{
    "paper_id": "W89-0101",
    "header": {
        "generated_with": "S2ORC 1.0.0",
        "date_generated": "2023-01-19T03:44:41.006599Z"
    },
    "title": "Automatisk morfologisk analyse af islandsk tekst",
    "authors": [
        {
            "first": "Stefan",
            "middle": [],
            "last": "Briem",
            "suffix": "",
            "affiliation": {},
            "email": ""
        },
        {
            "first": "A",
            "middle": [],
            "last": "Bstract",
            "suffix": "",
            "affiliation": {},
            "email": ""
        }
    ],
    "year": "",
    "venue": null,
    "identifiers": {},
    "abstract": "Automatic Morphological Analysis o f Icelandic Text One of the projects worked on at the Institute o f Lexicography at the University o f Iceland is a frequency analysis of Icelandic vocabulary and grammar. The most time-consuming part o f the work consists in morpho logical analysis of text samples containing in all more than half a million running words. For every single word the analysis results in registration of the word class, the flexion form and the lemma to which the text word belongs.",
    "pdf_parse": {
        "paper_id": "W89-0101",
        "_pdf_hash": "",
        "abstract": [
            {
                "text": "Automatic Morphological Analysis o f Icelandic Text One of the projects worked on at the Institute o f Lexicography at the University o f Iceland is a frequency analysis of Icelandic vocabulary and grammar. The most time-consuming part o f the work consists in morpho logical analysis of text samples containing in all more than half a million running words. For every single word the analysis results in registration of the word class, the flexion form and the lemma to which the text word belongs.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Abstract",
                "sec_num": null
            }
        ],
        "body_text": [
            {
                "text": "If manually performed, this kind o f analysis would be enormously monotonic work requiring high precision. A method has been developed to perform the analysis to a great extend automatically using a computer. However the manual work can not be eliminated, but it has already been reduced significantly, and at the same time the character o f the manual work is altered to be mainly a matter of correcting activity.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "",
                "sec_num": null
            },
            {
                "text": "The method of automatic analysis is based on a corpus o f tags and word forms originating in a previous manually performed analysis o f more than 54,000 text words and on a set of rules for possible relations between words of the same sentence. On the basis o f frequencies compiled by the previous analysis and o f points given by the rules when fulfilled, a computer program automatically selects the probably 'best sentence' among a (usually great) number o f homograph sentences. Furthermore, in case o f words not found in the collection o f word forms, the program makes use o f a collection of more than 5,000 back parts of word forms in order to make an intelligent guess.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "",
                "sec_num": null
            },
            {
                "text": "At the current stage the result o f the automatic analysis is completely correct for about 70% o f the text words and partly correct for about 15%. Our experience shows that the manual effort is reduced by about 2/3. By extension of the word form collection and by improvement o f the relation rules and points giving, a significant improvement o f the automatic analysis is expected in the near future, e.g. leading to 85%-90% o f text words being correctly analysed.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "",
                "sec_num": null
            },
            {
