more context classes. The estim~ted conditiollal probability of occurrence of the source word, ~iven tlle target word connected 1:O tll~ ';O~lrCe WC)Ld .Illd (1ivell t,he collt:ext, of t-lle target word, coml~rises ~t le~st one fnllcti,on h~villg ~ v~lue dependellt on t}le elass COIIt~illillg tlle cont:ext of the t~rget word.
Altel.ll~tiVe]y, the eStilll~lted ('Olldi t iOII~I] prol)ability of occurrence of t}le source word may ~onlprise a fun~t,ioll }l~vin9 a value depelldellt Oll the p~rt of speech in the target llypothesis of at le~st one word in t,}le context of tlle t,c~r~let, word W~liCll iS COIllleCt to tlle source word, or del)elldellt on tlle idellti,ty of at le~st one word in the context of tlle target word wllic~l is conllected to the ~ource word.
The me~ns for outputting tlle target hy~otllesi~ h~Vill9 the best match score may comprise a display. The input means may comprise a keyboard, a computer disk drive, or a computer tape drive.
By estimating the lexicaL probability of a source word as the conditional proba~ility of the source word given the target word 21252~û
conllected to tlle source word ill c~n aligllmellt, alld given tlle context of the target word connected to the source word, the present invention obtains translation match scores with improved accuracy.
Brief Description of the Drawinqs Figure 1 is a l~lock diagram of an example of an apparatus for trallslatillg words from a first language to a second language accordillg to tlle present inventioll.
Figure 2 scllematical.ly sl~ow.~ all ex.~ le of all aligllment betWeell a l~ypot)letical series of source words and a hyl)ot]~etical series of 1arget words.
Figure 3 scllematically s~ows an examl)le of a second alignmellt between t}le hypotlletical series of source words and the hypot}letical series of target words of Figure 2.
Figure 4 schematically sllows an example of a tllird aligllment between the hypothetical series of source words and the hypothetical series of target words of Figure 2.
Yo993-089 - 11 -~ 6 D
Desc~ ion ofthe Plerelled Embodiments Figure I is a block diagram of an eA~"p'e of an appalal~ls for tranQl~ting words from a first l~n~-~e to a second l~age accolding to the present invention. The apparatus co",p,ises a source text input device 10 for inputting a series of source words. The source text input device 10 may comprise, for example, a keyboard, a computer disk drive, or a computer tape drive.
The source text input device 10 may further comprise means for tran~ro"~ g the input series of source words into a series of ~,~nsro""ed source words. Each llansrolllled source word may ,~,~se,lt a set of related input source words. For example, each ofthe input source words in the set {etre, etant, ete, suis, es, est, sommes, etes, sont, fus, fumes, serai, serons, sois, soit, soyons, soyez, soient} are forms of the verb infinitive "etre", and may be ~ sr~""led to "etre" with a tag I epl ese"ling the tense of the original input source word.
The means for l~nsl~",ling the input series of source words into a series of llansr~Jlllled source words may col~p,ise the source tr~n~lucers desclil)ed in Sections 3, 4, and 11 of Brown et al, above.
In essence, these tr~nsd~lcers check and correct the spelling of the input source words, check and correct the case of the input source words, detect titles of documPnt~ in the input series of source , .
..
2 1 2 ~ 2 û O
words, and detect names in the input series of source words. The transducers also tag each input source word Wit]l the most likely part of speech of the source word, and flag unknown source words (which are not contained in a stored vocabulary of source words).
The means for transforming the input series of source words also collapses multi-word units of input source words into single transformed source words, and splits compound input source words into two or more transformed source words. Tlle means for tra~lsforming the input series of source words illtO a series of trallsformecl source worcls furtller ~)erforms linguistic or morpllological transformatiolls of differellt forms of a worcl illtO a sil~gle l~asic form. Finally, tlle mealls for trallsformillg tlle inl)ut series of source words may also estimcate tlle sense of each input so~lrce word, allclc assigl~ t:lnlt sense to t)le tr<lllc;formed sollrce word.
Tal)le 1 SIIOWS a llypotlletical exanll)l c? of all illpUt series of source words according to the invelltioll. In tllis example, the source words are Frenc}l words.
Input Series of Source Words, F
f, f2 f3 f4 fs f6 La clef est dans la porte Yo993-089 - 13 -yo9-93 -089 The translation apparatus according to the present invention further comprises a target hypothesis generator 12. The target hypothesis generator 12 generates at least two target hypotheses. Each target hypothesis comprises a series of target words selected from a vocabulary of words in the second language. The vocabulary of words in the second language may be stored in a target l~n~l~ge vocabulary store 14. Each target word in a target hypothesis has a context comprising at least one other word in the target hypothesis.
An example of a target hypothesis generator is described in Section 14 of Brown et al, cited above.
Table 2 shows a hypothetical example of target hypotheses El, E2, and E3. In this example, the target words are English language words.
Target Hypotheses Eh Eh eh~l eh,2 eh,3 eh~ll eh,S eh,6 El The key is in the door Ez The key is in the gate E3 The wrench is in the door ;- A-,~ r~
21252Q~
Preferably, each target hypotllesis comprises a series of target words selected from a vocabulary comprising words in the second language, and a null word representing tlle absence of a word. In Table 2, all of the target hypotheses Eh will be considered to include the "null" word.
Returning to Figure 1, the translatiol1 apparatus comprises a lallguage model match score generator 16 for generating, for eacl target 1~ypothesis, a lan~uage model match score comprising an estimate of tl-e pro~al)ility of occurrence of t}le series of words in t]~e tarc~et llypotl~esis. Se<tion~. 6 al1(1 7 of Brown et al, cite(1 above, describe an example of a language model match score generator~ Wl1ile ally known language model may }~e used to estimate l he pro})a}~i ] i ty of occurrence of t}le series of words in the target llypot11esis, Brown et al descri~e an n-qram langnage mo(1e] comprisillg ca 1-~lram model, a 2-gram model, and a 3--gram model com~ine<1 l~y parametersC. whose values are o~tained ~y interpolated estimation.
The translation apparatus furt11er comprises al1 aligllmel1t identifier 18 for identifying at least one alignmel1t ~etween tl1e input series of source words and each target hypot}1esis. Tl1e alignment connects each source word with at least one target word in the target hypothesis.
2125~
Figures 2, 3, and 4 schematically sllow examples of possi~le alignments between the hypothetical input series of source words in Table 1 and the hypothetical input series of target words in target hypothesis E~ in Table 2. In each alignment, each source word in the input series of source words F is connected with at least one target word in the target hypothesis E, by a solid line.
In tlle alignmellt of Figure 4, tlle second occurrellce of the source word "La" has no solid line therefrom, and is therefore considered to ~)e conllected to tl-e "null" word.
Ta~les 3, 4, and 5 provide alternate descriptions of the alignments of Figures 2, 3, and 4, respectively.
Alignmellt A,~, For llypot)lesis E,~
(h=1) and Source Words F
j fj aj eaj 1 La 1 The 2 clef 2 key 3 est 3 is 4 dans 4 in la 5 tB~
6 porte 6 door Yo993-089 - 16 -2125~1~0 Alignment Ah 2 For Hypothesis Eh (h=1) and Source Words F
j f; ai eAi 1 La 5 the 2 clef 2 key 3 est 3 is 4 dans 4 in la 1 Tl1e 6 porte 6 door TABI,E 5 Alignment A" 2 For l~ypothesis Eh (h=1) and Source Words F
i f~
1 La 1 The 2 clef 2 key 3 est 3 is 4 dal1s 4 in la 0 <null>
6 porte 6 door In each Table, for each parameter j ranging from 1 to m (w11ere m is the number of words in the input series of source words), there is another parameter aj having a single value in the range from 0 to l (where l is the number of words in the target hypothesis).
212 3 2 ~ ~
For a given alignment, each word fj in the input series of source words is connected to the word e~j in tlle tarqet llypot11esis.
In general, there are 2'm possible alignments between a series of m source words al1d a series of 1 noll-null target words, wllere eac)l source word may ~e connected to either the null source word or one or more non-null target words. If each source word is constrained to ~e conllected to only one null or nol1-null target word, then t~lere are m' ' possible alignmel1ts.
Preferably, only one alignmel1t between t~le input series of source words and each target hypot}lesis is identified for obtainil1g a word m~tcl1 score, descril)ed l)elow, for each source word. The one identified alignmellt for the series of source words and eacll target hypothesis is preferably that whic}1 is produced by t}le t<~rget hypot}lesis gener~tor, as descri~ed in section 14 of Brown et al, cited above~
If the source text input device 10 comprises means for transforming the input series of source words into a series of transformed source words, then the alignment identifier 18 identifies at least one alignment between the series of transformed source words and each target hypot}1esis. The alignment connects each transformed source word with at least one target word in the target hypothesis.
2 1 ~
.~eturning to Figure 1, tlle tratl.slatiol~ ~pparatu~ furtller comprise~, a word match score generator 20. The word match score generator 20 generates, for each source word and each target hypothesis, a word match score comprising an estimate of the conditional probability P(fle, X) of occurrence of the source word f, given the target word e in the target llypothesis whicll is connected to the source word f and given t}le context X of the target word e in the target hypotllesis wl-ich is conllected to the source word f.
Ta~le 6 illustrates a hypotl)etical example of the context X of ec~ target word e,~ tlle 1:arget llypotllesis E, of Table 2 for tl~e aligllmellt ~,, of Ta~le 3 witll tlre input series of source words of Tal)le 1.
Colltext X of Target Word eaj For ~ lllllCllt ~ I
fj eaj X =~e,Aj 3~ e(~j 2)1 e(aj-I)r e(Aj-~ e(Aj+2)~ e(~ ~ 3~) X~ x2 X3 X" X5 Xfi 1 La The <null> <null> <null> key is in 2 clef key <null> <null> The is in the 3 e~;t- i ~. <1~ll11 > Tlle key i 1~ e ~oor 4 dalls ill Tlle key i~; t l~e ~lool .
words, and detect names in the input series of source words. The transducers also tag each input source word Wit]l the most likely part of speech of the source word, and flag unknown source words (which are not contained in a stored vocabulary of source words).
The means for transforming the input series of source words also collapses multi-word units of input source words into single transformed source words, and splits compound input source words into two or more transformed source words. Tlle means for tra~lsforming the input series of source words illtO a series of trallsformecl source worcls furtller ~)erforms linguistic or morpllological transformatiolls of differellt forms of a worcl illtO a sil~gle l~asic form. Finally, tlle mealls for trallsformillg tlle inl)ut series of source words may also estimcate tlle sense of each input so~lrce word, allclc assigl~ t:lnlt sense to t)le tr<lllc;formed sollrce word.
Tal)le 1 SIIOWS a llypotlletical exanll)l c? of all illpUt series of source words according to the invelltioll. In tllis example, the source words are Frenc}l words.
Input Series of Source Words, F
f, f2 f3 f4 fs f6 La clef est dans la porte Yo993-089 - 13 -yo9-93 -089 The translation apparatus according to the present invention further comprises a target hypothesis generator 12. The target hypothesis generator 12 generates at least two target hypotheses. Each target hypothesis comprises a series of target words selected from a vocabulary of words in the second language. The vocabulary of words in the second language may be stored in a target l~n~l~ge vocabulary store 14. Each target word in a target hypothesis has a context comprising at least one other word in the target hypothesis.
An example of a target hypothesis generator is described in Section 14 of Brown et al, cited above.
Table 2 shows a hypothetical example of target hypotheses El, E2, and E3. In this example, the target words are English language words.
Target Hypotheses Eh Eh eh~l eh,2 eh,3 eh~ll eh,S eh,6 El The key is in the door Ez The key is in the gate E3 The wrench is in the door ;- A-,~ r~
21252Q~
Preferably, each target hypotllesis comprises a series of target words selected from a vocabulary comprising words in the second language, and a null word representing tlle absence of a word. In Table 2, all of the target hypotheses Eh will be considered to include the "null" word.
Returning to Figure 1, the translatiol1 apparatus comprises a lallguage model match score generator 16 for generating, for eacl target 1~ypothesis, a lan~uage model match score comprising an estimate of tl-e pro~al)ility of occurrence of t}le series of words in t]~e tarc~et llypotl~esis. Se<tion~. 6 al1(1 7 of Brown et al, cite(1 above, describe an example of a language model match score generator~ Wl1ile ally known language model may }~e used to estimate l he pro})a}~i ] i ty of occurrence of t}le series of words in the target llypot11esis, Brown et al descri~e an n-qram langnage mo(1e] comprisillg ca 1-~lram model, a 2-gram model, and a 3--gram model com~ine<1 l~y parametersC. whose values are o~tained ~y interpolated estimation.
The translation apparatus furt11er comprises al1 aligllmel1t identifier 18 for identifying at least one alignmel1t ~etween tl1e input series of source words and each target hypot}1esis. Tl1e alignment connects each source word with at least one target word in the target hypothesis.
2125~
Figures 2, 3, and 4 schematically sllow examples of possi~le alignments between the hypothetical input series of source words in Table 1 and the hypothetical input series of target words in target hypothesis E~ in Table 2. In each alignment, each source word in the input series of source words F is connected with at least one target word in the target hypothesis E, by a solid line.
In tlle alignmellt of Figure 4, tlle second occurrellce of the source word "La" has no solid line therefrom, and is therefore considered to ~)e conllected to tl-e "null" word.
