THE QUALITY OF MACHINE TRANSLATION: A CRITICAL APPRAISAL OF SYSTRAN’S HANDLING OF DUTCH Dirk Geens
Hogeschool-Universiteit Brussel Multiculturalism and thus translation has become part of our lives, very much the same way that makes it hard for man to conceive urban life without, for example, electricity. And Machine Translation (MT) can be seen as the industrialized manifestation of man’s linguistic ingenuity. Machine translation is presented on the internet in trial versions and a few hundred dollars will buy a CD-Rom with alleged easy to use, professional translation capabilities in quite regularly over 40 languages. Search bots like Google offer instant (machine) translation of websites in 25 different languages. Such omnipresence of MT is a clear sign of the demand our multilingual societies have for translation, for it is on the mountains one has to look back to understand the need for quick, inexpensive and yet adequate translations. Not the Alps nor the Pyrenees obviously, but the piles of books, papers, reports, electronic documents and the like which are awaiting global disclosure. Even though all major players in society, the industry, the administration, the educational system, and many others are constantly in want of translation, MT gives the impression to plod along in order to live up to these expectations. And not entirely without reason; certainly MT meets our need of speed and therefore will supply our demand for quantity as well. The price of machine translation is not the problem either. Precarious translation quality is the main reason why MT is fighting a battle for popularity it can hardly win. As it is widely accepted these days that machine translation is a skill different from human translation, using different methods and techniques, and aiming at different text domains and applications, it should not even be particularly disturbing that MT does not meet exactly the same standards as human translation. Human translators embrace translation technology as a tool they can mobilize to increase their efficiency. Several technologies, such as terminology management systems and translation memories have earned an incontestable place on the translators’ desk. A better quality of MT might only add to the productivity of translators. The quality of machine translation was already at issue in its early years; it was one of the fundamental shortcomings according to the ALPAC report [1966]. Although their appreciation nowadays is considered as biased, many of their recommendations have been very influential in computational linguistics for many years. And quite rightly so. Fundamental research into machine translation has led to considerable improvements, better translation
strategies, promising combinations of MT with CAT-tools, and by and large a better understanding of the mechanisms governing natural language.
1. SYSTRAN1 Yet it is a simple fact that from the point of view of practical machine translation, the best overall performance available is turned out by MTsystems like Google Translate and Systran, not purely by chance the best known operational systems. It is amazing and disturbing at the same time that a system dating back to the 1970’s like Systran can be a market leader in an area characterized by constant change and evolution. Theorists notice that these MT-systems share at best a very blurred relationship with modern linguistic insights, and thus regularly produce translations which are terribly defective. Admittedly Systran carries the burden of its age. It is not “State-ofthe-Art” machine translation, and therefore the motivation for this study is to be found in the huge popularity it enjoys among the major users of machine translation: large, important corporations, government institutions and the military Bad translations happen, but they are not a Systran trademark. In fact, within the bounds of certain domains like business, administration, finance, … the English-French language pair for example performs surprisingly well. Not a match for human translation, but quite acceptable translations with a high degree of fidelity and precision. From a purely practical point of view this is what makes Systran an interesting object of study: a reasonable quality of service with popular language pairs may account for its popularity on the market. Its commercial versions have 13 languages, which are combined in 52 language pairs. It is however, common knowledge that the quality of service (QoS) is very heterogeneous, i.e. that certain language pairs generate quite reliable translations, whilst other combinations produce notoriously poor translations. It should not come as a surprise that language pairs with English as their source language usually generate the best translations. The translation quality of language pairs with Oriental languages was not evaluated in this study, although informants are known to have expressed their disappointment with the quality of the language generated. The language pairs which interest Flemish and Dutch users most, i.e. the pairs with English or French and Dutch very often produce a poor output. For this reason a thorough examination of the treatment of Dutch in the Systran MT-system is a compelling task. The present study might be seen as a preamble to such a project. In its present form it is meant as a preparatory The Systran company was founded in 1968 by Peter Thoma, with roots going back to GAT. In 1972 Systran MT was thrown on the market. It was acquired by the European Commission in 1976. 1
study to examine the feasibility of a Systran-based MT-system for Dutch in combination with several Oriental languages.