                "text": "A u t o m a tis k m o r fo lo g is k a n a ly se a f isla n dsk tek st h \u00f8 re r til en a f L eksik ografisk b liv e m a n u elt u d f\u00f8 r t eller i h vert fa ld m a n u elt k on trolleret. M e n d e t har allerede v ist sig a t a r b e jd e t re d u cere s i b e ty d e lig g ra d v ed a t m an i f\u00f8 r s te o m g a n g u df\u00f8rer a n a ly sen a u to m a tis k v e d h jae lp a f en d a ta m a t. e m a n g e g a n g e p .g .a . forskelle i le k sik on sord o g forsk e llig e b \u00f8 jn in g s fo rm e r. T a llen e i a n d e n k o lo n n e er h y p p ig h e d sta l fr a d e n tid lig e re a n a ly se. ",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Indledning",
                "sec_num": "1"
            },
            {
                "text": "S Y S T E M E T S B E S T A N D D E L E",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "D en autom atiske hovedanedyse",
                "sec_num": "3.2"
            },
            {
                "text": "O R D F O R M E R : 421 1 v",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "D en autom atiske hovedanedyse",
                "sec_num": "3.2"
            },
            {
                "text": "R e su lta tet a f a n alysen a f en p r\u00f8 v e te k st p \u00e5 5 .0 15 % te k s to rd u k orrek t a n a ly sere t 15% te k s to rd slet ik ke a n a ly seret F oran alysen ta g er n ae sten in gen tid . H o v e d a n a ly se n ta g e r h eller ikk e la n g P \u00e5 g ru n d a f u reg elm ae ssig b r u g a f in te rp u n k tio n i isl^lndsk h a r m a n i den a u to m a tis k e a n a ly se h elt set b o r t fr a in te rp u n k tio n en , selv o m d en se lv f\u00f8 lg e lig i m a n g e tilfae ld e k u n n e g iv e v ae rd ifu ld e op ly sn in g e r. 1^9 . Ordabok H\u00e5sk\u00e5lans. Reykjavik.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Prae station og kvalitet",
                "sec_num": "3.3"
            },
            {
                "text": "Proceedings of NODALIDA 1989",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "",
                "sec_num": null
            }
        ],
        "back_matter": [],
        "bib_entries": {},
        "ref_entries": {
            "FIGREF0": {
                "num": null,
                "uris": null,
                "type_str": "figure",
                "text": "In s titu ts o p g a v e r. D e t d re je r sig o m en fa se a f et s t\u00f8 r r e p r o je k t, som er u dar b e jd e ls e a f en isla n d sk fr e k v e n s o r d b o g . B o g e n skal g iv e o p ly sn in g e r o m b ru g en a f isla n d sk n u tid s s p ro g , d et sk riftlig e s p ro g , i fo r m a f forsk ellig e slags oversigter o v e r h y p p ig h e d e n a f o r d , o r d fo r m e r , b \u00f8 jn in g s fo r m e r , o rd k la sser o .s .v . D e t kan g iv e en id \u00e9 o m p r o je k te t s o m fa n g a t d et v il o m fa tte ca . en h alv m illio n te k s to r d . L a n g t d en s t\u00f8 rs te d e l a f a rb e jd e t lig g e r i d en m o rfolog isk e a n a ly se , s o m b liv e r sae rlig o m h y g g e lig t u d f\u00f8 r t. I den sid ste en d e v il analysen"
            },
            "FIGREF1": {
                "num": null,
                "uris": null,
                "type_str": "figure",
                "text": "o rm \u00e5 le t m e d d en m o rfo lo g is k e a n a ly se u an set o m den bliv er u d f\u00f8 rt m an u elt eller m a s k in s t\u00f8 tte t er f\u00f8 lg e n d e ."
            },
            "FIGREF2": {
                "num": null,
                "uris": null,
                "type_str": "figure",