Ta~les 3, 4, and 5 provide alternate descriptions of the alignments of Figures 2, 3, and 4, respectively.
Alignmellt A,~, For llypot)lesis E,~
(h=1) and Source Words F
j fj aj eaj 1 La 1 The 2 clef 2 key 3 est 3 is 4 dans 4 in la 5 tB~
6 porte 6 door Yo993-089 - 16 -2125~1~0 Alignment Ah 2 For Hypothesis Eh (h=1) and Source Words F
j f; ai eAi 1 La 5 the 2 clef 2 key 3 est 3 is 4 dans 4 in la 1 Tl1e 6 porte 6 door TABI,E 5 Alignment A" 2 For l~ypothesis Eh (h=1) and Source Words F
i f~
1 La 1 The 2 clef 2 key 3 est 3 is 4 dal1s 4 in la 0 <null>
6 porte 6 door In each Table, for each parameter j ranging from 1 to m (w11ere m is the number of words in the input series of source words), there is another parameter aj having a single value in the range from 0 to l (where l is the number of words in the target hypothesis).
212 3 2 ~ ~
For a given alignment, each word fj in the input series of source words is connected to the word e~j in tlle tarqet llypot11esis.
In general, there are 2'm possible alignments between a series of m source words al1d a series of 1 noll-null target words, wllere eac)l source word may ~e connected to either the null source word or one or more non-null target words. If each source word is constrained to ~e conllected to only one null or nol1-null target word, then t~lere are m' ' possible alignmel1ts.
Preferably, only one alignmel1t between t~le input series of source words and each target hypot}lesis is identified for obtainil1g a word m~tcl1 score, descril)ed l)elow, for each source word. The one identified alignmellt for the series of source words and eacll target hypothesis is preferably that whic}1 is produced by t}le t<~rget hypot}lesis gener~tor, as descri~ed in section 14 of Brown et al, cited above~
If the source text input device 10 comprises means for transforming the input series of source words into a series of transformed source words, then the alignment identifier 18 identifies at least one alignment between the series of transformed source words and each target hypot}1esis. The alignment connects each transformed source word with at least one target word in the target hypothesis.
2 1 ~
.~eturning to Figure 1, tlle tratl.slatiol~ ~pparatu~ furtller comprise~, a word match score generator 20. The word match score generator 20 generates, for each source word and each target hypothesis, a word match score comprising an estimate of the conditional probability P(fle, X) of occurrence of the source word f, given the target word e in the target llypothesis whicll is connected to the source word f and given t}le context X of the target word e in the target hypotllesis wl-ich is conllected to the source word f.
Ta~le 6 illustrates a hypotl)etical example of the context X of ec~ target word e,~ tlle 1:arget llypotllesis E, of Table 2 for tl~e aligllmellt ~,, of Ta~le 3 witll tlre input series of source words of Tal)le 1.
Colltext X of Target Word eaj For ~ lllllCllt ~ I
fj eaj X =~e,Aj 3~ e(~j 2)1 e(aj-I)r e(Aj-~ e(Aj+2)~ e(~ ~ 3~) X~ x2 X3 X" X5 Xfi 1 La The <null> <null> <null> key is in 2 clef key <null> <null> The is in the 3 e~;t- i ~. <1~ll11 > Tlle key i 1~ e ~oor 4 dalls ill Tlle key i~; t l~e ~lool .
5 la the key is in door . <null>
6 porte door is in the . <llull> Cnull>
As shown in Table 6, in this hypothetical example the context X
of a selected target word consists of tlle three target words preceding the selected target word and the three target words following the selected target word in the target hypothesis. The context also includes punctuation and absence of words.
In general, the context of the target word e~j in the target llypothesis E which is connected to a source word fj may be contained in at least one of two or more context classes. T}le estimated conditiollal proba~ility of occurrence of a source word, given the target wor~ in the target hypothesis whic}l is connected to the source word and given the context of the target word connected to the source word, may comprise at least one function having a value dependent on the class containing tlle context of the target word which is connected to the source word.
Alternatively, the context may comprise at least one word having a part of speech in the target hypothesis. The estimated conditional probability of occurrence of a source word, given the target word in the target hypothesis which iS connected to the source word and given the context of the target word which is connected to the source word, may comprise at least one function having a value dependent on the part of speech in the target hypothesis of at least one word in the context of the target word which is connected to the source word.
In another example, the context of the target word in the target hypothesis which is connected to a source word comprises at least r~
2 1 2 ~ 2 ~ O
~ne word having an identity. T}le estimated conditiollal probability of occurrence of the source word, given the target word in the target hypothesis whic]l is connected to the source word and given the context of the target word which is connected to the source word, comprises at least one function having a value dependent on the identity of at least one word in the context of tlle target word whic]l is conllected to t]~e source word.
Equations 3, 4, 5, and 6 are hypotl-etical examples of functions h~villg values depen~lellt on tlle context of the tarc~et worcl WlliC) is conllected to a source worcl.
g~(f,e = key,X)= 1, if f = "clef", and tlle word X3 in X immediate]y l)rece(lil-g "key" is "tl~e"; (~) = 0, otherwise.
g2(f,e = key,X)= 1, if f = "clef", and tlle word X3 in X immediately prece~ing "key" is "car"; (4) = 0, otherwise.
g3(f,e - key,X)= 1, if f = "ton", and tlle word X3 in X immediately preceding "key" is "the"; (5) = 0, otherwise.
2l232a(~ , g~(f,e = key,X)= 1, if f = "ton", and the word X4 in X immediately foLlowing "key", or the word x5 in X next following ~key" is (6) an element of the set {A, B, C, D, E, F, G};
= O, otherwise.
In Equation 3 the function g has a value of 1 if the source word f is "clef" if the target word e is "key" and if the word in the ~ontext X immediately preceding "key" is "the". If these conditions are not met the context function g has a value of 0.
T}le llypot}letical context functioll g2 of Equation 4 }las a value of 1 if the source word f is "clef" if the target word e is "key"
and if tlle word in the context X immedic~tely preceding "key" is "car". If these conditions are not met the function g2 llas a value of 0.
In Equation 5 the context functioll g3 llas a value of 1 if the source word f is "ton" if the target word e is "key" and if the word in the context X immediately preceding "key" is "the". If tllese conditions are not met t}le context functioll g3 has a value of 0.
Finally the hypothetical context function g~ in Equation 6 has a value of 1 if the source word f is "ton" the target word e is "key" and if the word in the context X immediately following "key" or the word in the context X next following "key" is an 212~2~
element of the set {A, B, C, D, ~, F, G~. If these conditions are not met, the context function g4 has a value of 0.
Table 7 illustrates the evaluation of context functions g(f,e,X) for the source word f="clef", for the target word e="key", and for the context X of the target word "key" in the target hypothesis E, of Table 2.
TA131,E: 7 C~ x ~ , i OII ~ ( r ~ e X = (e(ai - 3~ e~a; - 2)' e(aj f j eAj e(aj ~ I)' e(~j ~ 2)' e(~j + 3)) g(fi~ ea j~X) Xl x2 Xl X~ X5 X~ gl ~~2 ~11 q'l 2 cl ef key <null> <null> Tlle is i n tlle 1 0 0 0 ~s shown in Table 7, the context function g, has a value of 1, and the context functiolls 92, g3 an-l g~ have values of 0.
As discussed above, the word matcJI score for each source word alld each target hypothesis comprises an estimate of the conditional probability P(fle, X) of occurrence of the source word f given the target word e in the target hypothesis which is connected to the source word f and given the context X of the target word e in the target hypothesis which is connected to the source word f. The 2l2~Da word matcll score may l)e obtained, for example, using a model defined by Equation 7 as follows.
~e, i~qi (f, e, Xl P(fle, X)= N(e X) ei (7) In Equation 7, the functions gj(f,e,X) are functions having values (lel)elldellt oll tl~e colltexL x of t~le target word e wllicll is collllecte-l to the source word f in an al.ignmellt ~etween an input series of source words and a target hypot}lesis. The parameters i~e;) are parameters representing the relative strengtll of each context functioll g, in predi.cting the source word f from the target word e in the context X. Tlle quantity N(e,X) is a normalizati.on factor which depends on the target word e and the context X of the target word e, as showll in E(luatioll 8.
N(e,X)= ~ ei (8) For the target word e = "key" and for the hypotlletical context functions g, through g4 of Equations 3-6, above, Equation 9 is a Yo993-089 - 24 -2 1 2 ~
Jlypothetical model for generating word match scores for source words f.
(fle = key~ X)= N(el X) x et(e=~'Yl)9l~r~ 'yx)~l(e=~ey~2)q2~e=~ey~x)]
[I/e = Icey, 3)93(~ .e ' Icey, X) + Z(e = l~ey 4)q4(~ ,e = I:ey, X)]
For the purpose of Equatioll 9, the normalization N(e,X) is given by Equation 10.
~ ~I(e = ~ i(f.~ ,x) N(e = key,X)= ~ el (10) To illustrate the use of this model, hyl)ot}letical values for the model parameters are ~e= k~y 1) = ~ 12~ ~(e=~y 2) = ~ 34~ y 3) = ~09~ ~nd A(e=)ey 4) = .40 Table 8 illustrates the computation of the wor~ match score from Equations 7 and 9 for the source word "clef", the target word "key", and the context X of "key" in the target hypothesis E, of Table 2.
2125~
Computation of Word Match Score P("clef" I "keyn,X) X = {e(aj _ 3~ e~j _ 2~' e~nj ~
f j e~,j e~aj + 1), e~2,j + 2)' e~j + 3)) g(f j, eaj, X) Xl x2 X3 X,l X5 X6 gl 9Z g3 9'1 clc~ key <null><null>The is in tlle 1 0 0 0 ~oll k~y <nllll><nllll>TII~ is in ~h~ O 0 1 0 ~e = Jcey, ~) = . 12 A(e = l~ey, 2~ - 34 '7(~ e~, 3~ = ~ 09 A~e = J~ey, ~1) 40 ~I(e, j)qj(clef, ~ey, x~
e = 1.127 ~ (e, i~'li(ton, l~ey, xi e = 1.094 ~I(e ~ e~ y, X) ~ o~ , X) N(~ = key,X) = e + ~ = 2.221 r("cl ef n I "key" ~ X) = O . 507 Ill this hypotlletical example, the conditiol1al probability P("clef"l"key", X) of occurrence of the source word "clef" given the target word "key" in the target hyl-othesis which is conllected to the source word "clef" and given the context X of the target word "key" in the target hypotl1esis whicll is connected to the source word "clef" is equal to 0.507 (from Equatiol1s 9 and 10).
If the alignment identifier 18 identifies two or more alignments between the input series of source words and each target 21252~0 llypot)lesis, the word match score generator generates a word match score for each source word and each alignment and each target hypothesis. Each word match score may be estimated using the model of Equation 7.
If the source text input device 10 comprises means for transforming the input series of source words into a series of transformed source words, thell the word match score generator generates a word match score for each transformed source word and eac}l tarqet hypot}lesis. The word matc}l score comprises an estimate of conditiollal prol~a~ility of occurrence of the transformed source word, given the target word in the target }IypOt}leSiS W}liC}I iS conllected to the transformed source word and given the context of the target word in the target hypot}lesis whic}l is connected to the transformed source word. The word match score for a transformed word may also })e estimated usillg the model of Equation 7.
T}le translatioll apparatus according to t}le illventioll furt}ler comprises a translation match score generator 22. The translatio match score generator 22 generates, for each target hypotllesis, a translation match score comprising a com~ination of the word match scores for the target hypot}lesis and the source word~s in the illpUt series of source words. The translation match score for a target hypothesis may, for example, comprise the product of the word Yo993-089 - 27 -212S~30 matcll scores foL tl~e tarc~et ~Iypotllesis alld t~le source wor~s in tlle input series of source words.
Table 9 illustrates the computation of a translation match score for the input series of source words fj of Table 1, and for the target words eAj in the target hypothesis E, of Table 2.
Translation Matcll Score: Alignment =
j fj aj eaj P(fjle~j, X) 1 La 1 The 0.91 2 clef 2 key 0.507 3 est 3 is 0.87 4 dans 4 in 0.63 la 5 the 0.91 6 porte 6 door 0.79 Translation M~tch Score = IIP(fjle~j, X)= 0.182 (to illustrate the invention, this example assumes the fertility and distortion scores, and the combinatorial factor are all equal to 1) Each word matcll score P(fjleaj, X) is obtained from Equation 7. The numbers in Table 9 are hypotlletical numbers. Por the purpose of illustrating the present invention, the fertility and distortion scores of Brown et al, described above, are assumed equal to one.
Yo993-o89 - 28 -2~2~200 Returning to Figure 1, the translatioll apparatus comprises a hypothesis match score generator 24 for generating a target hypothesis match score for each target hypothesis. Each target hypothesis match score comprises a combination (for example, the product) of the language model match score for the target hypothesis and the translation match score for the target hypothesis. As discussed above, the language model match score may be obtained from known language models such as the n-gram ]anguage models described by Brown et al.
If the alignment identifier 18 identifies two or more alignments between the input series of source words and each target hypothesis, the translation matc}l score for each target hypothesis comprises a combination of the word match scores for the target hypothesis and the alignments and the source words in the input series of source words.
