{
    "paper_id": "W89-0120",
    "header": {
        "generated_with": "S2ORC 1.0.0",
        "date_generated": "2023-01-19T03:44:43.282619Z"
    },
    "title": "Kunskap om v\u00e4rlden eller kunskap om texten? En metod f\u00f6r korpusst\u00f6dd maskin\u00f6vers\u00e4ttning",
    "authors": [
        {
            "first": "A",
            "middle": [],
            "last": "Bstract",
            "suffix": "",
            "affiliation": {},
            "email": ""
        }
    ],
    "year": "",
    "venue": null,
    "identifiers": {},
    "abstract": "World Knowledge or Text Knowledge? A Method fo r Corpus-based Machine Translation As the scope aind the quality demands on machine translation systems increase, the developers tend to direct their efforts not only on automating the translation process itself but also on automating the more labourintensive subprocesses o f the system development. This is the reason for a tendency towards more and more automated techniques of knowledge acquisition. Whereas commercial systems at present normally have dictio naries or knowledge banks which were generated in at best semi-automatic corpus-based or corpus-inspired ways, some o f the more advanced research projects attempt to approach fuUy automatically generated knowledge banks. From this idea it is a logical step to using a corpus directly as a knowledge source for the machine translation process. For the DLT system o f BSO in Utrecht a method is developed in which the translation process relies entirely on a Bilingual Knowledge Bank as its one and only source o f translation-relevant knowledge. The Bilingual Knowledge Bank consists o f parallel corpora of texts in a given source/t\u00a3irget language and DLT's intermediate language Esperanto. The texts are represented as dependency trees with the sentence trees being linked up by text-grammatical pointers (e.g., for deixis, reference, event chains). Corresponding elements in the parallel versions of the text are combined to form translation units. The translation process is carried out by means of various sets of generalization rules which apply the specimen translation units to a given text to be translated. Such generalizations are made at the levels o f monolingual syntax, metataxis (syntactic transfer) and semantics/pragmatics. Since the knowledge acquisition process is not carried out before the translation function requires a certain bit of information, it can be dynamically steered by the information contained in the part o f the text already translated.",
    "pdf_parse": {
        "paper_id": "W89-0120",
        "_pdf_hash": "",
        "abstract": [
            {
                "text": "World Knowledge or Text Knowledge? A Method fo r Corpus-based Machine Translation As the scope aind the quality demands on machine translation systems increase, the developers tend to direct their efforts not only on automating the translation process itself but also on automating the more labourintensive subprocesses o f the system development. This is the reason for a tendency towards more and more automated techniques of knowledge acquisition. Whereas commercial systems at present normally have dictio naries or knowledge banks which were generated in at best semi-automatic corpus-based or corpus-inspired ways, some o f the more advanced research projects attempt to approach fuUy automatically generated knowledge banks. From this idea it is a logical step to using a corpus directly as a knowledge source for the machine translation process. For the DLT system o f BSO in Utrecht a method is developed in which the translation process relies entirely on a Bilingual Knowledge Bank as its one and only source o f translation-relevant knowledge. The Bilingual Knowledge Bank consists o f parallel corpora of texts in a given source/t\u00a3irget language and DLT's intermediate language Esperanto. The texts are represented as dependency trees with the sentence trees being linked up by text-grammatical pointers (e.g., for deixis, reference, event chains). Corresponding elements in the parallel versions of the text are combined to form translation units. The translation process is carried out by means of various sets of generalization rules which apply the specimen translation units to a given text to be translated. Such generalizations are made at the levels o f monolingual syntax, metataxis (syntactic transfer) and semantics/pragmatics. Since the knowledge acquisition process is not carried out before the translation function requires a certain bit of information, it can be dynamically steered by the information contained in the part o f the text already translated.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Abstract",
                "sec_num": null
            }
        ],
        "body_text": [
            {
                "text": "A tt \u00f6 v e r s \u00e4 tta \u00e4 r en k r\u00e4 v a n d e in tellek tu ell h a n d lin g . F \u00f6 rs \u00f6 k e n a tt b y g g a m a sk in\u00f6v e rs\u00e4 ttn in g ssy ste m kan b e tra k ta s s o m e tt s tr\u00e4 v a n d e e fte r en a u to m a tis e r in g av d e n n a y tte r s t k o m p le x a v erk sa m h et. E tt p a r d ecen n iers erfjiren h eter m ed d e n n a str\u00e4 v a n h a r v is a t a tt d e t in te r\u00e4ck er a tt a u to m a tis e r a b a r a s j\u00e4 lv a \u00f6 v e r s\u00e4 ttn in g sp rocesse n . R e d a n i u p p b y g g n in g e n a v e tt m a sk in \u00f6 v e rs\u00e4 ttn in g ssy ste m in g\u00e5 r s\u00e5 k o m p le x a a rb e ts s te g a tt \u00e4 v en d essa m \u00e5 ste u tf\u00f6 ra s i s to r u tstr\u00e4 ck n in g a u to m a tisk t eller \u00e5 tm in s to n e p \u00e5 e tt a v a n cera t d a to r s t\u00f6 t t s\u00e4 tt. G iv e tv is rik ta r sig d e tta sek u n d \u00e4 ra a u to m a tiserin g sin tresse i f\u00f6 r s ta h a R e s o n e m a n g e t \u00e4r en k elt: G ra n sk n in g en a v a u to m a tisk t g e n e re ra d e le x ik o n in g \u00e5 n g a r \u00e4 r e tt tid s \u00f6 d a n d e o c h d \u00e4 rm e d d y r t a rb e te . D e t ligger d \u00e4 r f\u00f6 r n \u00e4 r a till h a n d s a tt rik ta a u tom a tise rin g sin tre sset ig en p \u00e5 d et m est ar b e tsin te n s iv a m o m e n te t o c h f\u00f6 rs \u00f6 k a a tt g \u00f6 r a g ra n sk n in g sa rb e te t \u00f6 v e r fl\u00f6 d ig t. O m d e tta v o r e m \u00f6 jlig t u ta n a tt g e avkall p \u00e5 \u00f6v e rs\u00e4 ttn