                "text": "o r d et f\u00f8 r s te skal a n a ly sen fo r h v ert en k elt te k sto rd s v ed k o m m e n d e f\u00f8 r e til re g istre rin g a f o rd k la sse o g b \u00f8 jn in g s fo r m . D e s u d e n registrerer m a n h vilk et kasus v e rb e r o g p r ae p o s itio n e r styrer. D e fle ste o r d , s o m m a n p le je r a t k lassificere so m p r ae p o s itio n e r , b ru g es i n ogen tilfae ld e s o m a d v e r b ie r , o g o m v e n d t, m a n g e o r d , s o m tr a d itio n e lt klassificeres s o m a d v e r b ie r , b ru g e s o g s \u00e5 s o m p r ae p o sitio n e r. I d e tte p r o je k t h ar m an derfor v a lg t a t b e h a n d le p r ae p o s itio n e r o g a d v e rb ie r u n d er \u00e9 t, h vilket m ed f\u00f8rer, at m an o g s \u00e5 reg istrerer a d v e rb ie rs kasu sstyrelse, n \u00e5r den fo re k o m m e r. E n d v id e r e registrerer m a n fo r h vert te k s to rd d et tilh \u00f8 re n d e lek sik on sord, o g s\u00e5 k a ld t le m m a . S o m e t e k se m p e l ta g e r v i f\u00f8 lg e n d e k o r te tek st o g d e t tilstrae b te resu ltat a f d en s a n a ly se:"
            },
            "FIGREF3": {
                "num": null,
                "uris": null,
                "type_str": "figure",
                "text": "d en m id terste k olon n e h ar v i te k sto rd e n e , \u00e9t i h ver lin ie, o g til h \u00f8 jr e d e tilh \u00f8 re n d e lek sik on sord . T il v e n stre h ar v i s \u00e5 teg n fo r d e g ra m m a tisk e o p ly s n in g e r so m a n alysen h ar f\u00f8 r t til. D e t f\u00f8 r s te te g n st\u00e5 r fo r ord k la sse. F o r h v ert te k sto rd er d er h \u00f8 js t 7 g ra m m a tisk e te g n i en b e s te m t rae k k ef\u00f8lg e . D e n kan m a n k alde en g ra m m a tisk stren g . E n b la n k lin ie b e te g n e r b e g y n d e ls e n a f en n y sae tn in g.3 Den automatiske analyse D en m e to d e , s o m m a n h er b e n y tte r til a u to m a tis k m o r fo lo g is k a n a ly se, er b a se ret p \u00e5 en tid lig e re m a n u elt u d f\u00f8 rt a n a ly se a f g o d t5 4 .0 0 0 te k s to r d (PriQrik M a g n u sson 1 988) o g fu n g erer v e d h jae lp a f et sae t a f m o r fo lo g is k e regler o g v e d b e tra g tn in g o g v u rd e rin g a f sa n d sy n lig h e d e r. H o v ed a n a ly sen u d f\u00f8 re s m e d \u00e9n sae tn in g a d g a n g e n . D e t m \u00e5 d e r fo r v ae re h elt k lart, h v or h ver sae tn in g b e g y n d e r . F o r a t o p n \u00e5 d e t u d f\u00f8 re s d e r en fora n a ly se so m har til fo rm \u00e5 l at registrere s ta rtfelte t fo r h ver sae tn in g ^lf d e n tek st, so m skal analyseres. Proceedings of NODALIDA 1989 ora n a ly sen u d f\u00f8res h a lv a u to m a tisk , d .v .s . i sa m a r b e jd e m ellem m en n esk e o g m a sk in . T E K S T : B E G Y N D E L S E S O R D ; had er l>ridjudagur i dag. Magnhs kemur \u00e5 morgun. Har m dvaldist \u00e5samt dr. Jdsteini Samuelssyni alllengi i Danmorku. F\u00e9lagi hans hef u r ordid eftir. P O I N T E R R E G I S T E R ; 0 < bad 26 # Magnhs 49 < Hann 115 < F\u00e9lagi b e g y n d e ls e s o r d -\u25ba # fo r e g e n n a v n e , ellers < sta rtfe lt i tek sten H er h a r v i d e n sa m m e tek st ig en . T il h \u00f8 jr e er v ist d e f\u00f8 rste 20 o rd i en sa m lin g cif 6 4 9 b e g y n d e ls e s o r d , o r d n e t e fte r h y p p ig h e d . D isse 20 o r d er a lts\u00e5 d e h y p p ig ste i b e g y n d e ls e n a f isla n dsk e sae tn in g e r, i h v ert fa ld i d en tekst so m sa m lin gen er b a se re t p \u00e5 . D e t p r o g r a m , s o m u d f\u00f8 re r fora n a ly se n , b ru g er d en n e sa m lin g til a u to m a tis k a t fin d e u d a f d e fleste sae tn in g s b e g y n d e lse r i en tek st. N \u00e5 r d er er tv iv l s to p p e r p r o g r a m m e t o p , in v ite re r m en n esk et til en a fg \u00f8 relse o g fo rtsae tte r n \u00e5 r en b e s lu tn in g er tru ffet. I d e tte e k se m p e l m \u00e5 et m en n esk e trae d e til fo r a t a fg \u00f8 re , at M a g n u s er et p r o p r iu m o g a t J \u00e5 stein n ikke er b e g y n d e ls e n a f en n y sae tn in g. 6 Proceedings of NODALIDA 1989 Stef\u00e5n Briem: Automatisk morfologisk analyse af islandsk tekst R e s u lta te t, et p o in te rre g iste r, b ru g e s s \u00e5 sa m m en m ed te k sten u n d er h o v e d a n alysen , d en e g e n tlig e a u to m a tisk e m o rfo lo g is k e a n a ly se , s o m v i nu g \u00e5 r o v e r til a t b e tra g te ."