2. The Texts Seven texts2 from journalistic sources were chosen randomly for this study: 2,620 words in all. Three texts were selected from De Tijd, a business newspaper, quite popular in management circles, with 1064 words altogether. Three texts, 504 words, came from Gazet van Antwerpen, a general interest newspaper with a more popular approach of business topics. Finally, one article selected from KNACK, a mainstream newsmagazine contributed 1,052 words. In a first round a raw translation of the unedited texts into English was generated by Systran. The first five paragraphs of each output translation was then submitted to a thorough error analysis.
3. Error Analysis It should be pointed out first of all that the user of Systran has the choice between five elementary techniques to remedy the performance of the translation system: 1. 2. 3. 4. 5.
Adding missing words and idioms to the personal dictionary Selecting and ordering the (topical) dictionaries to be used Adding phrases to the translation memory Pre-editing the source text Post-editing the target text
It is obvious that making an appropriate choice between these methods is essential to improve the overall performance of Systran in the user’s translation environment, i.e. any generalized solution, e.g. adding keywords to the personal dictionary, is to be preferred to ad hoc resolutions, e.g. preediting the source.
ABX: een bonafide verkoop of de grote treinroof?, KNACK, 29th February 2008 Belgacom TV groeit sneller dan Telenet-tv, De Tijd, 29th February 2008 België hinkt achterop bij ontwikkeling van windmolens, Gazet van Antwerpen, 29th February 2008 Groene lening pak duurder dan hypothecaire lening, Gazet van Antwerpen, 29th February 2008 Triodos absoluut toonbeeld duurzaam bankieren, De Tijd, 27th February 2008 Unizo veroordeelt doorgerekende Elia-heffing op energiefactuur, Gazet van Antwerpen, 29th February 2008 Wie Umicore wil overnemen zal een forse premie moeten betalen, De Tijd, 29th February 2008 2
Using the translation memory, pre-editing and post-editing have to be considered essentially as ad hoc techniques. That is, although such techniques may be helpful to arrive at a useful rendering of the source text, they do not provide a general solution for a translation problem. Any ad hoc solution will only pertain to one particular text, and has no effect on future translations of similar texts, with the exception perhaps on some occasions of items included in the translation memory. These may, however, be disappointingly rare. For machine translation to be successful, it is first and foremost necessary that all the words in the source text are understood correctly. Vocabulary errors will therefore be a category which is thoroughly scrutinized in the error analysis. Other categories of errors which will be distinguished concern: syntax errors both in the source and target languages and translation errors. However, in order to draft a useful inventory of the syntax and translation problems in Systran, it is important to get the vocabulary right, as dictionary imperfections will almost inevitably lead to analysis failure and translation flaws. Furthermore, it should be noted that the user of Systran in no way can interfere with the syntactic analysis process.
4. The Vocabulary Consequently the error analysis performed during the first round focussed mainly on vocabulary errors: 225 errors could be discerned, of which almost 50 %, 99 cases in all, were due to missing dictionary items or flawed, inappropriate usage of the information in the dictionary, as shown in Table 1. SOURCE TEXT
TRANSLATION
DICTIONARY SUGGESTION
dat is ruim 165000 abonnees goed anderhalf jaar later … dat Belgacom TV actief is in heel België
that is wide 165000 abonnees goods a year and a half latter … that Belgacom TV assets are in complete Belgium
about 165000 subscribers about a year and a half later
maar daartegenover stond dan weer …
but resisted on the other hand then
again
active
… dat de aandeelhouders niet op eender welk bod zullen ingaan
… and does not apply to already their activities … the shareholders to that not on alike which offer will enter
Belgacom TV wil dit jaar minstens even veel nieuwe kijkers winnen als in 2007
Belgacom TV want win this year at least even much new kijkers such as in 2007
as
Belgacom TV wil dit jaar minstens even veel nieuwe kijkers winnen als in 2007
Belgacom TV want win this year at least even much new kijkers such as in 2007
as many
zo snel mogelijk
as soon as possible
as quickly as possible
ik heb 's avonds even goed met mijn vrouw Valentijn gevierd als anders
I at night even goods with my woman have celebrated Valentijn as the other
as usual
… en niet voor al hun activiteiten gelden
all any
Belgacom trok het afgelopen jaar bijna 166000 nieuwe klanten aan bovendien kan er met een LOA niet geprofiteerd worden van … … en daarvoor geld lenen, … die toenmalig NMBS-baas Karel Vinck meekreeg …
