Swedish and Danish

Se introduktion til Apertium på dansk her: Dansk introduktion.

Pressemeddelese

From: Jacob Nordfalk <jacob.nordfalk@gm...> - 2009-10-12 13:29 http://sourceforge.net/mailarchive/message.php?msg_name=20cf28cd0910120629o572ede0i13542ee2737f2deb%40mail.gmail.com

Første Open Source maskinoversættelse mellem svensk og dansk

First open source machine translation between Swedish and Danish

Dansk (english version below)

Vi har netop frigivet version 0.5 af svensk-dansk til open source maskinoversættessystemet Apertium.

Det er det første frie maskinoversættelsesystem mellem svensk og dansk.

Det kan allerede nu bruges fra http://apertium.org/, men forhåbentlig vil fællesskabet omkring fri software tage det til sig og snart gøre det tilgængeligt på bl.a. alle Linux-arbejdsstationer.

Til udviklingen har vi brugt et antal frit tilgængelige kilder, bl.a. open source stavekontrollen Aspell, Den stora svenska ordlistan, http://dsso.se og den svenske og danske Wikipedia og Wiktionary.

Udviklingen er sponsoreret af Google Summer of Code (GSOC) og foretaget af student Michael Kristensen. Mentorer på projektet er Francis Tyers (Universitat d'Alacant og Prompsit Language Engineering) og Jacob Nordfalk (Ingeniørhøjskolen i København).

For nærmere oplysninger om udviklingen af sprogparret, se http://wiki.apertium.org/wiki/Swedish_and_Danish

For mere information om apertium og GSOC, se http://socghop.appspot.com/org/home/google/gsoc2009/apertium.

Tekniske specifikationer

Svensk morfologisk ordbog       5.230 ordrødder
Tosproget ordbog                6.854 ordrødder
Dansk morfologisk ordbog       10.694 ordrødder

Dækningen på Wikipedia-tekst er p.t. 72% og korpuset Europarl 80%.

Vi anvender 1-trins "shallow transfer" med 17 transferregler.

Vi har foretaget en sammenlignende vurdering med andre tilgængelige maskinoversættelsessystemer på 65 sætninger fra Wikipedia.

Resultaterne findes nedenfor (lavest tal er bedst)

System     Edit distance      WER (Word Error Rate)
Apertium    353                31 %
Gramtrans   308                26 %
Google      415                35 %

Yderligere oplysninger kan findes i artiklen "Shallow-transfer rule-based machine translation for Swedish to Danish" som vi vil præsentere på First International Workshop on Free/Open-Source Rule-Based Machine Translation (http://xixona.dlsi.ua.es/freerbmt09/).

For mere information, kontakt Jacob Nordfalk, Ingeniørhøjskolen i København (jano@ih...), telefon 26206512.

English

A new language pair, Swedish-Danish, has been released for the the free and open-source Apertium machine translator engine.

It's the first open source machine translator for Swedish and Danish.

The pair is immediately available for testing at http://apertium.org/, but will hopefully adopted by the free-software community and be available on i.a. the Linux desktop.

In developing this system, we used a number of freely available sources of information for constructing the system, i.a. high coverage spell-checkers available in the aspell project, Den stora svenska ordlistan, http://dsso.se and the Swedish and Danish Wikipedia and Wiktionaries.

This language pair was developed as part of a Google Summer of Code (GsoC) project by Michael Kristensen, mentored by Francis Tyers (Universitat d'Alacant and Prompsit Language Engineering) and Jacob Nordfalk (Ingeniørhøjskolen i København). For more information on Apertium and GsoC, see http://socghop.appspot.com/org/home/google/gsoc2009/apertium .

Many thanks to Thyge Larsen for his assistance with post-edition and evaluation.

For more details on development and the language pair, see http://wiki.apertium.org/wiki/Swedish_and_Danish

Technical details

Swedish monolingual dictionary           5,230 lemmas
Bilingual dictionary                     6,854 lemmas
Danish monolingual dictionary           10,694 lemmas

We measured coverage on Wikipedia to 72 % and the EuroParl corpus to 80 %.

The system used 1-stage shallow transfer with 17 transfer rules.

We have made a comparative evaluation to other available MT systems. The results for 65 Wikipedia sentences can be found below

System     Edit distance      WER (Word Error Rate)
Apertium    353                31 %
Gramtrans   308                26 %
Google      415                35 %

Further details can be found in the article "Shallow-transfer rule-based machine translation for Swedish to Danish" to be presented during the First International Workshop on Free/Open-Source Rule-Based Machine Translation (http://xixona.dlsi.ua.es/freerbmt09/).