If the source text input device 10 comprises means for transforming the input series of source words into a series of transformed source words, the translation match score comprises a combination of the word match scores for the target hypothesis and the transformed source words.
The translation apparatus according to the invention further comprises an output 26. The output 26 outputs the target 2~2~2~0 hypothesis having the best target hypothesis match score. The output 26 may comprise, for example, a display or a printer.
If the source text input device 10 comprises means for transforming the input series of source words into a series of transformed source words, then the output 26 comprises means for synthesizing a series of output words from the target hypothesis llaving the best target hypothesis match score. The means for synthesizing output words from the target hypotllesis may comprise target transducers such as those described at Section 5 of Brown et al (cited above)~
For example, the target word "be" with a tag representing the tense of the original input source word, may be transformed to one of the synthesized output words {be, was, were, been, am, are, is, being} which are forms of the verb infinitive "ben.
As described above, Equation 7 is an example of a model which may be used to obtain word match scores according to the present invention. The context functions gi(f,e,X) of the word match score model, and the parameters A~ei~, of the word match score model may be obtained as follows.
Candidate context functions gl(f,e,X) may be obtained, for example, by limiting the context X of the target word in the target hypothesis to three words to the left of the target word e and 2~52i~3 three words to the right of the target word e, as shown in the example of Table 6, above.
Next, a training text of source language sentences and target language sentences which are translations of each other is obtained. Corresponding source language sentences and target lallguage sentellces whic)l are translations of eacll other may be identified, for example, ~y a skilled translator. Preferably, correspolldillg source langllage sentellces and target language ~entellces wllicll are translatio]ls of eac~l otller may be i~entified alltomatically, for example, ~y tlle method described in Sections 12 and 13 of Brown et al (cited above), For each ~air of correspondillg source alld target sentences in the training text, the estimated most prol~able aliqnment between tl~e source words and the target words is found using the method described above used by the alignment identifier 18. Each occurrence of a target word eaj in the trainillg text is then identified and tagged with the source word fj connected to the target word eaj in each alignment. The target word eaj is also tagged with its context X in each target sentence.
Table 10 illustrates a hypothetical example of training events for a target word e~j = nkey" in a training text of aligned source language sentences and target language sentences.
Yo993-089 - 31 -21252~
TRAINING EYENTS
Occurrences of e~="key" in a training text of aligned source and target sentences, tlle source word fj connected to "key" in each alignment, and the context X of "key" in each tar~et sentence.
fj eaj X =(e(aj 3~ e(p,j_2~ e~ e~ "~ e~aj{2)~ e(~j+3)) Xl x2 X3 X" X5 x6 clef key <null> My car is on the clef key make me a for my door ton key she sings ill <null> <null> <null>
clef key <null> <null> Tlle unlocks t]le safe ton key play in the of B flat tOII key finger on the of middle C
Using tlle training events for a target word eaj from the trainillg text, candidate context functions gj(f,e,X) may be obtained by first identifying types of context functions. For example, one type of context functioll tests for tlle presence of a particular word in one or more positions in the context. Anot}ler type of context function tests for the presence of a particular word class (for example, part of speech) in one or more positions in the context.
The particular words or word classes to be tested for in a context function for a target word e can be obtained from the words and classes in the context of the target word e in the training text.
Yo993-089 - 32 -Alternatively, candidate context functions gj(f,e,X) may be obtained by clustering the training events according to their context X using the method described in Section 7 of Brown et al (cited above), Initially, all of the parameters A(ej~ of the word match score model of Equation 7 are set equal to ~.ero.
For each calldi.date context functioll gi(f,e,X), a "measure of merit"
G(i) is calculated to according to Equatioll 11.
G(i)= (Egj(f,e,X)- Egj(f,e,X)) (11 E(gj(f,e,X)- E(~j(f,e,X)¦X)) where Egl(f,e,X)- ~P(flX)P(X)gi(f,X) (1~) Yo993-OB9 - 33 -~12$~0~
Egj(f,e,X)= ~ P(f,X)gj(f,X) (13) E(gj(f,e,X)¦X)= ~P(f ¦ X)P(X)(~'P(f ¦ X)~ j(f, X)) ( 1 4) r,x In Equations 11 through 14, the conditional probability P(flX) of a source word f given the context X of the connected target word e is o~tained from Equatioll 7 using the most recent values of the model parameters. The probability P(X) of a context X is obtained l)y counting occurrences of the target word e and the context X in the trainil1g text events of the type showll in Table 10, al1d dividing the COUI1t by the total number of events in the training text in which the target word is e. The probability P(f,X) of a source word f and a context X are o~tained by counting occurrences of the target word e, the context X, and the source word f in the training text events of the ty~e showll in Ta~le 10, and dividing each count by the total number of events in the training text in which the target word is e.
The context function gj(f,e,X) having the highest "measure of merit"
G(i) from Equation 11 is selected as a context function to be used Yo993-089 - 34 -in E~uation 7. The parameter l(ej, is obtained by initially setting l(ei~= O, and then solving the following Equation 15 for the quantity Al(,i~.
~ ( )P(flX)gj(f~e~x)e[Al(e~ it ~,X f,X
~ new value of i~e;) is o~tailled l~y adding tlle value of Ai(e j) to t]~e previous value ~f A(e ;). Usillg tlle new value of A(e ;)~ Equatioll 15 is then again solved for a new value ~f ~(e ;). The process is repeated, each time updatillg the value of A(e i)~ until the value of ~A(ej~ falls below a selected tllresllold. T}lis metllod is known as iterative scaling.
Using the new model for the word match score (Equatioll 7), the "measures of merit" G(i) of Equation 11 are recomputed for tlle remaining candid~te context functions gi(f,e,X) to identify the remaining context function having the highest "measure of merit".
The best remaining context function is added to the word match score model of Equation 7 and new values of all of the parameters l,el) are calculated using the iterative scaling method and Equation 15. When the word match score model of Equation 7 contains two or more parameters l(el~ every parameter A~ei~ is updated exactly once per iteration, so all parameters l~e;) converge in the same Yo993-089 - 35 -~l2s2ao iteration. The process is repeate on the remainillg candidate context functions gj(f,e,X) until the "measure of merit" of the best context function falls below a selected threshold.
In the translation apparatus according to the present invention, the target hypothesis generator 12, the language model match score generator 16, the alignment identifier 18, the word match score generator 20, the translation match score generator 22, and the l~ypothesis matcll score generator 24 may be suitably programmed general purpose or special purpose digital signal processors. The target language vocabulary store 14 may be computer storage, suc}~
as random access memory. The means for transforming the input series of source words into a series of transformed source words in the source text input device 10, and the means for syntllesizil1g a series of output words from the target llypot}1esis llaving the best target hypothesis match score of the output 26 may also be suita~ly programme~ general purl)ose or special l)urpose digita1 signal processors.
As shown in Table 6, in this hypothetical example the context X
of a selected target word consists of tlle three target words preceding the selected target word and the three target words following the selected target word in the target hypothesis. The context also includes punctuation and absence of words.
In general, the context of the target word e~j in the target llypothesis E which is connected to a source word fj may be contained in at least one of two or more context classes. T}le estimated conditiollal proba~ility of occurrence of a source word, given the target wor~ in the target hypothesis whic}l is connected to the source word and given the context of the target word connected to the source word, may comprise at least one function having a value dependent on the class containing tlle context of the target word which is connected to the source word.
Alternatively, the context may comprise at least one word having a part of speech in the target hypothesis. The estimated conditional probability of occurrence of a source word, given the target word in the target hypothesis which iS connected to the source word and given the context of the target word which is connected to the source word, may comprise at least one function having a value dependent on the part of speech in the target hypothesis of at least one word in the context of the target word which is connected to the source word.
In another example, the context of the target word in the target hypothesis which is connected to a source word comprises at least r~
2 1 2 ~ 2 ~ O
~ne word having an identity. T}le estimated conditiollal probability of occurrence of the source word, given the target word in the target hypothesis whic]l is connected to the source word and given the context of the target word which is connected to the source word, comprises at least one function having a value dependent on the identity of at least one word in the context of tlle target word whic]l is conllected to t]~e source word.
Equations 3, 4, 5, and 6 are hypotl-etical examples of functions h~villg values depen~lellt on tlle context of the tarc~et worcl WlliC) is conllected to a source worcl.
g~(f,e = key,X)= 1, if f = "clef", and tlle word X3 in X immediate]y l)rece(lil-g "key" is "tl~e"; (~) = 0, otherwise.
g2(f,e = key,X)= 1, if f = "clef", and tlle word X3 in X immediately prece~ing "key" is "car"; (4) = 0, otherwise.
g3(f,e - key,X)= 1, if f = "ton", and tlle word X3 in X immediately preceding "key" is "the"; (5) = 0, otherwise.
2l232a(~ , g~(f,e = key,X)= 1, if f = "ton", and the word X4 in X immediately foLlowing "key", or the word x5 in X next following ~key" is (6) an element of the set {A, B, C, D, E, F, G};
= O, otherwise.
In Equation 3 the function g has a value of 1 if the source word f is "clef" if the target word e is "key" and if the word in the ~ontext X immediately preceding "key" is "the". If these conditions are not met the context function g has a value of 0.
T}le llypot}letical context functioll g2 of Equation 4 }las a value of 1 if the source word f is "clef" if the target word e is "key"
and if tlle word in the context X immedic~tely preceding "key" is "car". If these conditions are not met the function g2 llas a value of 0.
In Equation 5 the context functioll g3 llas a value of 1 if the source word f is "ton" if the target word e is "key" and if the word in the context X immediately preceding "key" is "the". If tllese conditions are not met t}le context functioll g3 has a value of 0.
Finally the hypothetical context function g~ in Equation 6 has a value of 1 if the source word f is "ton" the target word e is "key" and if the word in the context X immediately following "key" or the word in the context X next following "key" is an 212~2~
element of the set {A, B, C, D, ~, F, G~. If these conditions are not met, the context function g4 has a value of 0.
Table 7 illustrates the evaluation of context functions g(f,e,X) for the source word f="clef", for the target word e="key", and for the context X of the target word "key" in the target hypothesis E, of Table 2.
TA131,E: 7 C~ x ~ , i OII ~ ( r ~ e X = (e(ai - 3~ e~a; - 2)' e(aj f j eAj e(aj ~ I)' e(~j ~ 2)' e(~j + 3)) g(fi~ ea j~X) Xl x2 Xl X~ X5 X~ gl ~~2 ~11 q'l 2 cl ef key <null> <null> Tlle is i n tlle 1 0 0 0 ~s shown in Table 7, the context function g, has a value of 1, and the context functiolls 92, g3 an-l g~ have values of 0.
As discussed above, the word matcJI score for each source word alld each target hypothesis comprises an estimate of the conditional probability P(fle, X) of occurrence of the source word f given the target word e in the target hypothesis which is connected to the source word f and given the context X of the target word e in the target hypothesis which is connected to the source word f. The 2l2~Da word matcll score may l)e obtained, for example, using a model defined by Equation 7 as follows.
~e, i~qi (f, e, Xl P(fle, X)= N(e X) ei (7) In Equation 7, the functions gj(f,e,X) are functions having values (lel)elldellt oll tl~e colltexL x of t~le target word e wllicll is collllecte-l to the source word f in an al.ignmellt ~etween an input series of source words and a target hypot}lesis. The parameters i~e;) are parameters representing the relative strengtll of each context functioll g, in predi.cting the source word f from the target word e in the context X. Tlle quantity N(e,X) is a normalizati.on factor which depends on the target word e and the context X of the target word e, as showll in E(luatioll 8.
N(e,X)= ~ ei (8) For the target word e = "key" and for the hypotlletical context functions g, through g4 of Equations 3-6, above, Equation 9 is a Yo993-089 - 24 -2 1 2 ~
Jlypothetical model for generating word match scores for source words f.
(fle = key~ X)= N(el X) x et(e=~'Yl)9l~r~ 'yx)~l(e=~ey~2)q2~e=~ey~x)]
[I/e = Icey, 3)93(~ .e ' Icey, X) + Z(e = l~ey 4)q4(~ ,e = I:ey, X)]
For the purpose of Equatioll 9, the normalization N(e,X) is given by Equation 10.
~ ~I(e = ~ i(f.~ ,x) N(e = key,X)= ~ el (10) To illustrate the use of this model, hyl)ot}letical values for the model parameters are ~e= k~y 1) = ~ 12~ ~(e=~y 2) = ~ 34~ y 3) = ~09~ ~nd A(e=)ey 4) = .40 Table 8 illustrates the computation of the wor~ match score from Equations 7 and 9 for the source word "clef", the target word "key", and the context X of "key" in the target hypothesis E, of Table 2.
2125~
Computation of Word Match Score P("clef" I "keyn,X) X = {e(aj _ 3~ e~j _ 2~' e~nj ~
f j e~,j e~aj + 1), e~2,j + 2)' e~j + 3)) g(f j, eaj, X) Xl x2 X3 X,l X5 X6 gl 9Z g3 9'1 clc~ key <null><null>The is in tlle 1 0 0 0 ~oll k~y <nllll><nllll>TII~ is in ~h~ O 0 1 0 ~e = Jcey, ~) = . 12 A(e = l~ey, 2~ - 34 '7(~ e~, 3~ = ~ 09 A~e = J~ey, ~1) 40 ~I(e, j)qj(clef, ~ey, x~
e = 1.127 ~ (e, i~'li(ton, l~ey, xi e = 1.094 ~I(e ~ e~ y, X) ~ o~ , X) N(~ = key,X) = e + ~ = 2.221 r("cl ef n I "key" ~ X) = O . 507 Ill this hypotlletical example, the conditiol1al probability P("clef"l"key", X) of occurrence of the source word "clef" given the target word "key" in the target hyl-othesis which is conllected to the source word "clef" and given the context X of the target word "key" in the target hypotl1esis whicll is connected to the source word "clef" is equal to 0.507 (from Equatiol1s 9 and 10).