Belgacom TV drew new customers last year almost 166000 to moreover cannot become with LOA profited there of … … and for that money lend, … which then NMBS-baas got Karel Vinck …
de opsplitsing over de vier kwartalen
the opsplitsing over the four trimesters
breakdown
de overnemer was … oplaadbare batterijen als je de CO2-uitstoot wil beperken het concurrerend digitaal tv-product wat de andere onderzochte banken betreft, toont de enquête …
the take over was … oplaadbare batteries if you want CO2-uitstoot limit clashing digital tv-product to concerns, the survey shows the other examined bancs …
buyer chargeable CO2-exhaust compete
maar … overwegen nu opnieuw centrales te bouwen of open te houden
but … now again building call centers or open to hold
consider
… na het afstoten van de koper- … en zinkactiviteiten …
… after ricocheting the purchasers … and zinc activities …
copper
dat zegt Philippe Maystadt … in De Tijd
that says Philippe Maystadt … in the time
De Tijd
berekeningen van de redactie bevestigen calculations of the redactie confirm that dat dat meldt de ondernemersorganisatie that communicates Friday to the vrijdag entrepeneur organization
attract benefit from borrow boss of the NMBS
concerning
editorial board employers' association
een verklaring die … door logistieke experts … werd aangehoord België hinkt achterop in vergelijking met … hinkt onze industrie achterop
a declaration which was … listened to by logistical loss assessors … Belgium limps at the back in comparison with … our industry limps at the back
Test-Aankoop vindt dat vooral jammer
Test-Aankoop find that especially moan
feel sorry
… Maystadt see a little belgian proposals
few
… ziet Maystadt weinig Belgische aanvragen … dat de koers intussen grotendeels hersteld is … dat de koers intussen grotendeels hersteld is de forse koersval … baart Leysen niet echt zorgen [Chastel] stelde in de Kamer van Volksvertegenwoordigers de verkoop ter discussie de minderinkomsten ook vindt de zelfstandigenorganisatie dat … een lening op afbetaling de intercommunales willen zo de door hen voorgefinancierde Elia-heffing recupereren, …
… that the quotation has meanwhile mainly repaired … that the quotation has meanwhile mainly repaired the strong quotation fall … really does not bear Leysen looks [Chastel] brought up in for discussion the chamber of members of parliament sell the less income also the self-employed person organization finds that … a loan on payment
expert fall behind fall behind
for the most part recover give cause for concern in the Lower House income reduction independents' association installment loan
the intercomunales want recuperate this way Elia-heffing prefinanced by them, …
intercommunals
de transformatie naar een specialist in …
the transformation to a specialist in ...
into
… dat de aandeelhouders niet op eender welk bod zullen ingaan net even veel
… the shareholders to that not on alike which offer will enter network even much
jump at just as much/many
ik heb 's avonds even goed met mijn vrouw Valentijn gevierd als anders
I at night even goods with my woman have celebrated Valentijn as the other
maar … overwegen nu opnieuw centrales but … now again building call centers or te bouwen of open te houden open to hold
just as well keep open
die splitsing kom er …
that junction bowl there …
komt
… zal een forse premie moeten betalen
… a strong premium must pay
large
Unizo condemn calculated Elia-heffing …
levy
Unizo condemn calculated Elia-heffing …
recharged
Elia-heffing at a moment that but still limits … yield have a charge lowering
levy
Unizo veroordeelt doorgerekende Eliaheffing … Unizo veroordeelt doorgerekende Eliaheffing … de Elia-heffing op een moment dat … nog maar een beperkt rendement hebben een lastenverlaging dat zegt gedelegeerd bestuurder Thomas Leysen in een gesprek met De Tijd
limited lowering of the charges
that says delegate manager Leysen in an interview with the time to Thomas
managing director
Umicore heeft over die acht jaar een grote metamorfose achter de rug …
Umicore have which eight years large a metamorfose behind the back …
metamorphosis
de gewezen minister van Financiën absoluut toonbeeld
the former minister of finances absolutely tone picture
Minister of Finance model
in Duitsland … heeft men veel sneller fiscale voordelen toegekend …
in Germany … one has granted many more rapidly fiscal advantages …
much more rapidly
dat merkt … Dirk Saelens van KBC Securities op in zijn rapport
that notices … Dirk Saelens van KBC Securities in its report
observe
anderzijds is de rentevoet hoger dan bij een hypothecair krediet, …
on the other side the interest rate is superior than at a mortgage credit, …
on the other hand
er zijn maar weinig Belgische bedrijven … but there are a little belgian companies …
only a few
in Europa en erbuiten.