            },
            "FIGREF4": {
                "num": null,
                "uris": null,
                "type_str": "figure",
                "text": "D e tte billed e v iser s y ste m e ts b e s ta n d d e le . I ce n tru m er et p r o g r a m , h v o ri d e r er in d b y g g e t et sae t a f 75 reg ler, isae r m o rfo lo g is k e reg ler. P r o g r a m m e t g \u00f8 r b r u g a f en sam lin g a f o r d fo r m e r o g en sa m lin g a f sid ste d e le a f o r d . D e t er p r o g r a m m e ts o p g a v e fo r h vert en k elt o r d i en g iv en te k st a t fin d e fre m til d e n m est sa n d sy n lig e stren g b la n d t d e 566 forsk ellig e g ra m m a tisk e s tre n g e , sa m t a t fin d e d e t tilh \u00f8 re n d e leksik on sord. V i v il nu b e tr a g te d e en k elte b e s ta n d d e le n ae rm ere. Proceedings of NODALIDA 1989Computational Linguistics -Reykjavik 1989 3.2.1 St\u00f8tteregistre L a d o s b e g y n d e m e d d e 3 s t\u00f8 tte r e g is tr e so m p r o g r a m m e t b ru g er u ae n d re d e fra tek st til tek st. D e t f\u00f8 r s te s t\u00f8 tte r e g is te r in d e h o ld e r 566 g ra m m a tisk e stren ge i a lfa b etisk rae k k ef\u00f8lg e. T a lle n e er h y p p ig h e d sta l. G R A M M A T I S K E S T R E N G E : a n d e t s t\u00f8 tte r e g is te r in d e h o ld e r k n a p 18.000 o rd fo r m e r i a lfa b e tisk rae kke f\u00f8 lg e sa m m e n m ed d e t tilh \u00f8 re n d e lek sik on sord o g lae n gst til v en stre n u m m eret p \u00e5 en g r a m m a tis k stren g . H v er o r d fo r m kan o p tr ae d"
            },
            "FIGREF5": {
                "num": null,
                "uris": null,
                "type_str": "figure",
                "text": "Automatisk morfologisk analyse af islandsk tekst O g d e t tre d je s t\u00f8 tte re g is te r in d e h o ld e r g o d t 5 .0 0 0 sid ste d e le a f o r d fo r m e r i ba g lae n s a lfa b e tisk rae k k e f\u00f8 lg e, h v er sa m m en m e d d en tilsv a ren d e sid ste d e l a f etlek sik on sord o g n u m m ere t p \u00e5 en g ra m m a tis k stren g . T a llen e i a n d e n k o lo n n e er sa n d sy n lig h e d sta l s o m m a n kan k n y tte til h ver sid ste d e l fo r a t s ty re p r o g r a m m e ts an vendelse a f d e tte register. os nu b e tr a g te d et reg elsae t so m er d irek te in d b y g g e t i p r o g r a m m e t. T il h ver regel er d er k n y tte t e t a n ta l p o in ts , s o m g iv es h ver g a n g reglen er o p fy ld t .P ro g ra m m e t a r b e jd e r m ed \u00e9n sae tn in g a d g a n g e n . D e t sl\u00e5r te k s to r d e n e o p i sa m lin gen a f o r d fo r m e r . I d e fleste tilfae ld e fin d es d e r n o g le m u lig h e d e r fo r hvert te k sto rd . D e t kan f\u00f8 r e til et s to r t a n ta l a f h o m o g r a fe sae tn in g er, d e r b liv e r ka n did a ter