If the alignment identifier 18 identifies two or more alignments between the input series of source words and each target 21252~0 llypot)lesis, the word match score generator generates a word match score for each source word and each alignment and each target hypothesis. Each word match score may be estimated using the model of Equation 7.
If the source text input device 10 comprises means for transforming the input series of source words into a series of transformed source words, thell the word match score generator generates a word match score for each transformed source word and eac}l tarqet hypot}lesis. The word matc}l score comprises an estimate of conditiollal prol~a~ility of occurrence of the transformed source word, given the target word in the target }IypOt}leSiS W}liC}I iS conllected to the transformed source word and given the context of the target word in the target hypot}lesis whic}l is connected to the transformed source word. The word match score for a transformed word may also })e estimated usillg the model of Equation 7.
T}le translatioll apparatus according to t}le illventioll furt}ler comprises a translation match score generator 22. The translatio match score generator 22 generates, for each target hypotllesis, a translation match score comprising a com~ination of the word match scores for the target hypot}lesis and the source word~s in the illpUt series of source words. The translation match score for a target hypothesis may, for example, comprise the product of the word Yo993-089 - 27 -212S~30 matcll scores foL tl~e tarc~et ~Iypotllesis alld t~le source wor~s in tlle input series of source words.
Table 9 illustrates the computation of a translation match score for the input series of source words fj of Table 1, and for the target words eAj in the target hypothesis E, of Table 2.
Translation Matcll Score: Alignment =
j fj aj eaj P(fjle~j, X) 1 La 1 The 0.91 2 clef 2 key 0.507 3 est 3 is 0.87 4 dans 4 in 0.63 la 5 the 0.91 6 porte 6 door 0.79 Translation M~tch Score = IIP(fjle~j, X)= 0.182 (to illustrate the invention, this example assumes the fertility and distortion scores, and the combinatorial factor are all equal to 1) Each word matcll score P(fjleaj, X) is obtained from Equation 7. The numbers in Table 9 are hypotlletical numbers. Por the purpose of illustrating the present invention, the fertility and distortion scores of Brown et al, described above, are assumed equal to one.
Yo993-o89 - 28 -2~2~200 Returning to Figure 1, the translatioll apparatus comprises a hypothesis match score generator 24 for generating a target hypothesis match score for each target hypothesis. Each target hypothesis match score comprises a combination (for example, the product) of the language model match score for the target hypothesis and the translation match score for the target hypothesis. As discussed above, the language model match score may be obtained from known language models such as the n-gram ]anguage models described by Brown et al.
If the alignment identifier 18 identifies two or more alignments between the input series of source words and each target hypothesis, the translation matc}l score for each target hypothesis comprises a combination of the word match scores for the target hypothesis and the alignments and the source words in the input series of source words.
If the source text input device 10 comprises means for transforming the input series of source words into a series of transformed source words, the translation match score comprises a combination of the word match scores for the target hypothesis and the transformed source words.
The translation apparatus according to the invention further comprises an output 26. The output 26 outputs the target 2~2~2~0 hypothesis having the best target hypothesis match score. The output 26 may comprise, for example, a display or a printer.
If the source text input device 10 comprises means for transforming the input series of source words into a series of transformed source words, then the output 26 comprises means for synthesizing a series of output words from the target hypothesis llaving the best target hypothesis match score. The means for synthesizing output words from the target hypotllesis may comprise target transducers such as those described at Section 5 of Brown et al (cited above)~
For example, the target word "be" with a tag representing the tense of the original input source word, may be transformed to one of the synthesized output words {be, was, were, been, am, are, is, being} which are forms of the verb infinitive "ben.
As described above, Equation 7 is an example of a model which may be used to obtain word match scores according to the present invention. The context functions gi(f,e,X) of the word match score model, and the parameters A~ei~, of the word match score model may be obtained as follows.
Candidate context functions gl(f,e,X) may be obtained, for example, by limiting the context X of the target word in the target hypothesis to three words to the left of the target word e and 2~52i~3 three words to the right of the target word e, as shown in the example of Table 6, above.
Next, a training text of source language sentences and target language sentences which are translations of each other is obtained. Corresponding source language sentences and target lallguage sentellces whic)l are translations of eacll other may be identified, for example, ~y a skilled translator. Preferably, correspolldillg source langllage sentellces and target language ~entellces wllicll are translatio]ls of eac~l otller may be i~entified alltomatically, for example, ~y tlle method described in Sections 12 and 13 of Brown et al (cited above), For each ~air of correspondillg source alld target sentences in the training text, the estimated most prol~able aliqnment between tl~e source words and the target words is found using the method described above used by the alignment identifier 18. Each occurrence of a target word eaj in the trainillg text is then identified and tagged with the source word fj connected to the target word eaj in each alignment. The target word eaj is also tagged with its context X in each target sentence.
Table 10 illustrates a hypothetical example of training events for a target word e~j = nkey" in a training text of aligned source language sentences and target language sentences.
Yo993-089 - 31 -21252~
TRAINING EYENTS
Occurrences of e~="key" in a training text of aligned source and target sentences, tlle source word fj connected to "key" in each alignment, and the context X of "key" in each tar~et sentence.
fj eaj X =(e(aj 3~ e(p,j_2~ e~ e~ "~ e~aj{2)~ e(~j+3)) Xl x2 X3 X" X5 x6 clef key <null> My car is on the clef key make me a for my door ton key she sings ill <null> <null> <null>
clef key <null> <null> Tlle unlocks t]le safe ton key play in the of B flat tOII key finger on the of middle C
Using tlle training events for a target word eaj from the trainillg text, candidate context functions gj(f,e,X) may be obtained by first identifying types of context functions. For example, one type of context functioll tests for tlle presence of a particular word in one or more positions in the context. Anot}ler type of context function tests for the presence of a particular word class (for example, part of speech) in one or more positions in the context.
The particular words or word classes to be tested for in a context function for a target word e can be obtained from the words and classes in the context of the target word e in the training text.
Yo993-089 - 32 -Alternatively, candidate context functions gj(f,e,X) may be obtained by clustering the training events according to their context X using the method described in Section 7 of Brown et al (cited above), Initially, all of the parameters A(ej~ of the word match score model of Equation 7 are set equal to ~.ero.
For each calldi.date context functioll gi(f,e,X), a "measure of merit"
G(i) is calculated to according to Equatioll 11.
G(i)= (Egj(f,e,X)- Egj(f,e,X)) (11 E(gj(f,e,X)- E(~j(f,e,X)¦X)) where Egl(f,e,X)- ~P(flX)P(X)gi(f,X) (1~) Yo993-OB9 - 33 -~12$~0~
Egj(f,e,X)= ~ P(f,X)gj(f,X) (13) E(gj(f,e,X)¦X)= ~P(f ¦ X)P(X)(~'P(f ¦ X)~ j(f, X)) ( 1 4) r,x In Equations 11 through 14, the conditional probability P(flX) of a source word f given the context X of the connected target word e is o~tained from Equatioll 7 using the most recent values of the model parameters. The probability P(X) of a context X is obtained l)y counting occurrences of the target word e and the context X in the trainil1g text events of the type showll in Table 10, al1d dividing the COUI1t by the total number of events in the training text in which the target word is e. The probability P(f,X) of a source word f and a context X are o~tained by counting occurrences of the target word e, the context X, and the source word f in the training text events of the ty~e showll in Ta~le 10, and dividing each count by the total number of events in the training text in which the target word is e.
The context function gj(f,e,X) having the highest "measure of merit"
G(i) from Equation 11 is selected as a context function to be used Yo993-089 - 34 -in E~uation 7. The parameter l(ej, is obtained by initially setting l(ei~= O, and then solving the following Equation 15 for the quantity Al(,i~.
~ ( )P(flX)gj(f~e~x)e[Al(e~ it ~,X f,X
~ new value of i~e;) is o~tailled l~y adding tlle value of Ai(e j) to t]~e previous value ~f A(e ;). Usillg tlle new value of A(e ;)~ Equatioll 15 is then again solved for a new value ~f ~(e ;). The process is repeated, each time updatillg the value of A(e i)~ until the value of ~A(ej~ falls below a selected tllresllold. T}lis metllod is known as iterative scaling.
Using the new model for the word match score (Equatioll 7), the "measures of merit" G(i) of Equation 11 are recomputed for tlle remaining candid~te context functions gi(f,e,X) to identify the remaining context function having the highest "measure of merit".
The best remaining context function is added to the word match score model of Equation 7 and new values of all of the parameters l,el) are calculated using the iterative scaling method and Equation 15. When the word match score model of Equation 7 contains two or more parameters l(el~ every parameter A~ei~ is updated exactly once per iteration, so all parameters l~e;) converge in the same Yo993-089 - 35 -~l2s2ao iteration. The process is repeate on the remainillg candidate context functions gj(f,e,X) until the "measure of merit" of the best context function falls below a selected threshold.
In the translation apparatus according to the present invention, the target hypothesis generator 12, the language model match score generator 16, the alignment identifier 18, the word match score generator 20, the translation match score generator 22, and the l~ypothesis matcll score generator 24 may be suitably programmed general purpose or special purpose digital signal processors. The target language vocabulary store 14 may be computer storage, suc}~
as random access memory. The means for transforming the input series of source words into a series of transformed source words in the source text input device 10, and the means for syntllesizil1g a series of output words from the target llypot}1esis llaving the best target hypothesis match score of the output 26 may also be suita~ly programme~ general purl)ose or special l)urpose digita1 signal processors.
Claims (21)
1. An apparatus for translating a series of source words in a first language to a series of target words in a second language different from the first language, said apparatus comprising:
means for inputting said series of source words;
means for generating at least two target hypotheses, each target hypothesis comprising said series of target words selected from a vocabulary of words in the second language, each target word having a context comprising at least one other word in the target hypothesis;
means for generating, for each target hypothesis, a language model match score comprising an estimate of the probability of occurrence of the series of words in the target hypothesis;
means for identifying at least one alignment between the input series of source words and each target hypothesis, the alignment connecting each source word with at least one target word in the target hypothesis;
means for generating, for each source word and each target hypothesis, a word match score comprising an estimate of the conditional probability of occurrence of the source word, given the target word in the target hypothesis which is connected to the source word and given the context of the target word in the target hypothesis which is connected to the source word;
means for generating, for each target hypothesis, a translation match score comprising a combination of the word match scores for the target hypothesis and the source words in the input series of source words;
means for generating a target hypothesis match score for each target hypothesis, each target hypothesis match score comprising a combination of the language model match score for the target hypothesis and the translation match score for the target hypothesis; and means for outputting the target hypothesis having the best target hypothesis match score.
means for inputting said series of source words;
means for generating at least two target hypotheses, each target hypothesis comprising said series of target words selected from a vocabulary of words in the second language, each target word having a context comprising at least one other word in the target hypothesis;
means for generating, for each target hypothesis, a language model match score comprising an estimate of the probability of occurrence of the series of words in the target hypothesis;
means for identifying at least one alignment between the input series of source words and each target hypothesis, the alignment connecting each source word with at least one target word in the target hypothesis;
means for generating, for each source word and each target hypothesis, a word match score comprising an estimate of the conditional probability of occurrence of the source word, given the target word in the target hypothesis which is connected to the source word and given the context of the target word in the target hypothesis which is connected to the source word;
means for generating, for each target hypothesis, a translation match score comprising a combination of the word match scores for the target hypothesis and the source words in the input series of source words;
means for generating a target hypothesis match score for each target hypothesis, each target hypothesis match score comprising a combination of the language model match score for the target hypothesis and the translation match score for the target hypothesis; and means for outputting the target hypothesis having the best target hypothesis match score.
2. An apparatus as claimed in claim 1, characterized in that each target hypothesis comprises a series of target words selected from a vocabulary comprising words in the second language and a null word representing the absence of a word.
3. An apparatus as claimed in claim 2, characterized in that:
the means for identifying at least one alignment comprises means for identifying two or more alignments between the input series of source words and each target hypothesis, each alignment connecting each source word with at least one target word in the target hypothesis;
the word match score generator comprises means for generating, for each source word and each alignment and each target hypothesis, a word match score comprising an estimate of the conditional probability of occurrence of the source word, given the target word in the target hypothesis which is connected to the source word and given the context of the target word in the target hypothesis which is connected to the source word;
the translation match score generator comprises means for generating, for each target hypothesis, a translation match score comprising a combination of the word match scores for the target hypothesis and the alignments and the source words in the input series of source words.
the means for identifying at least one alignment comprises means for identifying two or more alignments between the input series of source words and each target hypothesis, each alignment connecting each source word with at least one target word in the target hypothesis;
the word match score generator comprises means for generating, for each source word and each alignment and each target hypothesis, a word match score comprising an estimate of the conditional probability of occurrence of the source word, given the target word in the target hypothesis which is connected to the source word and given the context of the target word in the target hypothesis which is connected to the source word;
the translation match score generator comprises means for generating, for each target hypothesis, a translation match score comprising a combination of the word match scores for the target hypothesis and the alignments and the source words in the input series of source words.