in Europa and.
outside
Umicore heeft over die acht jaar een grote metamorfose achter de rug …
Umicore have which eight years large a metamorfose behind the back …
over 8 years
ING doet het iets beter dan de andere grootbanken
ING do it something improve then the other large bancs
perform
maar … overwegen nu opnieuw centrales te bouwen of open te houden
but … now again building call centers or open to hold
power stations
… die behoren net tot de belangrijkste domeinen … afgeschilderd als het grote zorgenkind van de NMBS
… those belong network to most important the fields depicted … as large a care child of the NMBS the construction of purification railway stations … after ricocheting the purchasers … and zinc activities … the last three trimesters … for cultivating … house owners who decide to seize verbouwingswerken for verbouwingen
de bouw van zuiveringsstations … na het afstoten van de koper- … en zinkactiviteiten … de laatste drie kwartalen … voor het verbouwen … huiseigenaars die beslissen om verbouwingswerken aan te vatten … voor verbouwingen
precisely problem child purification stations push off quarter rebuilding rebuilding rebuilding
op een moment dat … nog maar een beperkt rendement hebben … pakweg 6,0 procent over de verkoop na de verzesvoudiging zo antwoordde Bank van de Post als enige niet op de vragenlijst van TestAankoop zonne-energie
at a moment that but still limits … yield have … pakweg 6.0 percent concerning sell after the verzesvoudiging thus as some on the questionnaire of Test-Aankoop did not answer banc of the mail zonne-energie
kernenergie was wat weggezakt
nuclear power was what considers sunk
somewhat put aside
dat was zowat de richtlijn die splitsing kom er … Telenet- woordvoerder Stefan Coenjaerts stelt … dat … en heeft vadertje staat aan het eind van de rit 29 procent van de verbouwing gesponsord Leysen zal een hoog bod op zijn bedrijf bestuderen, … dat is ruim 165000 abonnees van zijn voogdijminister ...
that was zowat the directive that junction bowl there … Telenet- woordvoerder Stefan Coenjaerts state … that … and has to vadertje state at the end of the ride 29 per cent of the verbouwing is sponsored Leysen high an offer on its company observate, … that is wide 165000 abonnees of its protection minister …
sort of split
… dat de federale regering de meerinkomsten … moet aanwenden …
… that the federal government the meerinkomsten … must apply …
surplus income
Chastel had het toen al over …
Chastel had it then already concerning
talk about
van de telecomgroep stootte de NMBS-holding … ABX logistics af
of the telecomgroep stootte NMBS-holding… ABX Logistics finished
telecomgroup
Umicore heeft over die acht jaar een grote metamorfose achter de rug …
Umicore have which eight years large a metamorfose behind the back …
to get over
Jannie Haek … leidde de operatie in
Jannie Haek … direct the operation in
to initiate
ABX Logistics … are again offered buy
to offer for sale
the commitment … determines in itself costs to take
to take sth o.oneself
Belgacom TV wil dit jaar minstens even veel nieuwe kijkers winnen als in 2007
Belgacom TV want win this year at least even much new kijkers such as in 2007
to want to
de grote treinroof
large the train treinroof
train robbery
een onding aangekocht door Karel Vincks voorganger, Etienne Schouppe.
a onding Karel bought Vincks predecessor, Etienne Schouppe.