in g sk v a lite te n , sk u lle m yck et v a ra v u n n e t. S \u00e5 d a n a a u to m a tis k t fra m s t\u00e4 lld a lexikoning\u00e5n g\u00a3ir u tg \u00f6 r d et fj\u00e4 rd e u tv eck lin g s ste g e t. In n a n m a n sa tsa r p \u00e5 d e tta , l\u00f6 n a r d e t sig a tt f\u00f6 r a ta n k eex p e rim e n te t t h a n \u00e5 g o n m en in g a tt d is k u te ra d e t fe m te d \u00e4 r m a n \u00e4 r tv u n g e n a tt la g ra en h el k o rp u s i st\u00e4 llet f\u00f6r n \u00e5 g r a r e d u n d a n sfria in g \u00e5 n g a r? A r in te d en k o m p a k ta re l\u00f6sn in g en u tan v id a re a tt f\u00f6 r e d r a ? J a g b e sk riv e r n e d a n en l\u00f6 s n in g so m sik ta r p \u00e5 d e t fe m te steget, ste g e t \u00e4 r m \u00f6 jlig t , s \u00e5 b e ty d e r d e t a tt m a n kan g e n e re ra f\u00f6 r \u00f6 v e rs\u00e4 ttn in g sb e h o v tillr \u00e4 c k lig a le x ik o n in g \u00e5 n g a r u r en k o rp u s m ed h j\u00e4 lp a v en p \u00e5 f\u00f6 rh a n d fa stst\u00e4 lld u p p s \u00e4 ttn in g regler. T a n k e e x p e rim e n te t g \u00e5 r u t p \u00e5 a tt m a n ur en g iv en k orp u s f\u00e5r fra m sa m m a in fo r m a tio n , o a v s e tt tid p u n k te n p \u00e5 v ilk en m a n till\u00e4 m p a r reglerna. D e t sp ela r a llts\u00e5 i d e tta a v see n d e in g en r o ll o m reg lern a a n v \u00e4 n d s in n a n eller m ed a n d e t f\u00f6re lig g e r en k on k ret \u00f6 v e r s \u00e4 ttn in g s u p p g ift. M e d a n d r a o r d , o m m a n \u00f6v e rh u v u d ta g e t kan in h \u00e4 m ta d en n \u00f6 d v \u00e4 n d ig a in fo r m a tio n e n h e la u to m a tis k t, s\u00e5 h ar m a n M h e te n a tt v \u00e4 lja o m m a n v ill f\u00f6 r l\u00e4 g g a in h \u00e4 m tn in g sp ro ce sse n till d en f\u00f6 rb e re d a n d e s y stem u tv eck lin g e n eller till sj\u00e4 lv a \u00f6 v e rs \u00e4 ttn in g s p ro ce s s e n . D e n n a a n led n in g h ar i b esk riv n in g en ov a n n \u00e5 g o r lu n v ara fr\u00e5 g a o m en a n n a n e n h e t, t e x e tt m o r fe m , e tt s y n ta g m o s v .) N \u00e4 r m a n gen erera r le x ik on in g \u00e5 n g a r, \u00e4 r d et m en in g en a tt u p p n \u00e5 en s \u00e5 a llm \u00e4 n g iltig o c h t\u00e4 ck a n d e b e sk riv n in g a v u p p s la g so rd e t s o m m \u00f6 jlig t (e lle r n \u00e5 g r a f\u00e5 s \u00e5 d a n a ). P \u00e5 fem te steg et d \u00e4 r e m o t a n v \u00e4 n d s k orp u sen d ire k t s o m k u n sk a p sk \u00e4 lla , o c h en k orp u s \u00e4r av en k v a lita tiv t a n n o rlu n d a karalet\u00e4r \u00e4n en le x ik o n in g \u00e5 n g . M e d a n en le x ik o n in g \u00e5 n g skall v a ra a llm \u00e4 n g iltig i d en m \u00e5 n d e tta \u00e4r m \u00f6 jlig t, in n eh \u00e5ller en k orp u s e n b a rt e x e m p e l. M e d a n a llts \u00e5 e tt sy ste m a v fj\u00e4 rd e ste g e t g \u00e5 r fr\u00e5n ex em p len i d e n u n d e rlig g a n d e k orp u s en g e n o m h \u00e4 rled n in g sreg ler till en i d e n n a sp e ciella b e m \u00e4 rk e ls e a llm \u00e4 n g iltig le x ik o n in g \u00e5 n g o c h d \u00e4 rifr\u00e5 n g e n o m till\u00e4 m p n in g sreg ler till ; o m k u n sk a p sb a n k en : P a p e g a a ij 1 9 8 6 ). ",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Kunskapsk\u00e4llor f\u00f6r avancerad maskin\u00f6vers\u00e4ttning",
                "sec_num": "1"
            },
            {
                "text": "S om illu stra tio n a n v \u00e4 n d er ja g f\u00f6 lja n d e m e n in g p \u00e5 d a n sk a o c h e s p e ra n to . \u00c4 v e n m era in g rip a n d e sy n ta k tisk a f\u00f6 rsk ju tn in g a r kan ia k tta s i d e n n a m en in g :",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "En illustration",
                "sec_num": "4"
            },
            {
                "text": "[4] m e d h en b lik p \u00e5 -kiu j celas D e t ta \u00e4r en \u00f6 v e r g \u00e5 n g fr\u00e5 n e tt p r e p o s itio n e llt till\u00e4 g g sle d ( D -P R E A ) i d a n sk a till en re la tiv b isa ts (R E L ) i e sp e ra n to {k iu j ' v ilk a ' , cela s ' a v se r' ). \u00d6 v ers\u00e4 ttn in g sen h e ten illu strerar b r a a tt k o rp u s in fo rm a tio n e n h ar ex e m p e lk a ra k t\u00e4 r. G iv e tv is \u00f6 v e r s \u00e4 tte r m a n in te a lltid in fin itiv k o n stru k tio n e n d et a t skabe m ed en su b s ta n tiv {la kreado 'sk a p a n d et' ), m en fo rm e n kan m y ck et v \u00e4 l t\u00e4n kas f\u00f6 r e k o m m a i v issa a n d r a k o n te x ter. In fo rm a tio n e n \u00e4 r a llts\u00e5 in te a llm \u00e4 n g iltig , m en illu stre ra r b a r a en a v m \u00e5 n g a m \u00f6 jlig a \u00f6 v e rs\u00e4 ttn in g a r f\u00f6 r d e tta sy n ta g m . I ta n k e e x p e rim e n te t ov a n a n to g ja g a tt en v \u00e4 s e n tlig d riv k ra ft f\u00f6 r a tt l\u00e5 ta u tv eck lin g en g \u00e5 v id a re fr\u00e5 n ",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "En illustration",
                "sec_num": "4"
            },
            {
                "text": "Kunskap om v\u00e4rlden eller kunskap om texten?",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "6",
                "sec_num": null
            },
            {
                "text": "S \u00e5 lu n d a till\u00e5 ter u tv e ck lin g e n a v m a sk in \u00f6 v e rs\u00e4 ttn in g ssy ste m e t D L T m ed sin n y a sy ste m stru k tu r e tt n y a rta t svar p \u00e5 fr\u00e5 g a n h u ru v id a g ru n ",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "6",
                "sec_num": null
            },
            {
                "text": "Proceedings of NODALIDA 1989, pages 218-228",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "",
                "sec_num": null
            },
            {
                "text": "Proceedings of NODALIDA 1989",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "",
                "sec_num": null
            },
            {
                "text": "Proceedings of NODALIDA 1989",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "",
                "sec_num": null
            }
        ],
        "back_matter": [],
        "bib_entries": {
            "BIBREF0": {
                "ref_id": "b0",
                "title": "Machine translation. Amsterdam/Phila^ delphia",
                "authors": [
                    {
                        "first": "John",
                        "middle": [],