4. An apparatus as claimed in claim 2, characterized in that:
the input means comprises means for transforming the input series of source words into a series of transformed source words;
the alignment means comprises means for identifying at least one alignment between the series of transformed source words and each target hypothesis, the alignment connecting each transformed source word with at least one target word in the target hypothesis;
the word match score generator comprises means for generating, for each transformed source word and each target hypothesis, a word match score comprising an estimate of the conditional probability of occurrence of the transformed source word, given the target word in the target hypothesis which is connected to the transformed source word and given the context of the target word in the target hypothesis which is connected to the transformed source word;the translation match score generator comprises means for generating for each target hypothesis, a translation match score comprising a combination of the word match scores for the target hypothesis and the transformed source words; and the output means comprises means for synthesizing a series of output words from the target hypothesis having the best target hypothesis match score, and means for outputting the synthesized output words.
the input means comprises means for transforming the input series of source words into a series of transformed source words;
the alignment means comprises means for identifying at least one alignment between the series of transformed source words and each target hypothesis, the alignment connecting each transformed source word with at least one target word in the target hypothesis;
the word match score generator comprises means for generating, for each transformed source word and each target hypothesis, a word match score comprising an estimate of the conditional probability of occurrence of the transformed source word, given the target word in the target hypothesis which is connected to the transformed source word and given the context of the target word in the target hypothesis which is connected to the transformed source word;the translation match score generator comprises means for generating for each target hypothesis, a translation match score comprising a combination of the word match scores for the target hypothesis and the transformed source words; and the output means comprises means for synthesizing a series of output words from the target hypothesis having the best target hypothesis match score, and means for outputting the synthesized output words.
5. An apparatus as claimed in claim 2, characterized in that:
the translation match score for a target hypothesis comprises the product of the word match scores for the target hypothesis and the source words in the input series of source words; and the target hypothesis match score for a target hypothesis comprises the product of the language model match score for the target hypothesis multiplied by the translation match score for the target hypothesis.
the translation match score for a target hypothesis comprises the product of the word match scores for the target hypothesis and the source words in the input series of source words; and the target hypothesis match score for a target hypothesis comprises the product of the language model match score for the target hypothesis multiplied by the translation match score for the target hypothesis.
6. An apparatus as claimed in claim 2, characterized in that:
the context of the target word in the target hypothesis which is connected to a source word is contained in at least one of two or more context classes; and the estimated conditional probability of occurrence of the source word, given the target word in the target hypothesis which is connected to the source word and given the context of the target word which is connected to the source word, comprises at least one function having a value dependent on the class containing the context of the target word which is connected to the source word.
the context of the target word in the target hypothesis which is connected to a source word is contained in at least one of two or more context classes; and the estimated conditional probability of occurrence of the source word, given the target word in the target hypothesis which is connected to the source word and given the context of the target word which is connected to the source word, comprises at least one function having a value dependent on the class containing the context of the target word which is connected to the source word.
7. An apparatus as claimed in claim 2, characterized in that:
the context of the target word in the target hypothesis which is connected to a source word comprises at least one word having a part of speech in the target hypothesis; and the estimated conditional probability of occurrence of the source word, given the target word in the target hypothesis which is connected to the source word and given the context of the target word which is connected to the source word, comprises at least one function having a value dependent on the part of speech in the target hypothesis of at least one word in the context of the target word which is connected to the source word.
the context of the target word in the target hypothesis which is connected to a source word comprises at least one word having a part of speech in the target hypothesis; and the estimated conditional probability of occurrence of the source word, given the target word in the target hypothesis which is connected to the source word and given the context of the target word which is connected to the source word, comprises at least one function having a value dependent on the part of speech in the target hypothesis of at least one word in the context of the target word which is connected to the source word.
8. An apparatus as claimed in claim 2, characterized in that:
the context of the target word in the target hypothesis which is connected to a source word comprises at least one word having an identity; and the estimated conditional probability of occurrence of the source word, given the target word in the target hypothesis which is connected to the source word and given the context of the target word which is connected to the source word, comprises at least one function having a value dependent on the identity of at least one word in the context of the target word which is connected to the source word.
the context of the target word in the target hypothesis which is connected to a source word comprises at least one word having an identity; and the estimated conditional probability of occurrence of the source word, given the target word in the target hypothesis which is connected to the source word and given the context of the target word which is connected to the source word, comprises at least one function having a value dependent on the identity of at least one word in the context of the target word which is connected to the source word.
9. An apparatus as claimed in claim 2, characterized in that the means for outputting the target hypothesis having the best match score comprises a display.
10. An apparatus as claimed in claim 2, characterized in that the input means comprises a keyboard.
11. An apparatus as claimed in claim 2, characterized in that the input means comprises a computer disk drive.
12. An apparatus as claimed in claim 2, characterized in that the input means comprises a computer tape drive.
13. A computer-implemented method of translating a series of source words in a first language to a series of target words in a second language different from the first language, said method comprising:
inputting a series of source words;
generating at least two target hypotheses, each target hypothesis comprising said series of target words selected from a vocabulary of words in the second language, each target word having a context comprising at least one other word in the target hypothesis;
generating, for each target hypothesis, a language model match score comprising an estimate of the probability of occurrence of the series of words in the target hypothesis;
identifying at least one alignment between the input series of source words and each target hypothesis, the alignment connecting each source word with at least one target word in the target hypothesis;
generating, for each source word and each target hypothesis, a word match score comprising an estimate of the conditional probability of occurrence of the source word, given the target word in the target hypothesis which is connected to the source word and given the context of the target word in the target hypothesis which is connected to the source word;
generating, for each target hypothesis, a translation match score comprising a combination of the word match scores for the target hypothesis and the source words in the input series of source words;
generating a target hypothesis match score for each target hypothesis, each target hypothesis match score comprising a combination of the language model match score for the target hypothesis and the translation match score for the target hypothesis; and outputting the target hypothesis having the best target hypothesis match score.
inputting a series of source words;
generating at least two target hypotheses, each target hypothesis comprising said series of target words selected from a vocabulary of words in the second language, each target word having a context comprising at least one other word in the target hypothesis;
generating, for each target hypothesis, a language model match score comprising an estimate of the probability of occurrence of the series of words in the target hypothesis;
identifying at least one alignment between the input series of source words and each target hypothesis, the alignment connecting each source word with at least one target word in the target hypothesis;
generating, for each source word and each target hypothesis, a word match score comprising an estimate of the conditional probability of occurrence of the source word, given the target word in the target hypothesis which is connected to the source word and given the context of the target word in the target hypothesis which is connected to the source word;
generating, for each target hypothesis, a translation match score comprising a combination of the word match scores for the target hypothesis and the source words in the input series of source words;
generating a target hypothesis match score for each target hypothesis, each target hypothesis match score comprising a combination of the language model match score for the target hypothesis and the translation match score for the target hypothesis; and outputting the target hypothesis having the best target hypothesis match score.
14. A method as claimed in claim 13, characterized in that each target hypothesis comprises a series of target words selected from a vocabulary comprising words in the second language and a null word representing the absence of a word.
15. A method as claimed in claim 14, characterized in that:
the step of identifying at least one alignment comprises identifying two or more alignments between the input series of source words and each target hypothesis, each alignment connecting each source word with at least one target word in the target hypothesis;
the step of generating a word match score comprises generating, for each source word and each alignment and each target hypothesis, a word match score comprising an estimate of the conditional probability of occurrence of the source word, given the target word in the target hypothesis which is connected to the source word and given the context of the target word in the target hypothesis which is connected to the source word;
the step of generating a translation match score comprises generating, for each target hypothesis, a translation match score comprising a combination of the word match scores for the target hypothesis and the alignments and the source words in the input series of source words.
the step of identifying at least one alignment comprises identifying two or more alignments between the input series of source words and each target hypothesis, each alignment connecting each source word with at least one target word in the target hypothesis;
the step of generating a word match score comprises generating, for each source word and each alignment and each target hypothesis, a word match score comprising an estimate of the conditional probability of occurrence of the source word, given the target word in the target hypothesis which is connected to the source word and given the context of the target word in the target hypothesis which is connected to the source word;
the step of generating a translation match score comprises generating, for each target hypothesis, a translation match score comprising a combination of the word match scores for the target hypothesis and the alignments and the source words in the input series of source words.
16. A method as claimed in claim 14, characterized in that:
the step of inputting comprises the step of transforming the input series of source words into a series of transformed source words;
the step of aligning comprises identifying at least one lignment between the series of transformed source words and each target hypothesis, the alignment connecting each transformed source word with at least one target word in the target hypothesis;
the step of generating a word match score comprises generating, for each transformed source word and each target hypothesis, a word match score comprising an estimate of the conditional probability of occurrence of the transformed source word, given the target word in the target hypothesis which is connected to the transformed source word and given the context of the target word in the target hypothesis which is connected to the transformed source word;the step of generating a translation match score comprises generating, for each target hypothesis, a translation match score comprising a combination of the word match scores for the target hypothesis and the transformed source words; and the step of outputting comprises the step of synthesizing a series of output words from the target hypothesis having the best target hypothesis match score.
the step of inputting comprises the step of transforming the input series of source words into a series of transformed source words;
the step of aligning comprises identifying at least one lignment between the series of transformed source words and each target hypothesis, the alignment connecting each transformed source word with at least one target word in the target hypothesis;
the step of generating a word match score comprises generating, for each transformed source word and each target hypothesis, a word match score comprising an estimate of the conditional probability of occurrence of the transformed source word, given the target word in the target hypothesis which is connected to the transformed source word and given the context of the target word in the target hypothesis which is connected to the transformed source word;the step of generating a translation match score comprises generating, for each target hypothesis, a translation match score comprising a combination of the word match scores for the target hypothesis and the transformed source words; and the step of outputting comprises the step of synthesizing a series of output words from the target hypothesis having the best target hypothesis match score.
17. A method as claimed in claim 14, characterized in that:
the translation match score for a target hypothesis comprises the product of the word match scores for the target hypothesis and the source words in the input series of source words; and the target hypothesis match score for a target hypothesis comprises the product of the language model match score for the target hypothesis multiplied by the translation match score for the target hypothesis.
the translation match score for a target hypothesis comprises the product of the word match scores for the target hypothesis and the source words in the input series of source words; and the target hypothesis match score for a target hypothesis comprises the product of the language model match score for the target hypothesis multiplied by the translation match score for the target hypothesis.
18. A method as claimed in claim 14, characterized in that:
the context of the target word in the target hypothesis which is connected to a source word is contained in at least one of two or more context classes; and the estimated conditional probability of occurrence of the source word, given the target word in the target hypothesis which is connected to the source word and given the context of the target word which is connected to the source word, comprises at least one function having a value dependent on the class containing the context of the target word which is connected to the source word.
the context of the target word in the target hypothesis which is connected to a source word is contained in at least one of two or more context classes; and the estimated conditional probability of occurrence of the source word, given the target word in the target hypothesis which is connected to the source word and given the context of the target word which is connected to the source word, comprises at least one function having a value dependent on the class containing the context of the target word which is connected to the source word.
19. A method as claimed in claim 14, characterized in that:
the context of the target word in the target hypothesis which is connected to a source word comprises at least one word having a part of speech in the target hypothesis; and the estimated conditional probability of occurrence of the source word, given the target word in the target hypothesis which is connected to the source word and given the context of the target word which is connected to the source word, comprises at least one function having a value dependent on the part of speech in the target hypothesis of at least one word in the context of the target word which is connected to the source word.
the context of the target word in the target hypothesis which is connected to a source word comprises at least one word having a part of speech in the target hypothesis; and the estimated conditional probability of occurrence of the source word, given the target word in the target hypothesis which is connected to the source word and given the context of the target word which is connected to the source word, comprises at least one function having a value dependent on the part of speech in the target hypothesis of at least one word in the context of the target word which is connected to the source word.
20. A method as claimed in claim 14, characterized in that:
the context of the target word in the target hypothesis which is connected to a source word comprises at least one word having an identity; and the estimated conditional probability of occurrence of the source word, given the target word in the target hypothesis which is connected to the source word and given the context of the target word which is connected to the source word, comprises at least one function having a value dependent on the identity of at least one word in the context of the target word which is connected to the source word.
the context of the target word in the target hypothesis which is connected to a source word comprises at least one word having an identity; and the estimated conditional probability of occurrence of the source word, given the target word in the target hypothesis which is connected to the source word and given the context of the target word which is connected to the source word, comprises at least one function having a value dependent on the identity of at least one word in the context of the target word which is connected to the source word.
21. A method as claimed in claim 14, characterized in that the step of outputting the target hypothesis having the best match score comprises displaying the target hypothesis having the best match score.