trash
meer nieuwe digitale kijkers zo antwoordde Bank van de Post als enige niet op de vragenlijst van TestAankoop
more new digital kijkers thus as some on the questionnaire of Test-Aankoop did not answer banc of the mail
ik heb 's avonds even goed met mijn vrouw Valentijn gevierd als anders
I at night even goods with my woman have celebrated Valentijn as the other
wordt ABX Logistics … opnieuw te koop aangeboden het engagement … om bepaalde kosten op zich te nemen
22 Dutch words are not recognized by the dictionaries used:
return roughly sale sextupling so solar power
spokesman state study subscribers supervising minister
to drop
viewer was the only one wife
abonnees CO2-uitstoot Elia-heffing intercommunales kijkers meerinkomsten
metamorfose NMBS-baas onding oplaadbare opsplitsing overwegen
pakweg redactie telecomgroep treinroof verbouwingen verbouwingswerken
verzesvoudiging woordvoerder zonne-energie zowat
Dictionaries are per definition incomplete. Supplying the missing terms is a standard procedure and should in many cases resolve the problem. Notably more disturbing are source language terms covered in the dictionaries but connected with interpretations, c.q. translations which do not fit within a given context. In some instances it may be preferable to treat them as idiomatic phrases. In these cases the idiomatic expression will be entered as such in the personal dictionary, or in the translation memory. A number of fixed expressions, metaphors, compounds and creative combinations are misinterpreted by the dictionary procedures in Systran; these errors may be easily avoided by adding the appropriate translation to the dictionary. The mere fact that some of these cases are extremely obvious, seems to suggest that the performance of the Dutch language module is absolutely no priority for the Systran team. aantrekken (to draw to) als anders (as the other) anderzijds (on the other side) bepaalde kosten op zich nemen (determines in itself costs to take) bestuderen, (observate) centrales (call centers) de grote treinroof (large the train treinroof) even goed (even goods) even goed (even goods) gedelegeerd bestuurder (delegate manager) grotendeels (mainly) heeft achter de rug (have behind the back …) het beter doen (do improve) in Europa en erbuiten. (in Europa and) ingaan op (enter on) inleiden (direct in) jammer vinden (find … moan) Kamer van Volksvertegenwoordigers (chamber of members of parliament) kwartaal (trimester) lastenverlaging (a charge lowering) lening op afbetaling (loan on payment) logistieke experts (logistical loss assessors) Minderinkomsten (less income) minister van Financiën (minister of finances) net (network) net even veel (network even much) ondernemersorganisatie (entrepreneur organization) open houden (open to hold)
to attract3 as usual on the other hand to take certain costs on oneself study power stations train robbery just as well just as well managing director for the most part to get over perform outside jump at to initiate to feel sorry about lower house quarter lowering of the charges installment loan logistical experts income reduction Minister of Finance precisely just as much/many employers’ association keep open
Of course the Systran algorithm does not recognize this construction as such; the translation generated by Systran is rather the result of a word by word translation process of this sentence. It may be expected though that by adding an item like this one to the dictionary the analysis process may be redirected correctly. 3
opmerken (notice) over die acht jaar (which eight years) overnemer (take over) overwegen () profiteren van (to profit of) rendement (yield) splitsing (junction) te koop aanbieden (offer buy) toonbeeld (tone picture) vadertje staat (vadertje state) veel sneller (many more rapidly) verbouwen (cultivating) voogdijminister (protection minister) wat weggezakt (what considers sunk) willen winnen (want win) zelfstandigenorganisatie (self-employed person organization) zo snel mogelijk (as soon as possible) zorgen baren (to bear looks) zorgenkind (care child) zuiveringsstations (purification railway stations)
observe over 8 years buyer to consider to benefit from return split to offer for sale model state much more rapidly rebuilding supervising minister somewhat put aside to want to win independents' association as quickly as possible to give cause for concern problem child purification stations