                        "last": "Lehrberger",
                        "suffix": ""
                    },
                    {
                        "first": "Laurent",
                        "middle": [],
                        "last": "Bourbeau",
                        "suffix": ""
                    }
                ],
                "year": 1988,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Lehrberger, John, Laurent Bourbeau. 1988. Machine translation. Amsterdam/Phila^ delphia, Benjamins.",
                "links": null
            },
            "BIBREF1": {
                "ref_id": "b1",
                "title": "Word expert semantics",
                "authors": [
                    {
                        "first": "B",
                        "middle": [
                            "C"
                        ],
                        "last": "Papegaaij",
                        "suffix": ""
                    }
                ],
                "year": 1986,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Papegaaij, B. C. 1986. Word expert semantics. Dordrecht/Riverton, Foris.",
                "links": null
            },
            "BIBREF2": {
                "ref_id": "b2",
                "title": "Working with analogical semantics. Disambiguation techniques in DLT",
                "authors": [
                    {
                        "first": "Victor",
                        "middle": [],
                        "last": "Sadler",
                        "suffix": ""
                    }
                ],
                "year": 1989,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Sadler, Victor. 1989. Working with analogical semantics. Disambiguation techniques in DLT. Dordrecht/Providence, Foris.",
                "links": null
            },
            "BIBREF3": {
                "ref_id": "b3",
                "title": "Metataxis in practice. Dependency syntax fo r multilingual machine translation",
                "authors": [
                    {
                        "first": "Ingrid",
                        "middle": [],
                        "last": "Schubert",
                        "suffix": ""
                    }
                ],
                "year": 1989,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "39--67",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Schubert, Ingrid 1989. A dependency syntax of Danish. Dan Maxwell, Klaus Schu bert [utg.]. Metataxis in practice. Dependency syntax fo r multilingual machine translation:39-67. Dordrecht/Providence, Foris.",
                "links": null
            },
            "BIBREF4": {
                "ref_id": "b4",
                "title": "Linguistic and extra^linguistic knowledge",
                "authors": [
                    {
                        "first": "Klaus",
                        "middle": [],
                        "last": "Schubert",
                        "suffix": ""
                    }
                ],
                "year": 1986,
                "venue": "Computers and trans lation",
                "volume": "1",
                "issue": "",
                "pages": "125--152",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Schubert, Klaus. 1986. Linguistic and extra^linguistic knowledge. Computers and trans lation, 1:125-152.",
                "links": null
            },
            "BIBREF5": {
                "ref_id": "b5",
                "title": "Metataxis. Contrastive dependency syntax for machine transla tion",
                "authors": [
                    {
                        "first": "Klaus",
                        "middle": [],
                        "last": "Schubert",
                        "suffix": ""
                    }
                ],
                "year": 1987,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Schubert, Klaus. 1987. Metataxis. Contrastive dependency syntax for machine transla tion. Dordrecht/Providence, Foris.",
                "links": null
            },
            "BIBREF6": {
                "ref_id": "b6",
                "title": "The architecture of DLT-interlingual or double direct? Dan Maxwell",
                "authors": [
                    {
                        "first": "Klaus",
                        "middle": [],
                        "last": "Schubert",
                        "suffix": ""
                    }
                ],
                "year": 1988,
                "venue": "New directions in machine trans-/alion",
                "volume": "",
                "issue": "",
                "pages": "131--144",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Schubert, Klaus. 1988. The architecture of DLT-interlingual or double direct? Dan Maxwell, Klaus Schubert, Toon Witkam [utg.]. New directions in machine trans- /alion:131-144. Dordrecht/Providence, Foris.",
                "links": null
            },
            "BIBREF7": {
                "ref_id": "b7",
                "title": "Metataxis in practice. Dependency syntax fo r multilingual machine translation",
                "authors": [
                    {
                        "first": "Klaus",
                        "middle": [],
                        "last": "Schubert",
                        "suffix": ""
                    }
                ],
                "year": 1989,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "207--232",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Schubert, Klaus 1989. A dependency syntax of Esperanto. Dan Maxwell, Klaus Schu bert [utg.]. Metataxis in practice. Dependency syntax fo r multilingual machine translation:207-232. Dordrecht/Providence, Foris.",
                "links": null
            },
            "BIBREF8": {
                "ref_id": "b8",
                "title": "Distributed Language Translation",
                "authors": [
                    {
                        "first": "A",
                        "middle": [
                            "P M"
                        ],
                        "last": "Witkam",
                        "suffix": ""
                    }
                ],
                "year": 1983,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Witkam, A. P. M. 1983. Distributed Language Translation. Utrecht, BSO.",
                "links": null
            },
            "BIBREF9": {