til stillin g en ' d e n b e d s te s ae tn in g ' . H v is te k s to r d e n e b e tr a g te s h ver fo r sig u anset deres stillin g i sae tn in g en , er d e t u m u lig t at a fg \u00f8 r e h vilk en a f d e h o m o g ra fe sae tn in ge r er d en rig tig e eller d e n b e d s te . M en h er trae d er reglern e til. F o r h ver en k elt a f d e h o m o g r a fe sae tn in g e r kalkulerer p ro g ra m m e t d e t to ta le a n ta l p o in ts s o m reg le rn e g iv e r h ver g a n g d e er o p fy ld t. D e n sae tn in g , s o m f\u00e5r d e fleste p o in ts , an ses fo r a t v ae r e d e n b e d s te o g b liv er va lgt so m an a ly sen s resu ltat. I p ra k sis kan a n ta llet a f h o m o g r a fe sae tn in g er b liv e s\u00e5 e n orm t, a t d e t ville ta g e d a ta m a te n m \u00e5 n e d e r eller e n d o g \u00e5 r at b e tr a g te d e m alle. M en d a a n tallet er k en d t tid lig t u n d er p ro ce s s e n , l\u00f8ses d e tte p r o b le m n em t v ed op tim a lise rin g a f k u n en d e l a f sae tn in g en a d g a n g e n . D e t m e d f\u00f8 re r en n ed sae ttelse a f an alysen s k o rre k th e d p \u00e5 ku n ca . 1% . D e fleste a f reg lern e er m eg et en k le o g kan d e r fo r n em t om sk rives til et p ro g r a m m e r in g s s p r o g . H er er v ist n o g e n f\u00e5 a f reg lern e i damsk oversae ttelse. T a lle n e e r d e tilh \u00f8 re n d e p o in ts . \u2022 H v is d e r e ftir e t fa ld s ty re n d e a d v e rb iu m f\u00f8 lg e r et ikke falds ty re n d e a d v e r b iu m o g d e r p \u00e5 f\u00f8 lg e r et faildb\u00f8jet o r d , s\u00e5 v il det fa J d b \u00f8jed e o r d s t\u00e5 i d et fa ld s o m d et f\u00f8rste a d v e rb iu m styrer. 4 0 0 \u2022 D e t er sa n d sy n lig t a t ad er en k o n ju n k tio n , hvis et u k en d t o rd f\u00f8 lg e r efter. 5 0 0 \u2022 H v is d er e fte r et aidjektiv f\u00f8 lg e r et su b sta n tiv , s\u00e5 har d e n ae sten a ltid sa m m e k \u00f8 n , ta l o g fa ld . 1 0 0 0 \u2022 H v is d e r e fte r e t a d je k tiv f\u00f8 lg e r et su b sta n tiv i b e k en d t fo r m , s\u00e5 er d e r s t\u00f8 r r e sa n d sy n lig h e d fo r a t a d je k tiv e t h ar b estem t fo rm en d u b e s te m t. 1 0 \u2022 E t v e r b u m i p e r fe k tp a r tic ip iu m er sa n d sy n lig t, hvis d et f\u00f8 lg e n d e o r d er v e r b e t v era eller hvis v e rb e t v era er et a f d e t o fo r a n st\u00e5 e n d e o r d . 2 0 0 K o n g r u e n s b \u00f8 jn in g er en a f d e stae rk este s t\u00f8 tte p ille r fo r den a u tom a tisk e a n a ly se . E t e k s e m p e l p \u00e5 k o n g r u e n s b \u00f8 jn in g h a r v i h er i tre d je regel i tilfae ld e a f su b s ta n tiv o g tilh \u00f8 r e n d e a d je k tiv . A n d r e a n a lo g e regler fo r k on g ru en sb \u00f8jn in g o m fa tte r o g s \u00e5 p ro n o m in e r o g ta lo rd ."