A great many cases can be imagined however, in which it will be impossible to determine a clear-cut, well defined context. As the user’s interaction with Systran is limited to 1. Adding terms and phrases to the personal dictionary or to the translation memory, and 2. Indicating preference, priority for dictionary types and topical glossaries it is impossible to formulate a general rule; i.e. the user can only formulate an ad hoc translation. This might for example be the case with phrases like … ziet Maystadt weinig Belgische aanvragen
translated in the first round as … Maystadt see a little belgian proposals
As the user cannot add a rule weinig + plural few the only way out is to add a dictionary entry with the content weinig Belgische aanvragen few Belgian proposals Examples of such dictionary improvements include: al hun activiteiten (already their activities)
all their activities
behoren net tot (belong network to) België hinkt achterop (Belgium limps at the back) concurrerend digitaal tv-product (clashing digital tv-product) de grote treinroof (large the train treinroof) de koers is hersteld (the quotation has repaired)) de koper- en zinkactiviteiten (the purchasers and zinc activities) een beperkt rendement (limits … yield) een forse premie (a strong premium) eender welk bod (alike which offer) even veel nieuwe kijkers (even much) geld lenen (money lend) goed anderhalf jaar later (goods a year and a half latter) maar weinig Belgische bedrijven (but a little Belgian companies) met mijn vrouw (with my woman) na het afstoten van (after ricocheting) onze industrie hinkt achterop (our industry limps at the back) ruim 165000 abonnees (wide 165000 abonnees) transformatie naar (transformation to) weinig Belgische aanvragen (a little belgian proposals)
belong precisely to Belgium falls behind competing digital tv-product the big train robbery the quotation has been recovered the copper and zinc activities a limited yield a large premium any offer as many new viewers to borrow money about a year and a half later only few Belgian companies with my wife after pushing off our industry falls behind about 165000 subscribers transformation into few Belgian proposals
As it is obvious that neither human nor machine translation is possible without adequate vocabulary information for all the lexical items in the source text, a first and foremost priority should be the availability of correct, reliable dictionary information. It seems a matter of course although a number of complications may arise.
5. The Systran Dictionaries Systran relies heavily on its dictionaries. As one of the oldest surviving MTsystems it presents all the characteristics of its age, i.e. it leans towards direct translation rather than performing extensive syntactic – not to mention semantic – analysis. The language pairs with Dutch4, for example, have grave problems with word-order. It is obvious that, for want of reliable, elaborate parsing techniques, there can be no structural transfer of any consequence. The strengths of an MT-system like this must therefore be in the rich dictionary environment. From the point of view of the user Systran5 uses essentially three different types of dictionaries: the core dictionaries6, over which the user has absolutely no
English-Dutch, French-Dutch, Dutch-English and Dutch-French The present study was effected with Systran 5.0, where users’ permissions are fairly restricted. 6 Systran uses one-word dictionaries and multiword (idiomatic) dictionaries, with morphological, syntactic and semantic information concerning the entries, plus a translation in the target language(s) and some morphological-syntactic information concerning the target language equivalents. 4
5
control, the topical glossaries7, over which the user has limited control, and the user dictionaries8, which are by and large under control of the user. The user defines the entire content of the user dictionaries. If, during a translation process a source term is recovered in one of the user dictionaries, then that information is employed, if not the source term is looked up in the topical glossaries. The user has only limited control over the topical glossaries, in the sense that users cannot alter the content of the glossaries, but do have the permission to indicate which topical glossaries should be used and which not. Again, if a source term which is not present in one of the user dictionaries can be recovered in one of the topical glossaries9, then that is the information that will be employed. Source terms which cannot be recovered in the user dictionaries nor in the topical glossaries will be looked up in the core dictionaries. Terms that cannot be recovered there will be left untranslated. The following screen shot (Figure 1) is an illustration of the basic content for each dictionary entry, i.e. the source language entry, the target language equivalent, a rough syntactic categorization, and a reliability score.