                "ref_id": "b9",
                "title": "Probabilistic methods in dependency grammar parsing",
                "authors": [
                    {
                        "first": "Job",
                        "middle": [
                            "M"
                        ],
                        "last": "Zuijlen",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Van",
                        "suffix": ""
                    }
                ],
                "year": 1989,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Zuijlen, Job M. van 1989. Probabilistic methods in dependency grammar parsing.",
                "links": null
            }
        },
        "ref_entries": {
            "FIGREF0": {
                "text": "n d p \u00e5 d e m est a rb e tsin te n s iv a stegen i sy ste m u tv e ck lin g sp ro ce sse n . D e t ta \u00e4 r f\u00f6 r d e flesta sy stem en s v id k o m m a n d e d e lexdkografiska (sa m t term in o g ra fisk a ) a rb e ts m o m e n te n . M e r a adlm \u00e4nt sa g t g \u00e4 lle r d e t i d essa m o m e n t a tt g e n o m lex ik og ra fl eller p \u00e5 a n n a t s \u00e4 tt in h \u00e4 m ta den k u n sk ap so m b e h \u00f6 v s f\u00f6 r \u00f6 v e rs \u00e4 ttn in g s p ro ce s s e n o c h a tt g \u00f6 r a ku n sk a p en tillg \u00e4 n g lig f\u00f6 r d a to r s y s te m e t. J \u00e4 m f\u00f6 r m a n d e m a sk in \u00f6 v e rs\u00e4 ttn in g ssy ste m so m har b y g g ts eller p r o je k te r a ts sed a n e tt fy r t io ta l \u00e5 r tillb a k a , s \u00e5 kan m an ia k tt a en u tv e ck lin g i k u n sk a p sk \u00e4 llorn a s o m st\u00e5 r i d irek t sa m b a n d m e d n \u00f6 d v \u00e4 n d ig h e te n a tt a u to m a tis e ra sj\u00e4 lv a k u n sk a p sin h \u00e4 m tn in g sp ro ce sse n . U tv e ck lin g e n g \u00e4 lle r b \u00e5 d e kunskapsk \u00e4llans in n eh \u00e5 ll o c h k u n sk a p sin h \u00e4 m tn in g ss\u00e4 ttet. T a r m a n m e d \u00e4 v en tillt\u00e4 n k ta fraim tida in no\\'ationer s\u00e5 f\u00f6 r l\u00f6 p e r u tv e ck lin g e n (n \u00e5 g o t f\u00f6 r e n k la tn d g jo r d , b a s e ra d p \u00e5 in tu itio n ) i O r d b o k (h a n d g jo r d , k o rp u s in sp ire ra d ) i O r d b o k /K u n s k a p s b a n k (h a lv a u to m a tis k k o r p u s s t\u00f6 d d fra m s t\u00e4 lln in g s p ro ce s s ) i K u n sk a p sb a n k (h e la u to m a tis k k o rp u s s t\u00f6 d d fra m s t\u00e4 lln in g s p ro ce s s ) i K o r p u s i d ire k t b ru k (a u to m a tis k a k u n sk a p sin h \u00e4 m tn in g sm e to d e r) D e f\u00f6 r s ta t v \u00e5 u tv eck lin g ssteg en va r m y ck et v a n lig a i m a sk in \u00f6 v e rs\u00e4 ttn in g e n s f\u00f6 r s ta d e cen n ier. D a g e n s sy stem h a r f\u00f6 r d e t m e s ta u p p n \u00e5 tt e tt s\u00e5 d a n t o m f\u00e5 n g o c h s\u00e5 d a n a k v a litetsk ra v a tt rent h a n d a r b e te h a r b liv it o g \u00f6 r lig t. F u llst\u00e4 n d ig t h a n d g jo r d a o r d b \u00f6 c k e r f\u00f6 re k o m m e r d \u00e4 r e m o t \u00e4 v en i d a g i m a sk in \u00f6 v e rs\u00e4 ttn in g s sy stem so m \u00e4r re la tiv t sm \u00e5 , d v s sy ste m s o m a n tin g en \u00e4r a v s e d d a e n b a rt f\u00f6 r e x p erim en te llt b ru k eller s o m \u00e4 r in sk r\u00e4 n k ta till en m y ck et s n \u00e4 v \u00e4 m n e s d o m \u00e4 n . 219 Proceedings of NODALIDA 1989 S t\u00f6 rre sy ste m m e d fria re te x ts o r t s o m \u00e4r b e r\u00e4 k n a d e f\u00f6 r p ra k tisk t b ru k h ar o fta en k u n sk a p sin h \u00e4 m tn in g sm e to d s o m m o tsv a ra r d et tr e d je u tv e ck lin g sste g et: E tt p r o g r a m s y s te m a n a ly sera r en k o rp u s s o m v a n lig tv is red a n \u00e4 r f\u00f6 rse d d m ed tillagd d is a m b ig u e rin g s in fo rm a tio n o c h g en erera r u r k orp u sm a te ria let lex ik on in g \u00e5 n g a r i m a s k in \u00f6 v e rs\u00e4 ttn in g ssy ste m e ts s p e cie lla fo r m a t. In g \u00e5 n g a rn a granskas sedan av en m \u00e4 n n isk a o c h g o d ta s eller k orrig era s o c h k om p lette ra s. I st\u00e4 llet f\u00f6 r (eller v id sid a n o m ) e n k o rp u s a v l\u00f6 p a n d e te x t a n v \u00e4 n d s ib la n d \u00e4 v e n v a n lig a o r d b \u00f6 c k e r i b o k fo r m s o m g \u00f6 r s tillg \u00e4 n g lig a f\u00f6 r d a ta b e h a n d lin g g e n o m o p tis k in l\u00e4 sn in g eller k on v e rte rin g a v d a ta filer fr\u00e5 n s \u00e4 tt-o c h try ck m a sk in er. 2 Autom atisera systemutvecklingen? M ig v ete rlig e n h a r d et h ittills in te m a rk n a d sf\u00f6rts n \u00e5 g o t m a sk in \u00f6v ers\u00e4 ttn in g ssy ste m s o m b y g g e r p \u00e5 en m e ra fra m sk rid en tek n ik \u00e4n d en ja g h \u00e4r b esk river s o m d e t tr e d je s te g e t. D \u00e4 r e m o t f\u00f6 rs\u00f6 k e r m a n i s o m lig a a v d e m est a v an cerade nu p \u00e5 g \u00e5 e n d e fo rsk n in g s-o c h u tv e ck lin g s p ro je k te n a tt u p p n \u00e5 d et fj\u00e4 rd e eller till o c h m e d d e t fe m te s te g e t.",
                "num": null,
                "uris": null,
                "type_str": "figure"
            },
            "FIGREF1": {
                "text": "id a re . O m d et sk u lle v is a sig v a ra m \u00f6 jlig t a tt h e la u to m a tisk t g en erera lex ik on in g \u00e5 n g a r m e d u tg \u00e5 n g s p u n k t i en k o rp u s , s \u00e5 b e ty d e r d e tta a tt d en in fo rm a tio n m^ln b e h \u00f6 v e r f\u00f6 r a tt k u n n a \u00f6 v e r s \u00e4 tta finn s i k o rp u s te x te n o c h kan h \u00e4m tas d \u00e4 rifr\u00e5 n p \u00e5 e tt h e la u to m a tis k t s\u00e4 tt. \" H e la u to m a tis k t\" b e ty d e r i d e tta sa m m a n h a n g fr a m f\u00f6 r a llt a tt in g en ku n sk ap b e h \u00f6 v e r l\u00e4 g g a s till a v m \u00e4n n isk an . O m d e tta \u00e4 r s\u00e5, d \u00e5 kan m a n ev e n tu e llt lika g \u00e4 r n a l\u00e5 ta b li a tt fra m st\u00e4 lla in g \u00e5 n g a r o c h i st\u00e4 llet a n lita k o rp u s en d irek t s o m k u n sk a p sk \u00e4 lla . D e t t a \u00e4r d et fe m te steg et i k u n sk a p sk \u00e4 llorn a s u tv eck lin g . T a b e lle n ovein p resen tera r d et fe m te ste g e t s o m en fo r ts \u00e4 ttn in g eller v id a re u tv e c k lin g a v d e t Q \u00e4rd e, m en m e d ta n k e p \u00e5 p ro g ra m sy ste m e n s sto rle k o c h s n a b b h e t u n d ra r m a n kanske o m d e t in te sn a ra re \u00e4 r e tt s te g tillb a k a . O m d et fj\u00e4 rd e o c h d e t fe m te ste g e t \u00e4 r lik v \u00e4 rd ig a , kan d e t d \u00e5 \u00f6 v e rh u v u d ta g e",
