Your SlideShare is downloading. ×
MOLTO poster for ACL 2010, Uppsala Sweden
Upcoming SlideShare
Loading in...5
×

Thanks for flagging this SlideShare!

Oops! An error has occurred.

×
Saving this for later? Get the SlideShare app to save on your phone or tablet. Read anywhere, anytime – even offline.
Text the download link to your phone
Standard text messaging rates apply

MOLTO poster for ACL 2010, Uppsala Sweden

62

Published on

MOLTO is funded by the European Union Seventh Framework Programme (FP7/2007-2013) under grant agreement FP7-ICT-247914. MOLTO's goal is to develop a set of tools for translating texts between multiple …

MOLTO is funded by the European Union Seventh Framework Programme (FP7/2007-2013) under grant agreement FP7-ICT-247914. MOLTO's goal is to develop a set of tools for translating texts between multiple languages in real time with high quality. Languages are separate modules in the tool and can be varied; prototypes covering a majority of the EU's 23 official languages will be built.

http://molto-project.eu

Published in: Education
0 Comments
0 Likes
Statistics
Notes
  • Be the first to comment

  • Be the first to like this

No Downloads
Views
Total Views
62
On Slideshare
0
From Embeds
0
Number of Embeds
0
Actions
Shares
0
Downloads
0
Comments
0
Likes
0
Embeds 0
No embeds

Report content
Flagged as inappropriate Flag as inappropriate
Flag as inappropriate

Select your reason for flagging this presentation as inappropriate.