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
US08/144,913 US5510981A (en) | 1993-10-28 | 1993-10-28 | Language translation apparatus and method using context-based translation models |
US08/144,913 | 1993-10-28 |
Publications (2)
Publication Number | Publication Date |
---|---|
CA2125200A1 CA2125200A1 (en) | 1995-04-29 |
CA2125200C true CA2125200C (en) | 1999-03-02 |
Family
ID=22510719
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CA002125200A Expired - Fee Related CA2125200C (en) | 1993-10-28 | 1994-06-06 | Language translation apparatus and method using context-based translation models |
Country Status (6)
Country | Link |
---|---|
US (1) | US5510981A (en) |
EP (1) | EP0651340B1 (en) |
AT (1) | ATE187564T1 (en) |
CA (1) | CA2125200C (en) |
DE (1) | DE69421999T2 (en) |
ES (1) | ES2139690T3 (en) |
Families Citing this family (234)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JPH06195373A (en) * | 1992-12-24 | 1994-07-15 | Sharp Corp | Machine translation system |
US5675815A (en) * | 1992-11-09 | 1997-10-07 | Ricoh Company, Ltd. | Language conversion system and text creating system using such |
GB2279164A (en) * | 1993-06-18 | 1994-12-21 | Canon Res Ct Europe Ltd | Processing a bilingual database. |
US6304841B1 (en) * | 1993-10-28 | 2001-10-16 | International Business Machines Corporation | Automatic construction of conditional exponential models from elementary features |
EP0720106A3 (en) * | 1994-12-28 | 1997-07-23 | Canon Kk | System for generating natural language information from information expressed by concept and method therefor |
JPH08329105A (en) * | 1995-05-31 | 1996-12-13 | Canon Inc | Method and device for processing document |
US5890123A (en) * | 1995-06-05 | 1999-03-30 | Lucent Technologies, Inc. | System and method for voice controlled video screen display |
EP0834139A4 (en) * | 1995-06-07 | 1998-08-05 | Int Language Engineering Corp | Machine assisted translation tools |
US5721939A (en) * | 1995-08-03 | 1998-02-24 | Xerox Corporation | Method and apparatus for tokenizing text |
JPH09128396A (en) * | 1995-11-06 | 1997-05-16 | Hitachi Ltd | Preparation method for bilingual dictionary |
SG49804A1 (en) * | 1996-03-20 | 1998-06-15 | Government Of Singapore Repres | Parsing and translating natural language sentences automatically |
JP2987099B2 (en) * | 1996-03-27 | 1999-12-06 | 株式会社日立国際ビジネス | Document creation support system and term dictionary |
US6161083A (en) * | 1996-05-02 | 2000-12-12 | Sony Corporation | Example-based translation method and system which calculates word similarity degrees, a priori probability, and transformation probability to determine the best example for translation |
US6233544B1 (en) | 1996-06-14 | 2001-05-15 | At&T Corp | Method and apparatus for language translation |
US5819265A (en) * | 1996-07-12 | 1998-10-06 | International Business Machines Corporation | Processing names in a text |
US5832480A (en) * | 1996-07-12 | 1998-11-03 | International Business Machines Corporation | Using canonical forms to develop a dictionary of names in a text |
GB9619165D0 (en) * | 1996-09-13 | 1996-10-23 | British Telecomm | Training apparatus and method |
US6085162A (en) * | 1996-10-18 | 2000-07-04 | Gedanken Corporation | Translation system and method in which words are translated by a specialized dictionary and then a general dictionary |
US6122606A (en) * | 1996-12-10 | 2000-09-19 | Johnson; William J. | System and method for enhancing human communications |
US5926652A (en) * | 1996-12-20 | 1999-07-20 | International Business Machines Corporation | Matching of wild card patterns to wild card strings associated with named computer objects |
DE19708184A1 (en) * | 1997-02-28 | 1998-09-03 | Philips Patentverwaltung | Method for speech recognition with language model adaptation |
US6233545B1 (en) * | 1997-05-01 | 2001-05-15 | William E. Datig | Universal machine translator of arbitrary languages utilizing epistemic moments |
US6052682A (en) * | 1997-05-02 | 2000-04-18 | Bbn Corporation | Method of and apparatus for recognizing and labeling instances of name classes in textual environments |
US5991710A (en) * | 1997-05-20 | 1999-11-23 | International Business Machines Corporation | Statistical translation system with features based on phrases or groups of words |
WO1999000789A1 (en) * | 1997-06-26 | 1999-01-07 | Koninklijke Philips Electronics N.V. | A machine-organized method and a device for translating a word-organized source text into a word-organized target text |
DE69837979T2 (en) * | 1997-06-27 | 2008-03-06 | International Business Machines Corp. | System for extracting multilingual terminology |
CN1067782C (en) * | 1997-07-02 | 2001-06-27 | 华建机器翻译有限公司 | Inference technology on imperfect knowledge |
CN1067784C (en) * | 1997-07-02 | 2001-06-27 | 华建机器翻译有限公司 | Processing technology for special language phenomenon |
US5995926A (en) * | 1997-07-21 | 1999-11-30 | Lucent Technologies Inc. | Technique for effectively recognizing sequence of digits in voice dialing |
US6151022A (en) * | 1997-12-01 | 2000-11-21 | Microsoft Corporation | Method and apparatus for statically testing visual resources |
US6157905A (en) * | 1997-12-11 | 2000-12-05 | Microsoft Corporation | Identifying language and character set of data representing text |
US6275831B1 (en) | 1997-12-16 | 2001-08-14 | Starfish Software, Inc. | Data processing environment with methods providing contemporaneous synchronization of two or more clients |
US6216131B1 (en) * | 1998-02-06 | 2001-04-10 | Starfish Software, Inc. | Methods for mapping data fields from one data set to another in a data processing environment |
US6195631B1 (en) * | 1998-04-15 | 2001-02-27 | At&T Corporation | Method and apparatus for automatic construction of hierarchical transduction models for language translation |
US6345243B1 (en) * | 1998-05-27 | 2002-02-05 | Lionbridge Technologies, Inc. | System, method, and product for dynamically propagating translations in a translation-memory system |
US6345244B1 (en) * | 1998-05-27 | 2002-02-05 | Lionbridge Technologies, Inc. | System, method, and product for dynamically aligning translations in a translation-memory system |
US6173250B1 (en) | 1998-06-03 | 2001-01-09 | At&T Corporation | Apparatus and method for speech-text-transmit communication over data networks |
US6092034A (en) * | 1998-07-27 | 2000-07-18 | International Business Machines Corporation | Statistical translation system and method for fast sense disambiguation and translation of large corpora using fertility models and sense models |
US6285978B1 (en) * | 1998-09-24 | 2001-09-04 | International Business Machines Corporation | System and method for estimating accuracy of an automatic natural language translation |
JP3879321B2 (en) * | 1998-12-17 | 2007-02-14 | 富士ゼロックス株式会社 | Document summarization apparatus, document summarization method, and recording medium recording document summarization program |
US6317708B1 (en) | 1999-01-07 | 2001-11-13 | Justsystem Corporation | Method for producing summaries of text document |
US6266642B1 (en) | 1999-01-29 | 2001-07-24 | Sony Corporation | Method and portable apparatus for performing spoken language translation |
US6442524B1 (en) | 1999-01-29 | 2002-08-27 | Sony Corporation | Analyzing inflectional morphology in a spoken language translation system |
US6356865B1 (en) * | 1999-01-29 | 2002-03-12 | Sony Corporation | Method and apparatus for performing spoken language translation |
US6278968B1 (en) | 1999-01-29 | 2001-08-21 | Sony Corporation | Method and apparatus for adaptive speech recognition hypothesis construction and selection in a spoken language translation system |
US6282507B1 (en) | 1999-01-29 | 2001-08-28 | Sony Corporation | Method and apparatus for interactive source language expression recognition and alternative hypothesis presentation and selection |
US6243669B1 (en) | 1999-01-29 | 2001-06-05 | Sony Corporation | Method and apparatus for providing syntactic analysis and data structure for translation knowledge in example-based language translation |
US6519562B1 (en) * | 1999-02-25 | 2003-02-11 | Speechworks International, Inc. | Dynamic semantic control of a speech recognition system |
US6374224B1 (en) | 1999-03-10 | 2002-04-16 | Sony Corporation | Method and apparatus for style control in natural language generation |
US6629087B1 (en) * | 1999-03-18 | 2003-09-30 | Nativeminds, Inc. | Methods for creating and editing topics for virtual robots conversing in natural language |
EP2336899A3 (en) | 1999-03-19 | 2014-11-26 | Trados GmbH | Workflow management system |
US6349282B1 (en) * | 1999-04-20 | 2002-02-19 | Larnout & Hauspie Speech Products N.V. | Compound words in speech recognition systems |
US6356869B1 (en) * | 1999-04-30 | 2002-03-12 | Nortel Networks Limited | Method and apparatus for discourse management |
US6401104B1 (en) * | 1999-07-03 | 2002-06-04 | Starfish Software, Inc. | System and methods for synchronizing datasets using cooperation among multiple synchronization engines |
US20060116865A1 (en) | 1999-09-17 | 2006-06-01 | Www.Uniscape.Com | E-services translation utilizing machine translation and translation memory |
US6330530B1 (en) * | 1999-10-18 | 2001-12-11 | Sony Corporation | Method and system for transforming a source language linguistic structure into a target language linguistic structure based on example linguistic feature structures |
US6778949B2 (en) | 1999-10-18 | 2004-08-17 | Sony Corporation | Method and system to analyze, transfer and generate language expressions using compiled instructions to manipulate linguistic structures |
US6928448B1 (en) | 1999-10-18 | 2005-08-09 | Sony Corporation | System and method to match linguistic structures using thesaurus information |
US6721697B1 (en) * | 1999-10-18 | 2004-04-13 | Sony Corporation | Method and system for reducing lexical ambiguity |
US6529865B1 (en) | 1999-10-18 | 2003-03-04 | Sony Corporation | System and method to compile instructions to manipulate linguistic structures into separate functions |
US6535886B1 (en) | 1999-10-18 | 2003-03-18 | Sony Corporation | Method to compress linguistic structures |
US6484136B1 (en) * | 1999-10-21 | 2002-11-19 | International Business Machines Corporation | Language model adaptation via network of similar users |
US7392174B2 (en) * | 2000-03-20 | 2008-06-24 | Freeman Robert J | Natural-language processing system using a large corpus |
US20010029455A1 (en) * | 2000-03-31 | 2001-10-11 | Chin Jeffrey J. | Method and apparatus for providing multilingual translation over a network |
US6965858B2 (en) * | 2000-04-03 | 2005-11-15 | Xerox Corporation | Method and apparatus for reducing the intermediate alphabet occurring between cascaded finite state transducers |
JP2004501429A (en) * | 2000-05-11 | 2004-01-15 | ユニバーシティ・オブ・サザン・カリフォルニア | Machine translation techniques |
US6782356B1 (en) * | 2000-10-03 | 2004-08-24 | Hewlett-Packard Development Company, L.P. | Hierarchical language chunking translation table |
US6885985B2 (en) * | 2000-12-18 | 2005-04-26 | Xerox Corporation | Terminology translation for unaligned comparable corpora using category based translation probabilities |
IT1315160B1 (en) | 2000-12-28 | 2003-02-03 | Agostini Organizzazione Srl D | SYSTEM AND METHOD OF AUTOMATIC OR SEMI-AUTOMATIC TRANSLATION WITH PREEDITATION FOR THE CORRECTION OF ERRORS. |
US6990439B2 (en) * | 2001-01-10 | 2006-01-24 | Microsoft Corporation | Method and apparatus for performing machine translation using a unified language model and translation model |
US7177792B2 (en) * | 2001-05-31 | 2007-02-13 | University Of Southern California | Integer programming decoder for machine translation |
US7191115B2 (en) | 2001-06-20 | 2007-03-13 | Microsoft Corporation | Statistical method and apparatus for learning translation relationships among words |
US8214196B2 (en) * | 2001-07-03 | 2012-07-03 | University Of Southern California | Syntax-based statistical translation model |
US7610189B2 (en) | 2001-10-18 | 2009-10-27 | Nuance Communications, Inc. | Method and apparatus for efficient segmentation of compound words using probabilistic breakpoint traversal |
US7447623B2 (en) * | 2001-10-29 | 2008-11-04 | British Telecommunications Public Limited Company | Machine translation |
EP1306773A1 (en) * | 2001-10-29 | 2003-05-02 | BRITISH TELECOMMUNICATIONS public limited company | Machine translation |
EP1306774A1 (en) * | 2001-10-29 | 2003-05-02 | BRITISH TELECOMMUNICATIONS public limited company | Machine translation |
WO2004001623A2 (en) * | 2002-03-26 | 2003-12-31 | University Of Southern California | Constructing a translation lexicon from comparable, non-parallel corpora |
US8234115B2 (en) * | 2002-03-29 | 2012-07-31 | At&T Intellectual Property Ii, L.P. | Systems and methods for determining the N-best strings |
JP4047053B2 (en) * | 2002-04-16 | 2008-02-13 | 富士通株式会社 | Retrieval apparatus and method using sequence pattern including repetition |