                "num": null,
                "uris": null,
                "type_str": "figure"
            },
            "FIGREF2": {
                "text": "s \u00e5 a tt d e t \u00e4r p \u00e5 sin p la ts a tt skaffa sig k la rh et o m d et p r e c is a f\u00f6 rh \u00e5 lla n d et m e lla n h e la u to m a tis k in g \u00e5 n g sg e n e re rin g o c h d irek t k o rp u s b ru k . O m d e t Q \u00e4rde",
                "num": null,
                "uris": null,
                "type_str": "figure"
            },
            "FIGREF3": {
                "text": "D e t l\u00f6n a r sig in te a tt f\u00f6 r a d e tta ta n k e e x p e rim e n t v id a re o m in te d e t fe m te steget e r b ju d e r v \u00e4 s e n tlig a f\u00f6 rd e la r j\u00e4 m f\u00f6 r t m e d d e t s o m \u00e4r m \u00f6 jlig t re d a n p \u00e5 det Q \u00e4rde. M a n m \u00e5 ste h a m y ck et \u00f6 v e r ty g a n d e a rg u m en t n \u00e4 r m a n v ill a v s t\u00e5 fr\u00e5n m \u00f6 jlig h e te n a tt u n d a n g \u00f6 ra en s\u00e5 sv \u00e5 r d e lp r o c e s s so m k o r p u s s t\u00f6 d d k u nsk a p sin h \u00e4 m tn in g o n e k lig e n \u00e4 r red a n i u tv eck lin g sfa se n o c h i st\u00e4 llet u p p s k ju ta den till sj\u00e4 lv a \u00f6 v e rs\u00e4 ttn in g sp ro ce sse n i ru n tim e. D e t e n d a g iltig a kan v a ra e tt a rgu m en t s o m b y g g e r p \u00e5 v ik tig tilla g d in fo r m a tio n s o m b lir tillg \u00e4 n g lig f\u00f6 rst n\u00e4r \u00f6 v e rs\u00e4 ttn in g sp ro ce sse n h ar k o m m it ig \u00e5 n g . B a r a o m k u n sk a p sin h \u00e4 m tn in g sp ro cesse n kan sty ra s eller a v se v \u00e4 rt f\u00f6 r b \u00e4 ttr a s g e n o m k u n sk ap eller v illk o r h \u00e4 m ta d e ur d e n te x t so m \u00e4 r u n d er b e a r b e tn in g , d \u00e5 kan d e t l\u00f6 n a s ig a tt t\u00e4 n k a p \u00e5 en l\u00f6 sn in g p \u00e5 fe m te steget. I d en l\u00f6 sn in g ja g skisserar i a v sn itt 3 t o m 5 \u00e4r k u n sk apsk \u00e4llan o c h d en redan \u00f6 v e r s a tta delen a v te x te n rep re sen tera d e i sa m m a y tn \u00e4 r a fo r m a t. B l a d e tta g \u00f6 r d et m \u00f6 jlig t a tt g e n o m f\u00f6 r a fre k v e n sb e r\u00e4 k n in g a r, p ro b a b ilistisk a k o n te x tj\u00e4 m f\u00f6 re lse r o c h lik n a n d e d e lp r o c e s s e r p \u00e5 e tt sp ecifik t s \u00e4 tt s o m \u00e4 r a n p a ssa t till d en k o n k re ta k o n te x te n o c h s o m d y n a m is k t ta r m ed i b e r\u00e4 k n in g e n d en kunskap s o m kan in h \u00e4 m ta s ur d en red a n b e h a n d la d e te x td e le n . P \u00e5 d e tta s \u00e4 tt b lir d en k u n sk a p sb eh a n d lin g sp rocess s o m s t\u00f6 d e r \u00f6 v e r s \u00e4 ttn in g s fu n k tio n e n i h \u00f6 g g ra d sty rd a v e tt v \u00e4 la v v \u00e4 g t sa m sp el m ella n d en a llm \u00e4 n n a o c h d e n f\u00f6 r tillf\u00e4 llet m est releva n ta sp ecie lla k u nskapsk\u00e4llan.U t\u00f6 v e r f\u00f6 rd e la r som kan u p p n \u00e5 s g e n o m en k u n sk a p sin h \u00e4 m tn in g sp ro ce ss i ru n tim e finn s d et y tte rlig a re en a n led n in g a tt in tressera sig f\u00f6 r d e t fe m te ste g e t.",
                "num": null,
                "uris": null,
                "type_str": "figure"
            },
            "FIGREF4": {
                "text": "d a d o lt s a v fra m st\u00e4 lln in g s s\u00e4 tte t i d en fe m ste g ig a u tv eck lin g en . J a g h a r h ittills b a r a d isk u te ra t d e t fe m te steg et u n d er f\u00f6 r u ts \u00e4 ttn in g a tt d et Q \u00e4rde \u00e4r g e n o m f\u00f6 r b a r t, o c h ja g h a r i t\u00e4 m lig en a llm \u00e4 n n a o r d a la g ta la t o m d e n in fo rm a tio n m a n kan in h \u00e4 m ta m ed d e t v \u00e5 a n ty d d a m e to d e rn a : P \u00e5 d e t fe m te steg et \u00e4r d e n n a in fo rm a tio n m in st lik v \u00e4 rd ig m e d d en m an f\u00e5 r p \u00e5 d e t Q \u00e4rde, o c h d et finn s a n led n in g a tt a n ta a tt d e t d \u00e4 r u t\u00f6 v e r \u00e4r m \u00f6 jlig t a tt in h \u00e4 m ta tilla g d in fo rm a tio n s o m b a r a \u00e4 r tillg \u00e4 n g lig t p \u00e5 d e t fem te steg et. D e t t a re so n e m a n g f\u00e5 r em ellertid in te d \u00f6 lja d en k v a lita tiv a sk illn a d so m \u00e4 n d \u00e5 b e st\u00e5 r m ella n d e t Q \u00e4rde o c h d et fe m te steg et. S k illn aden b lir ty d lig n \u00e4 r m an g \u00e5 r n \u00e4 rm a re in p \u00e5 i v ilk en fo r m in fo rm a tio n e n la gra s. P \u00e5 fj\u00e4 r d e ste g e t u tv \u00e4 rd e ra s k orpu sen f\u00f6 r a tt g en erera lex ik on in g \u00e5 n g a r. P r o c e s s e n s u td a ta \u00e4 r a llts\u00e5 en fa stla g d rep resen ta tion f\u00f6 r d e n in h \u00e4 m ta d e ku n sk a p en . K u n sk a p e n rep resen tera s s\u00e5 lu n d a p \u00e5 e tt e x p lic it s\u00e4 tt. E n lex ik o n in g \u00e5 n g skall v a ra till\u00e4 m p lig p \u00e5 vilk a s o m h elst f\u00f6 re k o m ste r a v u p p s la g s o rd e t. (I st\u00e4 llet f\u00f6 r e tt u p p s la g s o r d kan d e t g iv e tv is",
                "num": null,
                "uris": null,
                "type_str": "figure"
            },
            "FIGREF5": {
                "text": "d en k o n k re ta \u00f6 v e r s \u00e4 ttn in g s u p p g ifte n , s \u00e5 l\u00e4 ggs p \u00e5 fe m te ste g e t e tt om e d e lb a rt f\u00f6 r b a n d m ella n e x e m p le n o c h u p p g ifte n . D e t a llm \u00e4 n g iltig a m ella n steg et kan falla b o r t. D e t t a \u00e4 r en v \u00e4 s e n tlig ieikttagelse. F \u00f6 r a tt b e v is a till\u00e4 m p lig h e te n a v d e t fem te ste g e t b e h \u00f6 v e r m a n a llts \u00e5 in te f\u00f6 r u ts \u00e4 t ta a tt d et Q \u00e4rd e \u00e4r m \u00f6jU gt. D e t r\u00e4cker a tt b e v is a a tt m a n ur k o rp u s e x e m p e l d irek t Imn h \u00e4 rled a d en in fo rm a tio n som b e h \u00f6 v s f\u00f6 r \u00f6 v e rs \u00e4 ttn in g s p ro ce s s e n .",
                "num": null,
                "uris": null,
                "type_str": "figure"
            },
            "FIGREF6": {