The topical glossaries contain basically the terminology concerning a specific subject. The structure is similar to the structure of the core dictionaries. The following subjects are available: Business, Industries, Sciences, Life Sciences and Colloquial. 8 Again, their structure is similar to that of the core dictionaries. The user can define several personal dictionaries and associate them with specific content categories, or even define new content categories with which a dictionary should be associated. 9 Topical glossaries used in the first round were: Business, Industries, Sciences and Life Sciences. 7
Figure 1, Basic Dictionary Information
Even though many users will probably content themselves with this basic dictionary functionality, it can be advisable to add expert coding as well. The precise content of the expert coding dictionary screens will depend on the syntactic categorization of the entry word. One would expect that the underlying parser should benefit from complete and exact expert coding information in these dictionaries. Any effect, however, is hardly, if at all, noticeable. Thus adding expert coding, in consequence of the absence of any decent syntactic analysis procedure for Dutch source texts – though very time-consuming – is in all likelihood a superfluous task with Dutch source language texts. In the next screen shot (Figure 2) an example is shown of the expert coding fields10. It should be noted that – depending on syntactic subcategorization – the expert coding fields content may differ considerably 10
Figure 2, Expert-coding of a dictionary entry
A heuristic technique was used to edit the personal dictionary:
Dictionary entries can be added, deleted or modified
Words, idioms and recurrent collocations of the source text are used as entry points
Entries are inserted in the dictionary for missing source words, or lexical items that were translated mistakenly, or somehow caused erroneous interpretations of the source text.
Dictionary completion and modification was continued in several steps. After each step the effect of the editing process was evaluated, i.e. the quality of the resulting translations was measured and compared with the original translation. As long as the translations improved, dictionary editing was continued. This modus operandi was continued until no significant effect of dictionary editing on the translation quality could be noticed.
In quite a number of cases an entry had to be made in the personal dictionary to avoid or to mend mistaken translations due to parsing errors,
misrepresented entries in the other, regular Systran dictionaries, or word order difficulties. A very specific word order problem Systran encounters with Dutch source texts concerns the structure of verb phrases: in Dutch the standard word order of main clauses will always be something like S Vf … O Vi and in subclauses CONJ S … O Vf Vi It is quite obvious that the Systran parser of Dutch does not even attempt to discover the relationship between Vf and Vi. In a number of those cases it was possible to have a better translation generated by including the nonlemmatized verb forms in the dictionary. Often the object plus the non-finite verb form were entered in the dictionary as one entry. Similarly combinations of verbs with prepositions or adverbs in the source texts are never recognized as such11. Again in most of these cases it was possible to improve Systran’s performance by adding an appropriate entry in the dictionary. In spite of all this, one should be aware of the fact that it is almost next to impossible to produce an acceptable translation of Dutch verb forms only on the basis of dictionary information, i.e. without access to the parser. Other entries had to be included in order to override faulty entries in one of the other dictionaries; for example, bank had to be entered because originally it was “translated” erroneously as banc. Similarly, and this is indeed quite alarming, Systran was profoundly mixed up with the modal auxiliary kunnen as in … en haar klanten kunnen op geen enkele manier te weten komen … … and its customers are possible on absolutely no manner to be possible come …
The translation seems to suggest that in one of the core dictionaries or in one of the topical glossaries there is an item that instructs Systran to translate kunnen by to be possible. A ploy that might make sense in certain contexts perhaps, but which ought to be applied correctly. Remarkably, in the same sentence, weten was interpreted by Systran exactly the same way, leading to 11
The preposition/adverb as a rule occupies the position of Vi in the clause
the translation to be possible come. This example illustrates nicely the reach, and the limits, of dictionary strength: and its customers can in absolutely no manner finding out
Finally, in the absence of any mechanism to interfere with the parser’s rules, a number a structures were interpreted as idiomatic, though only from sheer necessity; here are some examples: in heel België in Europa en erbuiten had het toen al over dat is ruim 165.000 abonnees weinig Belgische bedrijven
in complete Belgium in Europe and had it then already that is wide 165,000 abonnees a little Belgian companies
throughout Belgium in Europe and abroad then already talked about that is about 165,000 subscribers few Belgian companies