            },
            "FIGREF6": {
                "num": null,
                "uris": null,
                "type_str": "figure",
                "text": "o r r e k th e d e n a f d en a u to m a tis k e a n a ly se fre m g \u00e5 r a f sa m m e n lig n in g m ellem den p \u00e5 n ae ste sid e a n f\u00f8 rte o g d e n tid lig e re v iste k orrek te an a ly se a f sa m m e tek st. I d e t t e e k s e m p e l h a r p r o g r a m m e t b l.a . ta g e t fejl a f te k sto rd e t f \u00e9 l a g i . If\u00f8 lg e den a u to m a tis k e a n a ly se sk u lle d e t v ae re sin gu la ris d a tiv a f substam tivet f\u00e9 ia g som b e ty d e r fo r e n in g , m en i v irk e lig h e d e n d re je r d e t sig o m sin gu laris n o m in a tiv a f su b s ta n tiv e t f\u00e9la g i s o m b e ty d e r ka m m era t. L id t sen ere f\u00e5 r v i flere ek sem p ler p\u00e5 h o m o g r a f!. A U T O M A T I S K A N A L Y S E :"
            },
            "FIGREF8": {
                "num": null,
                "uris": null,
                "type_str": "figure",
                "text": "tid o g d et er j o d a ta m a te n s tid . S t\u00f8 rste d e le n a f a r b e jd s tid e n er s ta d ig v ae k d en m en n esk elige a r b e jd s tid s o m krae ves til k orrek tu rlae sn in g a f d en a u to m a tis k e an alyses resu ltat. K v a lite te n a f d en a u to m a tis k e a n a ly se p \u00e5 d e t n u v ae ren d e sta d iu m er v ist her i p r o c e n te r ."
            },
            "FIGREF9": {
                "num": null,
                "uris": null,
                "type_str": "figure",
                "text": "e s t\u00f8 r s te p r o b le m e r h id r\u00f8 re r fra f\u00f8 lg e n d e tre fa k torer: 1. M a n g e u k en d te te k s to rd 2. U reg elm ae ssig in te rp u n k tio n 3. H o m o g r a fe r b la n d t h y p p ig e o r d fo r m e r U k e n d te o r d m e d f\u00f8 r e r a t d e t b liv e r s v ae rt at a n alysere d e n ae rm est lig g en d e o r d k orrek t o g p rae cist."
            },
            "FIGREF10": {
                "num": null,
                "uris": null,
                "type_str": "figure",
                "text": "e g v il nu g iv e et p a r e k sem p ler p \u00e5 h o m o g ra fe r b la n d t h y p p ig e o rd fo rm e r, s o m tit b liv e r fe jla g tig t a n a ly seret u n d e r d en a u tom a tisk e analyse. o r d f o r m le k s ik o n s o r d dansk r vid p ro n . pers. 1. p. pi. n om . vi v i d < vid p r ae p ./a d v . ved 1 (v id u r sb . m a se. sg. akk. v ed ) O rd fo r m e n v i d h^lr tre h elt forsk ellig e m e n in g er. D e m est a lm in d elig e er p e r s o n lig t p ro n o m e n o g p r ae p o s itio n . D en tre d je er h er sat i p a ren teser, fo r d i den er ik k e n ae r s\u00e5 h y p p ig s o m d e a n d re; o g d en er fa k tisk slet ikke m ed i sam lin gen eif o r d fo r m e r . eu tr. p i. n o m ./a k k . bek . ord en e D e t o a n d re o r d fo r m e r , o r d i d o g o r d i n h ar h ver fo r sig t o h elt forsk ellige m en in g e r. D isse b e s te m te h o m o g r a fe r u d g \u00f8 r m \u00e5 sk e ikk e ret s to r e p ro b le m e r. M en der k rae v es i h v ert fa ld m e re p rae cise regler en d d e h id tid ig e til a t skelne i m ellem d em . skal v i b e tr a g te d e m u lig h ed er d e r g ives fo r a t fo r b e d r e d en a u to m a tis k e analyse. \u2022 S t\u00f8 rre k o rp u s, d .v .s . flere o r d fo r m e r o g flere s id sted ele \u2022 F le re o g m e re p rae cise regler \u2022 P rae cis e rin g a f p o in ts g iv n in g \u2022 U d n y ttelse a f in te rp u n k tio n D e n fo r b e d r in g so m m a n v en ter a t o p n \u00e5 u d en s t\u00f8 r r e b e s v ae r sk u lle f\u00f8 r e til 8 5 % -9 0 % k orrek t an alyse. Litteratur Magnusson, Fridrik. 1988. Hvad er titt? Jon Hilmar Jonsson [ed.]. I Ord og tunga 1:"
            }
        }
    }
}