                "text": "\u00f6 r m a sk in \u00f6 v e rs\u00e4 ttn in g ssy ste m e t D L T p r o je k te r a s n u m era en k o rp u s s t\u00f6 d d k u nsk a p s b e h a n d lin g s m e to d s o m str\u00e4 v a r e fte r a tt n \u00e4 rm a sig skalans fe m te steg.In n a n ja g ta r u p p m e to d e n n \u00e5 g o t m e ra i d e ta lj kan e tt p a r in le d a nd e o r d \u00f6v e r D L T v a ra n \u00f6 d v \u00e4 n d ig a . D istrib u ted L a n g u a g e T ranslation (D L T ) \u00e4 r n a m n e t p \u00e5 ett m a s k in \u00f6 v e r s \u00e4 ttn in g s p r o je k t s o m b e d r iv s a v d et n ed erl\u00e4 n d sk a m ju k v a ru f\u00f6 re ta g et B u r o v o o r S y ste e m o n tw ik k e lin g (B S O /R e s e a r c h ) i U tre ch t, d e lv is m e d s ta tlig t a n sla g . E fte r en f\u00f6 r s tu d ie (W itk a m 1 9 8 3 ) in tr\u00e4 d d e D L T \u00e5 r 1985 i im p lem e n te rin g sfa sen . D e n f\u00f6 r s t a p r o to ty p e n b le v f\u00e4 rd ig 1987, d en a n d r a 1988. D L T skall b li e tt m \u00e5 n g sp r\u00e5 k ig t sy ste m , b l a f\u00f6 r till\u00e4 m p n in g a r i d a ta k om m u n ik a tio n sn \u00e4 t. U n d e r u tg \u00e5 n g ssp r\u00e5 k sa n a ly sen f\u00f6 rs en sy ste m in itie ra d d isa m b ig u e rin g sd ia lo g m ed a n v \u00e4 n d a re n . D ia lo g & \u00e5 g o r n a st\u00e4lls p \u00e5 u tg \u00e5 n g ssp r\u00e5 k et o c h d et k r\u00e4 v s in gen p o s te d i te rin g , s\u00e5 a tt a n v \u00e4 n d a re n in te b e h \u00f6 v e r k \u00e4 n n a till m \u00e5 lsp r\u00e5 k en . F \u00f6 rb in d elsel\u00e4 n k en m e lla n u tg \u00e5 n g s -o c h m \u00e5 lsp r\u00e5 k en \u00e4 r m ella n sp r\u00e5 k et e sp e ra n to . D e f\u00f6 r s t a p r o to ty p v e r s io n e r n a \u00f6 v e r s \u00e4 tte r fr\u00e5 n en gelsk a g e n o m e sp e ra n to till fran ska. S o m k u n sk a p sk \u00e4 llor anU tar d e tre m o rfo sy n ta k tisk a o r d b \u00f6 c k e r (en g el ska, e s p e r a n to , f\u00f6a n sk a ), t v \u00e5 tv \u00e5 s p r \u00e5 k ig a m e ta ta x o r d b \u00f6 c k e r (en g elsk a -esp e ra n to, e sp e ra n to -fra n sk a ; o m te rm e n m e ta ta x j f i S ch u b ert 1 987) o c h en e n sp r\u00e5 k ig lexikal k u n sk a p sb a n k (e s p e r a n to ). D e o lik a fra m st\u00e4 lln in g s p ro ce ss e rn a l\u00e5 g m ellan det f\u00f6 r s t a o c h d et tr e d je ste g e t p \u00e5 skalan (jfr o m p r o to ty p e n s a rk itek tu r: S ch u bert 1986",
                "num": null,
                "uris": null,
                "type_str": "figure"
            },
            "FIGREF7": {
                "text": "tv \u00e4 r d e r in g e n a v e rfa re n h e te rn a m ed p r o to ty p e r n a hsir lett fra m till en v id a r e u tv e c k lin g a v k u n sk a p sk \u00e4 llorn a s o m b e ty d e r en in g rip a n d e f\u00f6 r\u00e4 n d rin g i s y s te m e t D L T :s s \u00e4 tt a tt fu n g era . D L T \u00e4 r (re d a n i p ro to ty p v e r s io n e r n a ) ett m o d u l\u00e4 r t sy s te m . D e t b e s t\u00e5 r a v s p r\u00e5 k p a rs m o d u le r s o m a lltid h a r e s p e ra n to p \u00e5 d e n e n a sid a n . E n te x t s o m \u00f6 v e rs \u00e4 tts till e tt e n d a m \u00e5 lsp r\u00e5 k passerar p \u00e5 s\u00e5 s \u00e4 tt t v \u00e5 s \u00e5 d a n a s p r\u00e5 k p a rs m o d u le r. D e t \u00e4 r b l a p \u00e5 g ru n d a v d e n n a arkitek tu r s o m s y s te m e t kan b e tr a k ta s s o m e tt d u b b e lt d irek t \u00f6v e rs\u00e4 ttn in g ssy stem (S c h u b e r t 1 9 8 8 ). I D L T :s t r e d je sy ste m v e rsio n , s o m b efin n er sig i pla n erin gso c h m o d e llim p le m e n te rin g s fa s e n , h a r a lla k u n sk a p sk \u00e4 llor so m in g \u00e5 r i sa m m a s p r \u00e5 k p a r s m o d u l s a m m a n fa tta ts till e tt e n d a sy ste m . D e t ta sy ste m h ar f\u00e5 tt n am n et T v\u00e5 spr\u00e5 k ig kunskapsbank. D L T :s T v \u00e5 s p r \u00e5 k ig a k u n sk a p sb a n k b e s t\u00e5 r a v en p a ra llell k o rp u s , d v s en o c h sa m m a te x t p a ra llellt p \u00e5 t v \u00e5 sp r\u00e5 k (o rig in a l o c h \u00f6 v e r s \u00e4 ttn in g , \u00e4 v en t v \u00e5 \u00f6 v e r s\u00e4 ttn in g a r fr\u00e5 n e tt tr e d je s p r\u00e5 k ). I d e t p \u00e5 g \u00e5 e n d e p r o v im p le m e n te r in g s a r b e te t in g\u00e5r p aren en g elsk a resp era n to o c h fra n sk a -esp era n to. I d e n slu tg iltig a im p le m en terin gen a v d en tr e d je sy stem v ersion en skall m in st t v \u00e5 sp r\u00e5 k k o m m a till; sen are ve rsion er p r o je k te r a s f\u00f6 r tv \u00e5 p a k e t a v s e x sp r\u00e5 k var, v a refter fle ra sp r\u00e5 k kan l\u00e4 g g a s till e fte r b e h o v ta c k vare D L T :s m o d u l\u00e4 r a m ella n sp r\u00e5 k sa rk itek tu r. J ag illu strera r d e n T v \u00e5 s p r \u00e5 k ig a k u n sk ap sban k en n ed a n m e d sp r\u00e5 k p a re t d a n sk a esp era n to. K o r p u s t e x t e m a la gra s i d e n T v \u00e5 s p r \u00e5 k ig a k u n sk a p sb a n k en i d isa m b ig u e ra d fo rm . D en g ru n d l\u00e4 g g a n d e re p re se n ta tio n sform en \u00e4 r d e p e n d e n ssy n ta k tisk a tr \u00e4 d d ia g ra m , u td a ta a v en paxser (jfr S ch u b e rt 1987: 2 8 -1 2 9 ). D e s sa \u00e4r sy n ta k tiskt o a m b ig u \u00f6 s a . T e x tg ra m m a tisk a pek a re (d e ix is, referen s, s k e e n d e k e d jo r m m ) f\u00f6 rb in d e r sa tsern a o c h m en in g a rn a till s a m m a n h \u00e4 n g a n d e te x te r . V id sid a n o m dessa e n sp r\u00e5 k ig a m a rk \u00f6 re r \u00e4 r te x te r n a f\u00f6 r s e d d a m ed s p e cie lla tv \u00e5 s p r \u00e5 k ig a p ek a re so m b y g g e r u p p \u00f6 v e rs\u00e4 ttn in g se n h e te r. E n \u00f6v e rs\u00e4 ttn in g se n h e t \u00e4 r e tt o r d , en o r d g r u p p eller b a ra e tt m o rfe m m ed d ess m o tsv a rig h e t i d e t a n d r a sp r\u00e5 k et. M a rk \u00f6 re rn a f\u00f6 r d e sy n ta k tisk a re la tio n e rn a s o m \u00e4 r u ts a tta i d e p e n d e n s tr \u00e4 d e t in g\u00e5r \u00e4ven i \u00f6v ers\u00e4 ttn in g se n h e te n . E n s t\u00f6 r r e \u00f6v e rs\u00e4 ttn in g se n h e t kan in n e h \u00e5 lla m in d re en h eter. E n h e te rn a \u00e4 r d o c k in te m in d re \u00e4n a tt d en \u00f6 v e r s \u00e4 ttn in g s m o tsvarigh et d e in n eh \u00e5ller kan a n v \u00e4 n d a s \u00e4 v en i a n d r a k o n te x te r.",
                "num": null,
                "uris": null,
                "type_str": "figure"
            },
            "FIGREF8": {