US7107207B2 (en) | 2002-06-19 | 2006-09-12 | Microsoft Corporation | Training machine learning by sequential conditional generalized iterative scaling |
US7110937B1 (en) * | 2002-06-20 | 2006-09-19 | Siebel Systems, Inc. | Translation leveraging |
US20040012643A1 (en) * | 2002-07-18 | 2004-01-22 | August Katherine G. | Systems and methods for visually communicating the meaning of information to the hearing impaired |
JP3696231B2 (en) * | 2002-10-08 | 2005-09-14 | 松下電器産業株式会社 | Language model generation and storage device, speech recognition device, language model generation method and speech recognition method |
US7149688B2 (en) * | 2002-11-04 | 2006-12-12 | Speechworks International, Inc. | Multi-lingual speech recognition with cross-language context modeling |
US20040111272A1 (en) * | 2002-12-10 | 2004-06-10 | International Business Machines Corporation | Multimodal speech-to-speech language translation and display |
US7356457B2 (en) * | 2003-02-28 | 2008-04-08 | Microsoft Corporation | Machine translation using learned word associations without referring to a multi-lingual human authored dictionary of content words |
JP2004362249A (en) * | 2003-06-04 | 2004-12-24 | Advanced Telecommunication Research Institute International | Translation knowledge optimization device, computer program, computer and storage medium for translation knowledge optimization |
US7711545B2 (en) * | 2003-07-02 | 2010-05-04 | Language Weaver, Inc. | Empirical methods for splitting compound words with application to machine translation |
US8548794B2 (en) * | 2003-07-02 | 2013-10-01 | University Of Southern California | Statistical noun phrase translation |
US20070136251A1 (en) * | 2003-08-21 | 2007-06-14 | Idilia Inc. | System and Method for Processing a Query |
CA2536265C (en) | 2003-08-21 | 2012-11-13 | Idilia Inc. | System and method for processing a query |
US8135575B1 (en) * | 2003-08-21 | 2012-03-13 | Google Inc. | Cross-lingual indexing and information retrieval |
US7774333B2 (en) * | 2003-08-21 | 2010-08-10 | Idia Inc. | System and method for associating queries and documents with contextual advertisements |
WO2005033909A2 (en) * | 2003-10-08 | 2005-04-14 | Any Language Communications Inc. | Relationship analysis system and method for semantic disambiguation of natural language |
US7677723B2 (en) | 2003-10-09 | 2010-03-16 | Ipventure, Inc. | Eyeglasses with a heart rate monitor |
US7983896B2 (en) * | 2004-03-05 | 2011-07-19 | SDL Language Technology | In-context exact (ICE) matching |
WO2005089340A2 (en) * | 2004-03-15 | 2005-09-29 | University Of Southern California | Training tree transducers |
US8296127B2 (en) * | 2004-03-23 | 2012-10-23 | University Of Southern California | Discovery of parallel text portions in comparable collections of corpora and training using comparable texts |
US20050216253A1 (en) * | 2004-03-25 | 2005-09-29 | Microsoft Corporation | System and method for reverse transliteration using statistical alignment |
US8666725B2 (en) * | 2004-04-16 | 2014-03-04 | University Of Southern California | Selection and use of nonstatistical translation components in a statistical machine translation framework |
US20060015323A1 (en) * | 2004-07-13 | 2006-01-19 | Udupa Raghavendra U | Method, apparatus, and computer program for statistical translation decoding |
US20060031071A1 (en) * | 2004-08-03 | 2006-02-09 | Sony Corporation | System and method for automatically implementing a finite state automaton for speech recognition |
JP5452868B2 (en) * | 2004-10-12 | 2014-03-26 | ユニヴァーシティー オブ サザン カリフォルニア | Training for text-to-text applications that use string-to-tree conversion for training and decoding |
JP4473702B2 (en) * | 2004-11-02 | 2010-06-02 | 株式会社東芝 | Machine translation system, machine translation method and program |
US7451125B2 (en) * | 2004-11-08 | 2008-11-11 | At&T Intellectual Property Ii, L.P. | System and method for compiling rules created by machine learning program |
US7672830B2 (en) * | 2005-02-22 | 2010-03-02 | Xerox Corporation | Apparatus and methods for aligning words in bilingual sentences |
JP2006277103A (en) * | 2005-03-28 | 2006-10-12 | Fuji Xerox Co Ltd | Document translating method and its device |
US7882116B2 (en) * | 2005-05-18 | 2011-02-01 | International Business Machines Corporation | Method for localization of programming modeling resources |
US7607918B2 (en) * | 2005-05-27 | 2009-10-27 | Dybuster Ag | Method and system for spatial, appearance and acoustic coding of words and sentences |
US20060282256A1 (en) * | 2005-06-13 | 2006-12-14 | Werner Anna F | Translation method utilizing core ancient roots |
US8676563B2 (en) | 2009-10-01 | 2014-03-18 | Language Weaver, Inc. | Providing human-generated and machine-generated trusted translations |
WO2006133571A1 (en) * | 2005-06-17 | 2006-12-21 | National Research Council Of Canada | Means and method for adapted language translation |
US8886517B2 (en) | 2005-06-17 | 2014-11-11 | Language Weaver, Inc. | Trust scoring for language translation systems |
US7974833B2 (en) | 2005-06-21 | 2011-07-05 | Language Weaver, Inc. | Weighted system of expressing language information using a compact notation |
US20070010989A1 (en) * | 2005-07-07 | 2007-01-11 | International Business Machines Corporation | Decoding procedure for statistical machine translation |
US7389222B1 (en) | 2005-08-02 | 2008-06-17 | Language Weaver, Inc. | Task parallelization in a text-to-text system |
US7813918B2 (en) * | 2005-08-03 | 2010-10-12 | Language Weaver, Inc. | Identifying documents which form translated pairs, within a document collection |
US7624020B2 (en) * | 2005-09-09 | 2009-11-24 | Language Weaver, Inc. | Adapter for allowing both online and offline training of a text to text system |
US7697827B2 (en) | 2005-10-17 | 2010-04-13 | Konicek Jeffrey C | User-friendlier interfaces for a camera |
US10319252B2 (en) | 2005-11-09 | 2019-06-11 | Sdl Inc. | Language capability assessment and training apparatus and techniques |
WO2007068123A1 (en) * | 2005-12-16 | 2007-06-21 | National Research Council Of Canada | Method and system for training and applying a distortion component to machine translation |
US7958164B2 (en) | 2006-02-16 | 2011-06-07 | Microsoft Corporation | Visual design of annotated regular expression |
EP2511833B1 (en) | 2006-02-17 | 2020-02-05 | Google LLC | Encoding and adaptive, scalable accessing of distributed translation models |
US7860881B2 (en) * | 2006-03-09 | 2010-12-28 | Microsoft Corporation | Data parsing with annotated patterns |
US20070214153A1 (en) * | 2006-03-10 | 2007-09-13 | Mazzagatti Jane C | Method for processing an input particle stream for creating upper levels of KStore |
US20070239631A1 (en) * | 2006-03-28 | 2007-10-11 | Nokia Corporation | Method, apparatus and computer program product for generating a graphical image string to convey an intended message |
US8943080B2 (en) | 2006-04-07 | 2015-01-27 | University Of Southern California | Systems and methods for identifying parallel documents and sentence fragments in multilingual document collections |
US9020804B2 (en) * | 2006-05-10 | 2015-04-28 | Xerox Corporation | Method for aligning sentences at the word level enforcing selective contiguity constraints |
US8886518B1 (en) * | 2006-08-07 | 2014-11-11 | Language Weaver, Inc. | System and method for capitalizing machine translated text |
US7860705B2 (en) * | 2006-09-01 | 2010-12-28 | International Business Machines Corporation | Methods and apparatus for context adaptation of speech-to-speech translation systems |
JP2008065395A (en) * | 2006-09-04 | 2008-03-21 | Fuji Xerox Co Ltd | Translation device, translation method and translation program |
US8521506B2 (en) | 2006-09-21 | 2013-08-27 | Sdl Plc | Computer-implemented method, computer software and apparatus for use in a translation system |
US8214199B2 (en) * | 2006-10-10 | 2012-07-03 | Abbyy Software, Ltd. | Systems for translating sentences between languages using language-independent semantic structures and ratings of syntactic constructions |
US8145473B2 (en) | 2006-10-10 | 2012-03-27 | Abbyy Software Ltd. | Deep model statistics method for machine translation |
US9235573B2 (en) | 2006-10-10 | 2016-01-12 | Abbyy Infopoisk Llc | Universal difference measure |
US9047275B2 (en) | 2006-10-10 | 2015-06-02 | Abbyy Infopoisk Llc | Methods and systems for alignment of parallel text corpora |
US9633005B2 (en) | 2006-10-10 | 2017-04-25 | Abbyy Infopoisk Llc | Exhaustive automatic processing of textual information |
US20080086298A1 (en) * | 2006-10-10 | 2008-04-10 | Anisimovich Konstantin | Method and system for translating sentences between langauges |
US8195447B2 (en) * | 2006-10-10 | 2012-06-05 | Abbyy Software Ltd. | Translating sentences between languages using language-independent semantic structures and ratings of syntactic constructions |
US8548795B2 (en) * | 2006-10-10 | 2013-10-01 | Abbyy Software Ltd. | Method for translating documents from one language into another using a database of translations, a terminology dictionary, a translation dictionary, and a machine translation system |
US9645993B2 (en) | 2006-10-10 | 2017-05-09 | Abbyy Infopoisk Llc | Method and system for semantic searching |
US9984071B2 (en) | 2006-10-10 | 2018-05-29 | Abbyy Production Llc | Language ambiguity detection of text |
US8433556B2 (en) | 2006-11-02 | 2013-04-30 | University Of Southern California | Semi-supervised training for statistical word alignment |
US9122674B1 (en) * | 2006-12-15 | 2015-09-01 | Language Weaver, Inc. | Use of annotations in statistical machine translation |
US20080162117A1 (en) * | 2006-12-28 | 2008-07-03 | Srinivas Bangalore | Discriminative training of models for sequence classification |
CA2675208A1 (en) * | 2007-01-10 | 2008-07-17 | National Research Council Of Canada | Means and method for automatic post-editing of translations |
US8468149B1 (en) | 2007-01-26 | 2013-06-18 | Language Weaver, Inc. | Multi-lingual online community |
US8615389B1 (en) | 2007-03-16 | 2013-12-24 | Language Weaver, Inc. | Generation and exploitation of an approximate language model |
US7895030B2 (en) * | 2007-03-16 | 2011-02-22 | International Business Machines Corporation | Visualization method for machine translation |
US8959011B2 (en) | 2007-03-22 | 2015-02-17 | Abbyy Infopoisk Llc | Indicating and correcting errors in machine translation systems |
US8831928B2 (en) * | 2007-04-04 | 2014-09-09 | Language Weaver, Inc. | Customizable machine translation service |
US8825466B1 (en) | 2007-06-08 | 2014-09-02 | Language Weaver, Inc. | Modification of annotated bilingual segment pairs in syntax-based machine translation |
US8812296B2 (en) | 2007-06-27 | 2014-08-19 | Abbyy Infopoisk Llc | Method and system for natural language dictionary generation |
US8639509B2 (en) * | 2007-07-27 | 2014-01-28 | Robert Bosch Gmbh | Method and system for computing or determining confidence scores for parse trees at all levels |
US8229232B2 (en) * | 2007-08-24 | 2012-07-24 | CVISION Technologies, Inc. | Computer vision-based methods for enhanced JBIG2 and generic bitonal compression |
US8209164B2 (en) * | 2007-11-21 | 2012-06-26 | University Of Washington | Use of lexical translations for facilitating searches |
US8706477B1 (en) | 2008-04-25 | 2014-04-22 | Softwin Srl Romania | Systems and methods for lexical correspondence linguistic knowledge base creation comprising dependency trees with procedural nodes denoting execute code |
US9411800B2 (en) * | 2008-06-27 | 2016-08-09 | Microsoft Technology Licensing, Llc | Adaptive generation of out-of-dictionary personalized long words |
US9262409B2 (en) | 2008-08-06 | 2016-02-16 | Abbyy Infopoisk Llc | Translation of a selected text fragment of a screen |
US20100121630A1 (en) * | 2008-11-07 | 2010-05-13 | Lingupedia Investments S. A R. L. | Language processing systems and methods |
CN102227723B (en) * | 2008-11-27 | 2013-10-09 | 国际商业机器公司 | Device and method for supporting detection of mistranslation |
US20100204977A1 (en) * | 2009-02-09 | 2010-08-12 | Inventec Corporation | Real-time translation system that automatically distinguishes multiple languages and the method thereof |
GB2468278A (en) * | 2009-03-02 | 2010-09-08 | Sdl Plc | Computer assisted natural language translation outputs selectable target text associated in bilingual corpus with input target text from partial translation |
US9262403B2 (en) | 2009-03-02 | 2016-02-16 | Sdl Plc | Dynamic generation of auto-suggest dictionary for natural language translation |
GB201016385D0 (en) * | 2010-09-29 | 2010-11-10 | Touchtype Ltd | System and method for inputting text into electronic devices |
US8762130B1 (en) | 2009-06-17 | 2014-06-24 | Softwin Srl Romania | Systems and methods for natural language processing including morphological analysis, lemmatizing, spell checking and grammar checking |