Cancel
No notes for slide

Transcript

  • 1. MOLTO’s goal is to develop tools for web content providers to translate texts between multiplelanguages in real time with high quality. Languages are separate modules in the tool and can be varied;prototypes covering a majority of the EU’s 23 official languages will be built.FP7- 247914 molto-project.eunon multa, sed multumM LTOOTools for Multilingual Grammar-Based Translation on the WebGrammar Engineer’s ToolsBuilding a multilingual translation system amounts to building amultilingual GF grammar, namely:✤ a language-independent abstract syntax;✤ for each language, a concrete syntax mapping abstractsyntax trees to strings in that language.Abstract syntax construction is an extra task compared to othertranslation methods, but it is technically relatively simple, andgets amortized as the system is extended to new languages.Concrete syntax construction can be much more demanding interms of programming skills and linguistic knowledge and GFeases it by two main assets:✤ Programming language support: GF is a modern functionalprogramming language, with a powerful type system andmodule system allowing collaborative programming andreuse of code.✤ RGL, the GF Resource Grammar Library, implementing thebasic linguistic details of languages: inflectional morphologyand syntactic combination functions. The RGL covers fifteenlanguages at the moment.To give an example, let us consider the inflectional morphology. Itis presented as a set of lexicon-building functions such as, inEnglish,! mkV : Str -> Vi.e. function mkV, which takes a string (Str) as its argument andreturns a verb (V) as its value. The verb is, internally, an inflectiontable containing all forms of a verb. The function mkV derives allthese forms from its argument string, which is the infinitive form.It predicts all regular variations: (mkV "walk") yields thepurely agglutinative forms walk-walks-walked-walked-walking whereas (mkV "cry") gives cry-cries-cried-cried-crying, and so on. For irregular English verbs, RGLgives a three-argument function taking forms such assing,sang,sung, but it also has a fairly complete lexicon ofirregular verbs, so that the normal application programmer whobuilds a lexicon only needs the regular mkV function.Typically, most of the effort in writing a concrete syntax goes intoextending a lexicon with domain-specific vocabulary. RGL’sinflection functions are designed as “intelligent” as possible andthereby ease the work of authors unaware of morphology. Forinstance, even Finnish, whose verbs have hundreds of forms andare conjugated in accordance with around 50 conjugations, has aone argument function mkV that yields the correct inflection tablefor 90% of Finnish verbs.As an example of a syntactic combination function of RGL,consider a function for predication with two place adjectives. Thisfunction takes three arguments: a two-place adjective, a subjectnoun phrase, and a complement noun phrase. It returns asentence as value:pred : A2 -> NP -> NP -> SIt is available in all languages of RGL, even though the details ofsentence formation differ vastly. Thus, the concrete syntaxes ofthe abstract (semantical) predicate:! div x y (”x is divisible by y”),are, for English and German:div x y = pred (mkA2 "divisible" "by") x ydiv x y = pred (mkA2 "teilbar" durch_Prep) x yThe German generates a much more complex structure: thecomplement preposition durch Prep takes care of rendering theargument y in the accusative case, and the sentence produced hasthree forms, as needed in grammatically different positions:! x ist teilbar durch y (in main clauses)! ist x teilbar durch y (after adverbs)! x durch y teilbar ist (in subordinate clauses).The translation equivalents in a multilingual grammar need notuse the same syntactic combinations in different languages. Forinstance, the transitive verb construction y delar x (literally, ”ydivides x”), in Swedish, can be expressed using the transitive verbpredication function of the RGL and switching the subject andobject:div x y = pred (mkV2 "dela") y xThus, even though GF translation is interlingua-based, there is acomponent of transfer between English and Swedish. But thistransfer is performed at compile time. In general, the use of thelarge-coverage RGL as a library for restricted grammars is calledgrammar specialization. The way GF performs grammarspecialization is based on techniques for optimizing functionalprogramming languages, in particular partial evaluation.Translator’s ToolsFor the translator’s tools, there are three different use cases:✦ restricted source‣ production of new source‣ modification/editing of source✦ unrestricted sourceThe translating tool can easily handle a restricted source languagerecognizable by a GF grammar, except when there is ambiguity inthe text. Authoring within the restricted language is helped bypredictive parsing, a technique recently developed for GF.Incremental parsing yields grammatically correct wordpredictions, sensitive to the context, which guide the author in away similar to the T9 method in mobile phones.The author has started a sentence as la femme qui remplitle formulaire est co (”the woman who fills the form isco”), and a menu shows a list of words beginning with co that aregiven in the French grammar and possible in the context at hand;all these words are adjectives in the feminine form. Notice howthis is difficult for n-gram-based statistical translators: theadjective is so far from the subject with which it agrees that itcannot easily be related to it.Predictive parsing is a good way to help users producetranslatable content in the first place. When modifying the contentlater it may not be optimal, in particular if the text is long. Thetext can contain parts that depend on each other but are locatedfar apart. For instance, if the word femme (”woman”) is changedto homme, the preceding article la has to be changed to l’, andthe adjective connue (”known”) would become connu, and soon. Such changes are notoriously difficult and can easily leave adocument in an inconsistent state. In the GF syntax editor,changes can be performed on the abstract syntax tree and arepropagated to the concrete syntax strings, that automatically obeyall the agreement rules.Pred known A(Rel woman N (Compl fill V2 form N))! the woman who fills the form is knownla femme qui remplit le formulaire est connuePred known A (Rel man N (Compl fill V2 form N))! the man who fills the form is knownl’ homme qui remplit le formulaire est connuGF - Grammatical FrameworkThe core translation tools are based on the GF system, a grammarformalism that is, a mathematical model of natural language. It isequipped with a formal notation for writing grammars and withcomputer programs implementing parsing and generation thatare declaratively defined by grammars. The novel feature of GF isthe notion of multilingual grammars, used to describe severallanguages simultaneously by a common representation calledabstract syntax. The abstract syntax enables meaning-preservingtranslation as a composition of parsing and generation. Thus, GFtranslation scales up linearly to new languages without thequadratic blowup of transfer-based systems: n + 1 componentsare sufficient to cover n languages, because the mappings fromthe abstract syntax to each language are usable for bothgeneration and parsing.Multilingual GF grammars can be seen as an implementation ofCurry’s distinction between tectogrammatical (abstract syntax),defined by using a logical framework, whose mathematical basisis in the type theory of Martin-Löf, and phenogrammaticalstructure.GF improves over state-of-the-art grammar-based translationmethods:❖ the translation interlingua is a powerful logical formalism,able to express the finest semantical structures such ascontextdependencies and anaphora. In particular, it is moreexpressive than the simple type theory used in Montaguegrammar and employed in the Rosetta translation project.❖ GF uses a framework for interlinguas, rather than oneuniversal interlingua, making it more light-weight andfeasible than systems based on one universal interlingua,such as Rosetta and UNL, Universal Networking Language.It also gives more precision to special-purpose translation:the interlingua of a GF translation system can encodeprecisely structures and distinctions relevant to the task.Thus an interlingua for mathematical proofs is differentfrom one for commands for operating an MP3 player.One important source of inspiration for GF was the WYSIWYMsystem, which used domain-specific interlinguas and producedexcellent quality in multilingual generation. However thegeneration components were hard-coded in the program, insteadof being defined declaratively as in GF, and they were not usablein the direction of parsing.GF is open source and available, for major platforms, fromwww.grammaticalframework.org, licensed under GPL (theprogram) and LGPL (the libraries).

×