                "text": "T r\u00e4 d d ia r g ra m m e n b y g g e r p \u00e5 d e p e n d e n s s y n ta x e rn a s o m h ar u ta r b e ta ts en lig t D L T -m o d e ll f\u00f6 r da n sk a (In g rid S ch u b ert 1 989) o c h e s p e r a n to (S c h u b e r t 1 9 8 9 ) d \u00e4 r d e h \u00e4 r a n v \u00e4 n d a e tik e tte rn a f\u00f6 r d e p e n d e n sre la tio n e r f\u00f6rk la ra s i d e ta lj. A v u try m m e s sk \u00e4 l ta r ja g b a r a en e n d a m en in g o c h v isa r b a r a d e n \u00f6 v r e d elen a v tr \u00e4 d d ia g ra m m e n . T r\u00e4 d a v sn itt m e d sa m m a n u m m er i b \u00e5 d a tr\u00e4 d e n u tg \u00f6 r \u00f6 v e rs\u00e4 ttn in g se n h e te r.F \u00f6 r \u00f6 v e rsik tlig h e te n s sku ll m a rk era r ja g l\u00e5 n g t ifr\u00e5n a lla \u00f6 v e rs\u00e4 ttn in g se n h e te rn a .T e x tg ra m m a tisk a pek a re u tel\u00e4 m n a s h elt.F or a t g \u00f8 r e d e t le tte re at sk a b e b e d r e \u00f8 k o n o m isk e b e tin g e lse r, fre m sae tte r K o m m is s io n e n 80 fo rs la g m e d h en b lik p \u00e5 a t n e d b r y d e m a r k edssk rankern e o g sae tte v irk so m h e d e rn e i sta n d til fu ld t u d a t d ra g e fo r d e l a f d en eu ro p ae isk e d im en sion . P o r fa cilig i la k re a d o n d e pli b o n a j e k o n o m ia j k o n d ic o j la K o m is io n o faxas 80 p r o p o n o jn , kiu j cela s fa lig i la b a r ilo jn d e la m e rk a to kaj e b lig i al la e n tr e p r e n o j m a k sim u m e elu zi la a v a n ta g o jn d e la e iir o p a d im en sio. (M e d h \u00e4 n syn till e n ty d ig h e t i o rd s tru k tu re n m a rk era s m o rfe m g r\u00e4 n s e rn a i D L T :s m ella n sp r\u00e5 k s o m \u00e4x fu llst\u00e4 n d ig t a g g lu tin e ra n d e . M o r fe m te c k n e n ses i e sp e ra n to tr\u00e4 d e t m ed h a r u te l\u00e4 m n a ts h \u00e4 r.) I d e n n a m e n in g f\u00f6 r e k o m m e r b \u00e5 d e e n k la o c h k o m p le x a \u00f6vers\u00e4 ttn in g se n h eter. T ill d e e n k la re h \u00f6 r e tt o r d s e n h e te m a : [10] \u00f8 k o n o m is k e -e k o n o m ia j O v e rs\u00e4 ttn in g se n h e te n [2] fo r s la g -p r o p o n o jn in n eh \u00e5 ller p \u00e5 e s p e ra n to s id a n en a ck u sa tiv ( -n ) , s\u00e5 a tt en g iltig m otsv a rig h e t b a ra f\u00f6 re lig g e r n \u00e4 r o b je k te t ik e tte n ta s m e d i en h e ten . O b s e r v e r a a tt fo r s la g m o tsv a ra r p r o p o n o jn m en a tt fr e m s ae tt e r m t e m otsv a ra r fa ra s. V e r b e t fa r i b e ty d e r 'g \u00f6 r a ' o c h kan in te b e tra k ta s s o m b ru k b a r \u00f6 v e rs\u00e4 ttn in g a v fr e m s ae t t e i a n d r a k o n te x t. D \u00e4 r f\u00f6 r o m fa tta r e n h eten flera o r d : [1] fr e m s ae tte r fo r s la g -faxas p r o p o n o jn E n m e r a k o m p le x m o tsv a rig h e t \u00e4r [5] a t g \u00f8 r e le tte r e -fa cilig i Figur 2:",
                "num": null,
                "uris": null,
                "type_str": "figure"
            },
            "FIGREF9": {
                "text": "[6] d e t a t sk a b e -la k rea d on d e",
                "num": null,
                "uris": null,
                "type_str": "figure"
            },
            "FIGREF10": {
                "text": "\u00d6vers\u00e4ttarens kunnande i maskin\u00f6vers\u00e4ttningssystemet D e t e s p e ra jito -d a n s k a m e n in g sp a re t \u00e4 r g o d ty c k lig t valt. A tt illu stra tion en u t\u00f6v er e n k la fa ll s o m [10] eller [3] \u00e4 v en in n eh \u00e5 ller s\u00e5 m \u00e5 n g a stru k tu rellt o c h lexik alt in tre ssa n ta m o ts v a rig h e te r \u00e4 r e tt teck en p \u00e5 a tt en k o rp u s st\u00f6 d d \u00f6 v e rs\u00e4 ttn in g sm e to d h a r tillg \u00e5 n g till en m \u00e4 n g d k o n stru k tio n e r o c h m o tsv a rig h e te r so m i d en n a fo rm a ld rig u p p ta s i v a n lig a o r d b \u00f6 c k e r . H \u00e4r \u00e5 te rsp eg la s i im p licit fo r m \u00f6vers\u00e4 tta ren s k u n n a n d e.I D L T :s k u n sk a p sb a n k in g \u00e5 r en s to r m \u00e4 n g d e x e m p e l p \u00e5 hur v issa te x t b ita r i k o n k re ta tillf\u00e4 llen h ar \u00f6 v e rs a tts a v fa ck k u n n ig a \u00f6 v e rs\u00e4 tta re . \" Uppiinn^lren\"a v d en (p a te n ta n m \u00e4 ld a ) T v \u00e5 s p r \u00e5 k ig a k u n sk apsban k en , V ic t o r S adler, b esk river i d e ta lj h ur m a n kan \u00f6 v e r s \u00e4 tta m e d h j\u00e4 lp av d e n n a k u n sk ap (S a d ler 1989). H a n s fra m st\u00e4 lln in g , s o m in te kan \u00e5 te rg es h\u00e4r, vill ja g i k orth e t k om p le tte ra m ed t v \u00e5 a s p e k te r s o m kan b e ly s a d e n n a k o r p u s s t\u00f6 d d a \u00f6 v e r s \u00e4 ttn in g s m e to d m ed u tg \u00e5 n g s p u n k t i re so n e m a n g e t o m d en fe m ste g ig a skalan i m a sk in \u00f6v ers\u00e4 ttn in g ssy ste m e n s u tv eck lin g . D e n f\u00f6 r s t a a sp ek ten b e tr\u00e4 ffa r u p p b y g g n in g e n a v en T v \u00e5sp r\u00e5 k ig k u n sk a p sb a n k o c h d en a n d r a till\u00e4 m p n in g sreg lern a . I d isk u ssion en o m D L T :s f\u00f6 r n y a d e s y ste m a rk ite k tu r b \u00f6 r t v \u00e5 fu n k tio n e r h\u00e5llas k la rt \u00e5 tsk ild a : \u00e5 e n a sid a n k u n sk a p sb eh a n d lin g en u n d er \u00f6 v e rs\u00e4 ttn in g sp ro ce sse n i ru n tim e o c h \u00e5 a n d r a sid an k u n sk a p sin h \u00e4 m tn in g en . D e n sen are p ro cessen , i v ilk en u p p b y g g n in g e n a v d en T v \u00e5 s p r \u00e5 k ig a k u n sk ap sban k en in g \u00e5 r, \u00e4r f\u00f6 rla g d till D L T -\" fa b rik e n \" , m e d a n sj\u00e4 lv a \u00f6 v e rs \u00e4 ttn in g s p ro ce s s e n g iv e tv is \u00e4 g er ru m h os a n v \u00e4 n d a re n . \u00c4 v e n o m d e t n a tu rlig tv is \u00e4 r \u00f6 n s k v \u00e4 rt a tt a u to m a tis e r a kunskapsin h \u00e4 m tn in g s p ro ce s s e n i h \u00f6 g g ra d , \u00e4r d et d o c k in te u te slu te t a tt u tf\u00f6 r a m an u ella in g rip a n d e n o c h a tt a n lita s p e c ia lis th j\u00e4 lp s o m in te \u00e4r tillg \u00e4 n g lig u n d er \u00f6 v e r s\u00e4 ttn in g s p r o c e s s e n . P \u00e5 e tt lik n a n d e s \u00e4 tt kan in h \u00e4 m tn in g en o c k s \u00e5 u tn y ttja st\u00f6rre o c h a n n o r lu n d a d a to r s y s te m \u00e4n a n v \u00e4 n d a rm o d u le rn a . U n d e r \u00f6 v e rs\u00e4 ttn