US8762131B1 (en) | 2009-06-17 | 2014-06-24 | Softwin Srl Romania | Systems and methods for managing a complex lexicon comprising multiword expressions and multiword inflection templates |
US8990064B2 (en) | 2009-07-28 | 2015-03-24 | Language Weaver, Inc. | Translating documents based on content |
US8380486B2 (en) | 2009-10-01 | 2013-02-19 | Language Weaver, Inc. | Providing machine-generated translations and corresponding trust levels |
KR101377459B1 (en) * | 2009-12-21 | 2014-03-26 | 한국전자통신연구원 | Apparatus for interpreting using utterance similarity measure and method thereof |
US20120254333A1 (en) * | 2010-01-07 | 2012-10-04 | Rajarathnam Chandramouli | Automated detection of deception in short and multilingual electronic messages |
US8478581B2 (en) * | 2010-01-25 | 2013-07-02 | Chung-ching Chen | Interlingua, interlingua engine, and interlingua machine translation system |
US9152484B2 (en) | 2010-02-26 | 2015-10-06 | Red Hat, Inc. | Generating predictive diagnostics via package update manager |
US10534624B2 (en) * | 2010-02-26 | 2020-01-14 | Red Hat, Inc. | Generating and storing translation information as package metadata |
US10417646B2 (en) | 2010-03-09 | 2019-09-17 | Sdl Inc. | Predicting the cost associated with translating textual content |
US8265923B2 (en) * | 2010-05-11 | 2012-09-11 | Xerox Corporation | Statistical machine translation employing efficient parameter training |
GB201200643D0 (en) | 2012-01-16 | 2012-02-29 | Touchtype Ltd | System and method for inputting text |
US9123339B1 (en) | 2010-11-23 | 2015-09-01 | Google Inc. | Speech recognition using repeated utterances |
US9128929B2 (en) | 2011-01-14 | 2015-09-08 | Sdl Language Technologies | Systems and methods for automatically estimating a translation time including preparation time in addition to the translation itself |
US20120221571A1 (en) * | 2011-02-28 | 2012-08-30 | Hilarie Orman | Efficient presentation of comupter object names based on attribute clustering |
US11003838B2 (en) | 2011-04-18 | 2021-05-11 | Sdl Inc. | Systems and methods for monitoring post translation editing |
US8694303B2 (en) | 2011-06-15 | 2014-04-08 | Language Weaver, Inc. | Systems and methods for tuning parameters in statistical machine translation |
US9201862B2 (en) * | 2011-06-16 | 2015-12-01 | Asociacion Instituto Tecnologico De Informatica | Method for symbolic correction in human-machine interfaces |
US8886515B2 (en) | 2011-10-19 | 2014-11-11 | Language Weaver, Inc. | Systems and methods for enhancing machine translation post edit review processes |
US8942973B2 (en) | 2012-03-09 | 2015-01-27 | Language Weaver, Inc. | Content page URL translation |
US8989485B2 (en) | 2012-04-27 | 2015-03-24 | Abbyy Development Llc | Detecting a junction in a text line of CJK characters |
US8971630B2 (en) | 2012-04-27 | 2015-03-03 | Abbyy Development Llc | Fast CJK character recognition |
US10261994B2 (en) | 2012-05-25 | 2019-04-16 | Sdl Inc. | Method and system for automatic management of reputation of translators |
KR101629415B1 (en) * | 2012-08-10 | 2016-06-10 | 에스케이텔레콤 주식회사 | Method for detecting grammar error and apparatus thereof |
US10638221B2 (en) | 2012-11-13 | 2020-04-28 | Adobe Inc. | Time interval sound alignment |
US10249321B2 (en) * | 2012-11-20 | 2019-04-02 | Adobe Inc. | Sound rate modification |
US9152622B2 (en) | 2012-11-26 | 2015-10-06 | Language Weaver, Inc. | Personalized machine translation via online adaptation |
US10455219B2 (en) | 2012-11-30 | 2019-10-22 | Adobe Inc. | Stereo correspondence and depth sensors |
US9519858B2 (en) * | 2013-02-10 | 2016-12-13 | Microsoft Technology Licensing, Llc | Feature-augmented neural networks and applications of same |
US9342794B2 (en) * | 2013-03-15 | 2016-05-17 | Bazaarvoice, Inc. | Non-linear classification of text samples |
US9213694B2 (en) | 2013-10-10 | 2015-12-15 | Language Weaver, Inc. | Efficient online domain adaptation |
US9779087B2 (en) * | 2013-12-13 | 2017-10-03 | Google Inc. | Cross-lingual discriminative learning of sequence models with posterior regularization |
RU2592395C2 (en) | 2013-12-19 | 2016-07-20 | Общество с ограниченной ответственностью "Аби ИнфоПоиск" | Resolution semantic ambiguity by statistical analysis |
RU2586577C2 (en) | 2014-01-15 | 2016-06-10 | Общество с ограниченной ответственностью "Аби ИнфоПоиск" | Filtering arcs parser graph |
US9740687B2 (en) | 2014-06-11 | 2017-08-22 | Facebook, Inc. | Classifying languages for objects and entities |
RU2596600C2 (en) | 2014-09-02 | 2016-09-10 | Общество с ограниченной ответственностью "Аби Девелопмент" | Methods and systems for processing images of mathematical expressions |
US9626358B2 (en) | 2014-11-26 | 2017-04-18 | Abbyy Infopoisk Llc | Creating ontologies by analyzing natural language texts |
US9864744B2 (en) | 2014-12-03 | 2018-01-09 | Facebook, Inc. | Mining multi-lingual data |
US10067936B2 (en) | 2014-12-30 | 2018-09-04 | Facebook, Inc. | Machine translation output reranking |
US9830404B2 (en) | 2014-12-30 | 2017-11-28 | Facebook, Inc. | Analyzing language dependency structures |
US9830386B2 (en) | 2014-12-30 | 2017-11-28 | Facebook, Inc. | Determining trending topics in social media |
US9477652B2 (en) | 2015-02-13 | 2016-10-25 | Facebook, Inc. | Machine learning dialect identification |
US9940321B2 (en) | 2015-03-15 | 2018-04-10 | Graham MOREHEAD | System for machine translation |
EP3089159B1 (en) | 2015-04-28 | 2019-08-28 | Google LLC | Correcting voice recognition using selective re-speak |
US9734142B2 (en) * | 2015-09-22 | 2017-08-15 | Facebook, Inc. | Universal translation |
US9959271B1 (en) | 2015-09-28 | 2018-05-01 | Amazon Technologies, Inc. | Optimized statistical machine translation system with rapid adaptation capability |
US10185713B1 (en) * | 2015-09-28 | 2019-01-22 | Amazon Technologies, Inc. | Optimized statistical machine translation system with rapid adaptation capability |
US10268684B1 (en) * | 2015-09-28 | 2019-04-23 | Amazon Technologies, Inc. | Optimized statistical machine translation system with rapid adaptation capability |
US10133738B2 (en) | 2015-12-14 | 2018-11-20 | Facebook, Inc. | Translation confidence scores |
US9734143B2 (en) | 2015-12-17 | 2017-08-15 | Facebook, Inc. | Multi-media context language processing |
US10002125B2 (en) | 2015-12-28 | 2018-06-19 | Facebook, Inc. | Language model personalization |
US9805029B2 (en) | 2015-12-28 | 2017-10-31 | Facebook, Inc. | Predicting future translations |
US9747283B2 (en) | 2015-12-28 | 2017-08-29 | Facebook, Inc. | Predicting future translations |
US10902221B1 (en) | 2016-06-30 | 2021-01-26 | Facebook, Inc. | Social hash for language models |
US10902215B1 (en) | 2016-06-30 | 2021-01-26 | Facebook, Inc. | Social hash for language models |
US10180935B2 (en) | 2016-12-30 | 2019-01-15 | Facebook, Inc. | Identifying multiple languages in a content item |
US10592614B1 (en) | 2017-01-19 | 2020-03-17 | Amdocs Development Limited | System, method, and computer program for translating unified ticketing system (UTS) messages |
US9996328B1 (en) * | 2017-06-22 | 2018-06-12 | Archeo Futurus, Inc. | Compiling and optimizing a computer code by minimizing a number of states in a finite machine corresponding to the computer code |
US10481881B2 (en) * | 2017-06-22 | 2019-11-19 | Archeo Futurus, Inc. | Mapping a computer code to wires and gates |
US10380249B2 (en) | 2017-10-02 | 2019-08-13 | Facebook, Inc. | Predicting future trending topics |
US10635863B2 (en) | 2017-10-30 | 2020-04-28 | Sdl Inc. | Fragment recall and adaptive automated translation |
US10817676B2 (en) | 2017-12-27 | 2020-10-27 | Sdl Inc. | Intelligent routing services and systems |
US11048885B2 (en) | 2018-09-25 | 2021-06-29 | International Business Machines Corporation | Cognitive translation service integrated with context-sensitive derivations for determining program-integrated information relationships |
US11256867B2 (en) | 2018-10-09 | 2022-02-22 | Sdl Inc. | Systems and methods of machine learning for digital assets and message creation |
US11106873B2 (en) * | 2019-01-22 | 2021-08-31 | Sap Se | Context-based translation retrieval via multilingual space |
CN111402864A (en) * | 2020-03-19 | 2020-07-10 | 北京声智科技有限公司 | Voice processing method and electronic equipment |
US20220335075A1 (en) * | 2021-04-14 | 2022-10-20 | International Business Machines Corporation | Finding expressions in texts |
CN115293177B (en) * | 2022-08-09 | 2023-08-01 | 上海交通大学 | Small language neural network machine translation method and system based on double transfer learning |
Family Cites Families (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JPS61105671A (en) * | 1984-10-29 | 1986-05-23 | Hitachi Ltd | Natural language processing device |
US4754489A (en) * | 1985-10-15 | 1988-06-28 | The Palantir Corporation | Means for resolving ambiguities in text based upon character context |
US4829580A (en) * | 1986-03-26 | 1989-05-09 | Telephone And Telegraph Company, At&T Bell Laboratories | Text analysis system with letter sequence recognition and speech stress assignment arrangement |
US4882759A (en) * | 1986-04-18 | 1989-11-21 | International Business Machines Corporation | Synthesizing word baseforms used in speech recognition |
US5146405A (en) * | 1988-02-05 | 1992-09-08 | At&T Bell Laboratories | Methods for part-of-speech determination and usage |
DE68928230T2 (en) * | 1988-09-15 | 1998-02-05 | Oce Tech Bv | System for the grammatical processing of a sentence composed of natural language |
JPH02288960A (en) * | 1989-02-27 | 1990-11-28 | Hitachi Ltd | Method and system for supporting preparation of document |
US5033087A (en) * | 1989-03-14 | 1991-07-16 | International Business Machines Corp. | Method and apparatus for the automatic determination of phonological rules as for a continuous speech recognition system |
US5477451A (en) * | 1991-07-25 | 1995-12-19 | International Business Machines Corp. | Method and system for natural language translation |
GB9209346D0 (en) * | 1992-04-30 | 1992-06-17 | Sharp Kk | Machine translation system |
-
1993
- 1993-10-28 US US08/144,913 patent/US5510981A/en not_active Expired - Fee Related
-
1994
- 1994-06-06 CA CA002125200A patent/CA2125200C/en not_active Expired - Fee Related
- 1994-07-18 EP EP94111148A patent/EP0651340B1/en not_active Expired - Lifetime
- 1994-07-18 ES ES94111148T patent/ES2139690T3/en not_active Expired - Lifetime
- 1994-07-18 AT AT94111148T patent/ATE187564T1/en not_active IP Right Cessation
- 1994-07-18 DE DE69421999T patent/DE69421999T2/en not_active Expired - Lifetime
Also Published As
Publication number | Publication date |
---|---|
ES2139690T3 (en) | 2000-02-16 |
EP0651340B1 (en) | 1999-12-08 |
CA2125200A1 (en) | 1995-04-29 |
DE69421999T2 (en) | 2000-06-08 |
ATE187564T1 (en) | 1999-12-15 |
EP0651340A3 (en) | 1997-11-05 |
EP0651340A2 (en) | 1995-05-03 |
US5510981A (en) | 1996-04-23 |
DE69421999D1 (en) | 2000-01-13 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CA2125200C (en) | Language translation apparatus and method using context-based translation models | |
Knight et al. | Translation with finite-state devices | |
CN104573099B (en) | The searching method and device of topic | |
US6823301B1 (en) | Language analysis using a reading point | |
Tanaka et al. | Universal dependencies for Japanese | |
US6098035A (en) | Morphological analysis method and device and Japanese language morphological analysis method and device | |
Van Wyk | Linguistic assumptions and lexicographical traditions in the African languages | |
Vositov | CLASSIFICATION THEORY OF TURKIC BORROWINGS | |
Gaussier | General considerations on bilingual terminology extraction | |
Franz | Independence assumptions considered harmful | |
Al-Onaizan et al. | Named entity translation | |
KR100784730B1 (en) | Method and apparatus for statistical HMM part-of-speech tagging without tagged domain corpus | |
Sahlgren | Towards a flexible model of word meaning | |
Moerenhout et al. | Remnants of the old order: OV in the Paston letters | |
Hajič | Linguistics meets exact sciences | |
Feinberg | Possible prehistoric contacts between Tonga and Anuta | |
JPH0578058B2 (en) | ||
EP0429057A1 (en) | Text-to-speech system having a lexicon residing on the host processor | |
Dardjowidjojo | Acronymic patterns in indonesian | |
Tadjievich | LEXICOLOGICAL PROBLEMS OF TRANSLATION | |
Scherpenisse | Scrambling in German | |
Lee et al. | Ambiguity resolution for machine translation of telegraphic messages | |
Valentínyová | Michael Skiba (2021). Participial prepositions and conjunctions in the history of English. München: Utzverlag. | |
Al-Onaizan | Named entity translation: A statistical approach | |
Mohammed et al. | Senseval-3 logic forms: A system and possible improvements |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
EEER | Examination request | ||
MKLA | Lapsed |