For more information, pls. contact Jacob Nordfalk, Ingeniørhøjskolen i København (jano@ih...), phone 26206512.

-- 
Jacob Nordfalk
एस्पेरान्तो के हो? http://www.esperanto.org.np/.
Memoraĵoj de KEF -. http://kef.saluton.dk/memorajoj/

Swedish and Danish

Swedish and Danish are closely related languages. Their differences are mainly found on the morphological level, the main lexicon is identical (or rather, very similar, with systematic differences), and the syntax is very similar. There are some differences, though.

Syntax

Particle order

Swedish keeps the verb together with a conjoined adverbial particle, where Danish separates them.

(sv) Vill du köra in bilen

(da) Vil du køre bilen ind

Swedish moves the reflexive pronoun sig along with the verb to V2 position, where Danish leaves it behind:

(sv) I går tvättade sig Peter äntligen

(da) I går vaskede Peter sig endelig

NP structure

Danish and Swedish have different NP patterns.

(sv) Vita huset

(da) Det hvide hus

(nb) Det hvite huset

(nn) Det kvite huset

In most NPs, Swedish has both the determiner den and the definite form of the noun. Danish, as always, cannot have both. Here, nn patterns with sv and nb with both sv and da (beware of non-idiomatic da, sv word choices, but the patterns are correct).

(sv) Den stora utmaningen är att göra det rätta. Utmaningen er svår.

(da) Den store udfordring er at gøre det rette. Udfordringen er vanskelig.

Existential sentences

Swedish can use "det" as an equivalent to the English "there", where Danish prefers "der",

(sv) Det kommer en bil

(da) Der kommer en bil

Relative clauses

In N + RC constructions, where the relativised constituent is subject, Danish uses either som or der as relativiser, whereas Swedish has som:

(da) manden som er her (the man who is here)

(da) manden der er her (the man who is here)

(sv) mannen som är här

When the relativised constituent is the object, on the other hand, the relativiser must be som, also in Danish:

(da) manden som jeg så (the man who I saw)

(sv) mannen som jag såg (the man who I saw)

Passives

(sv) Ytterligare prov kommer att tas under måndagen. (further test will be taken some time Monday)

(da) Yderligere prøve vil blive taget i løbet af mandagen

Grammatical words

Modal verbs

Danish and Swedish use more or less the same set of modal verbs, but with different meaning.

(allow) sv: Man får inte röka här 
        da: Man må  ikke ryge her 
        en: one is not allowed to smoke here

Some verbs also take different modals,

Swedish	Danish	Modal	Gloss	Example
åka	tage	ha → være	to go
föra	føre	ha → være	to take	Två personer har förts → To personer er ført
komma	komme	ha → være	to come	Min fru har kommit → Min kone er kommet hjem

There is a list of the most frequent 250 verbs with the modal they take here.

Resources

• http://spraakbanken.gu.se/sal/eng/ -- GPL morph. for Swedish
• http://w3.msi.vxu.se/~nivre/research/Talbanken05.html (A 300,000-word tree-bank: it is in XML, all words are nicely tagged with PAROLE-style tags, and it should be easy to build a morphological analyser and a PoS tagger from it; authors are likely be happy to let us use it if we cite them).
• http://www.isv.cbs.dk/~mbk/treebank/ (Danish tree bank, 100,000-word, as above, under the GPL)
• http://www.ling.su.se/staff/sofia/suc/suc.html (Stockholm Umeå Corpus: 1,000,000 Swedish words, tagged; a license has to be granted by authors - it was used for apertium-sv-da)

• http://www.woxikon.se Ordbok for svenska<->engelsk, tysk, nederlandsk...
• http://ordbok.nada.kth.se/ "Tvärslå är en nordisk ordbok bestående av många sammanslagna ordböcker"
• http://www.klid.dk/dansk/ordlister/samling.html - Kelds samling a resurser

See also

• /Pending tests
• /Regression tests
• http://gramtrans.com/ - state of the art MT between Swedish and Danish (closed source)

Further reading

• LUNDIN AKESSON Katarina (2003) "Constructions with låta, LET, reflexives and passive-s: a comment on some differences, similarities and related phenomena". Working papers in Scandinavian syntax ISSN 1100-097X
• TYERS, Francis M.; NORDFALK, Jacob. "Shallow-transfer rule-based machine translation for Swedish to Danish". En: Proceedings of the First International Workshop on Free/Open-Source Rule-Based Machine Translation / Edited by Juan Antonio Pérez-Ortiz, Felipe Sánchez-Martínez, Francis M. Tyers. Alicante : Universidad de Alicante. Departamento de Lenguajes y Sistemas Informáticos, 2009, pp. 27-33