in g sp ro cessen \u00e4 r d e n e n d a m \u00e4 n sk lig a h j\u00e4 lp s y ste m e t kan f\u00e5 d e svar so m g es i d en in te ra k tiv a d isa m b ig u e rin g sd ia lo g e n . D ia lo g fr \u00e5 g o r n a st\u00e4lls p \u00e5 u tg \u00e5 n g ssp r\u00e5 k et o ch m \u00e5 s te k u n n a b e sv a ra s a v sp r\u00e5 k -o c h d a ta v e te n sk a p lig a lek m \u00e4n . D e n n a skilln ad f\u00f6 rk la ra r h u r d e t \u00e4 r m \u00f6 jlig t a tt d en T v \u00e5 s p r \u00e5 k ig a k u n sk apsban k en i s y stem u tv e ck lin gsfa sen f\u00f6 rse s m ed all d e n u to m s p r \u00e5 k lig a in fo r m a tio n ja g ov a n h ar n \u00e4 m n t: o a m b ig u \u00f6 s a sy n ta k tis k a tr \u00e4 d , te x tg ra m m a tis k a pek a re, \u00f6 v e rs\u00e4 ttn in g sm o tsv a righ e te r m m . I D L T -fa b r ik e n ge n erera s d e ssa stru k tu rer a u to m a tis k t, t e x g e n o m p a rsn in g , i d en m \u00e5 n d e t t a \u00e4r m \u00f6 jlig t o c h sed a n gran sk as o c h k o m p le ttera s d e a v m \u00e4 n n isk o r. M \u00e4 n sk lig t a r b e te b e h \u00f6 v s fr a m f\u00f6 r a llt f\u00f6 r iden tifierin gen av \u00f6 v e rs \u00e4 ttn in g s m o ts v a rig h e te rn a . M e d a n D L T -a n v \u00e4 n d a re n kan v a ra en en sp r\u00e5k ig p e r s o n s o m b a r a f\u00f6 r s t\u00e5 r u tg \u00e5 n g s te x te n , k r\u00e4 v er k u n sk a p sin h \u00e4 m tn in g sp rocessen s p e c ia lis ta r b e te , b l a a v y rk e s\u00f6 v e rs\u00e4 tta re . D e n T v \u00e5 s p r \u00e5 k ig a k u n sk apsban k en \u00e4r a llts \u00e5 in te in sk r\u00e4 n k t till d e n k v a litet i b e ty d e ls e a n a ly se n so m i d a g en s l\u00e4ge kan u p p n \u00e5 s m e d h e la u to m a tis k a k u n sk a p sb eh a n d lin g sp roce sse r, utam d en har tillg \u00e5 n g till m \u00e4 n sk lig fa ck k u n sk a p . D e n a n d r a a sp e k t s o m ja g v ill n \u00e4 m n a h \u00e4 r g \u00e4 lle r g e n era liserin g sfu n k tion en . S o m e tt s p e cie llt sla gs k o rp u s in n eh \u00e5 ller d en T v \u00e5 s p r \u00e5 k ig a ku n sk apsban k en ex 226 Proceedings of NODALIDA 1989 e m p e l p \u00e5 fa k tisk t b ru k a v o r d , u ttr y c k o c h \u00f6 v e rs \u00e4 ttn in g s m o ts v a rig h e te r. M e n va rje k orp u s \u00e4r n \u00f6 d v \u00e4 n d ig tv is \" f\u00f6 r lite n \" , d v s en e x e m p e ls a m lin g , h u r s to r den \u00e4n b lir, kan a ld rig in n e h \u00e5 lla va rje a n v \u00e4 n d n in g s m \u00f6 jlig h e t a v v a rje en sta k a o r d o c h v a rje en stak a k o n stru k tio n (jfr L e h r b e r g e r /B o u r b e a u 1988: 1 2 9 ). S ta tistik en v isa r a tt u n g ef\u00e4 r h \u00e4 lfte n a v a lla lem m a n i en k o rp u s f\u00f6 r e k o m m e r b a r a en e n d a g \u00e5 n g i l\u00f6 p te x te n . D e reg ler m e d vars h j\u00e4 lp k o rp u s in fo rm a tio n e n till\u00e4 m p a s p \u00e5 k on k re ta \u00f6 v e r s \u00e4 ttn in g s u p p g ifte r (p a rsn in g , s y n ta k tis k tra n sfer, lexik a l tra n sfer, se m a n tisk -p ra g m a tisk d isa m b ig u e rin g o s v ) b \u00f6 r d \u00e4 r f\u00f6 r g e n e ra lise ra m e d u tg \u00e5 n g s p u n k t i k o rp u s e x e m p le n . S \u00e5 d a n a g e n era lise rin g sfu n k tion er b e h \u00f6 v s p \u00e5 m in st tre plan: 1. en sp r\u00e5 k ig s y n ta x (p a rsn in g [jfr Z u ijlen 1989], m o r fo s y n ta k tis k s y n te s ); 2. m e ta ta x (sy n ta k tisk tra n sfer); 3. s e m a n tik -p ra g m a tik (m \u00e4 tn in g a v b e ty d e ls e a v s t\u00e5 n d m m ). I \u00f6 v e ren sst\u00e4 m m else m ed D L T :s a r k ite k tu rp rin cip e r f\u00f6 lje r d e ssa g en era lise rin gsregler im p licite ts id \u00e9 n , v ilk e t b l a in n e b \u00e4 r a tt d e u n d v ik e r a tt a n lita en e x p licit a llm \u00e4 n g iltig m e lla n re p re se n ta tio n s o m d e n s o m sk u lle b e h \u00f6 v a s i e tt sy stem p \u00e5 skalans Q \u00e4rde steg.",
                "num": null,
                "uris": null,
                "type_str": "figure"
            },
            "FIGREF11": {
                "text": "d et tr e d je till d e t fj\u00e4 r d e o c h fe m te ste g e t \u00e4 r in tresset i a tt e rs \u00e4 tta d e t m \u00e4 n sk lig a a r b e te s o m \u00e4r n \u00f6 d v \u00e4 n d ig t i k u n sk a p sk o d erin g sp rocessen m ed a u tom a tisk a p ro ce sse r. D e n l\u00f6 s n in g ja g sk isserar ov a n realiserar d e n n a v id a re u tv e ck lin g , m en d en sa k n a r \u00e4 n d \u00e5 in te a r b e ts m o m e n t d \u00e4 r in fo r m a tio n l\u00e4 g g s till a v m \u00e4 n n isk or. M \u00e4 n n isk a n , sy ste m u tv e ck la re n , h a r a llts \u00e5 in te ra tion a lise ra ts b o r t . M en D L T :s k o r p u s s t\u00f6 d d a \u00f6 v e r s \u00e4 ttn in g s m e to d in n e b \u00e4 r a tt d et b id ra g sp e cia liste rn a l\u00e4 m n a r till k u n sk a p sin h \u00e4 m tn in g e n m o tsv a ra r i m y ck et h \u00f6 g re g ra d \u00e4 n v id tr e d je u tv eck lin g ssteg et e tt v a n lig t s \u00e4 tt a tt re son era \u00f6 v e r sp r\u00e5 k . B er m a n en sp e cia list a tt g e e x e m p e l p \u00e5 sp r\u00e5 k b ru k i sitt \u00e4 m n e s o m r\u00e5 d e eller a tt gran sk a f\u00f6 re sla g n a fo rm u le rin g a r s\u00e5 \u00e4r u p p g ifte n e n k la re o c h svciren p \u00e5 litlig a re \u00e4n n \u00e4r m a n \u00e4 r tv u n g e n a tt b e o m en a llm \u00e4 n g iltig m e ta sp r\u00e5 k lig red o g \u00f6 re lse . D e som m e d a r b e ta r i D L T :s sy stem u tv eck lin g sfa s kan d \u00e4 r f\u00f6 r i h \u00f6 g r e u tstr\u00e4 ck n in g vara s p e cia lisera d e i \u00e4 m n e so m r\u00e5 d e t o c h i \u00f6 v e r s \u00e4 ttn in g o c h b e h \u00f6 v e r i m in d re g ra d k on cen trera sig p \u00e5 te o re tis k g ra m m a tik eller lex ik og ra fi.",
                "num": null,
                "uris": null,
                "type_str": "figure"
            },
            "FIGREF12": {
                "text": "d v a le t f\u00f6 r m a sk in \u00f6 v e rs\u00e4 ttn in g s\u00e4 n d a m \u00e5 let b \u00f6 r v a ra (u to m s p r \u00e5 k lig ) k u n sk ap o m v \u00e4 rld e n eller (in o msp r\u00e5 k lig ) k u nskap o m te x te n . D L T :s svar \u00e4r a tt d et \u00e4r k u nskap ur te x te r.",
                "num": null,
                "uris": null,
                "type_str": "figure"
            }
        }
    }
}