The resulting personal dictionary, NLUK_BUFI.DCT is shown in the table below. 66ste aangekocht door aanstippen aantrekken abonnee achter de rug hebben achterophinken actief is afstoten al hun activiteiten als als anders als enige anderzijds bank banken baren België hinkt achterop bepaalde kosten op zich nemen beperkt besluiten beste beste mensen bestuderen brief centrales CO2-uitstoot concurreren conjunctuurgevoelig daags na de overname dat is ruim 165.000 abonnees de grote treinroof divisie doorrekenen duurzaam bankieren
66th bought by to mention to attract subscriber to get over to fall behind is active to push off all their activities as as usual as the only one on the other hand bank banks to give cause for Belgium falls behind to take certain costs on oneself limited to conclude best best people to study letter power stations CO2 emissions to compete susceptible for market trends a day after the take-over that is about 165,000 subscribers the big train robbery division to calculate durable banking
een beetje een onding eender welk even goed even veel expert finaal forse premie gedelegeerd bestuurder goed anderhalf jaar later grotendeels had het toen al over happig herstellen het grote zorgenkind het iets beter doen dan het verbouwen in de media in Europa en erbuiten in heel België in marstempo ingaan op inleiden intercommunale Italianen jammer vinden kijker koper kunnen kwartalen lastenverlaging leidde de operatie in lenen lening op afbetaling meekrijgen (prep: van) meerinkomsten metamorfose minderinkomsten minister van Financiën net net even veel nochtans nonchalance om ABX af te stoten om bepaalde kosten op zich te nemen om de voorfinanciering door de intercommunales te vergoeden onder Europese druk ondernemersorganisatie onding op haar beurt open houden oplaadbaar opmerken opsplitsing
a bit trash any just as well as many expert ultimately large premium managing director about a year and a half later for the most part then already talked about eager to recover the large problem child to perform a little bit better than rebuilding in the media in Europe and abroad throughout Belgium at top speed to jump at to initiate intermunicipal company Italians to feel sorry about viewer copper can quarter lowering of the charges initiated the operation to borrow installment loan to get (prep: from) surplus revenue metamorphosis income reduction Minister of Finance precisely just as much yet casualness to push off ABX to take certain costs on themselves to compensate for prefinancing by the intermunicipal companies under European pressure employers' association absurdity in turn to keep open chargeable to observe breakdown
over die acht jaar overnemer overwegen pakweg partijsoldaat profiteren van redactie rendement roamingtarieven schijnzelfstandige splitsing splitsing stootte af te koop aanbieden te weten komen telecomgroep terninatie toenmalig toonbeeld toonbeeld transformatie naar treinroof trimester tv-kijkers tv-product universitairen vadertje staat vanochtend vastelijndiensten veel sneller verbolgen verbouwen verbouwing verbouwingen verbouwingen verbouwingswerken verbouwingswerken vereenzelvigen verkoop verzesvoudiging via het fiscaal pact vice-premier vlot aangevraagd voluit voogdijminister vrijdag vrouw wat weggezakt weer weinig Belgische aanvragen weinig Belgische bedrijven weinige tijd later wie willen winnen
over eight years buyer to consider roughly party soldier to benefit from editorial board yield roaming tariifs mock-independent division split pushed off to offer for sale to find out telecom group termination former example model transformation into train robbery quarter television viewers television product academics the state this morning fixed line services much more rapidly incensed to rebuild renovation alterations rebuilding alterations rebuilding to identify sale sextupling through the fiscal pact vice-PM requested smoothly in full supervising minister on Friday wife somewhat put aside again few Belgian proposals few Belgian companies little time later who to want to win
zelfs al zelfstandigenorganisatie zich houden aan zo zo snel mogelijk zonne-energie zorgen zorgen baren zowat zuiveringsstations
even though independents' association adhere to so as quickly as possible solar power concern to give cause for concern sort of purification stations
A second personal dictionary, PROPER NAMES.DCT deals with names; in some cases this was necessary to avoid “translation” of proper names, as in Bank van de Post (banc of the mail) and stranger even, Olivier Chastel , in the target text rendered as oil four Chastel12; in other cases it was necessary to avoid erroneous translations like … a onding Karel bought Vincks predecessor…,. It is interesting that the mistaken translation, due to a flawed syntactictic analysis process can be mended by simply adding an appropriate entry in the proper names dictionary. Another type of entry in the proper names dictionary concerns compound words with a proper name morpheme, like ABX-baas, NMBS-bestuurder, PSpartijsoldaat, Corelio-kranten, … ABX ABX Logistics ABX moest de deur uit ABX-baas Bank van de Post Belgacom Coenjaerts Corelio-kranten De Tijd Dirk Dirk Saelens een LOA Elia-heffing Etienne Schouppe ICS International Carrierdiensten Kamer van Volksvertegenwoordigers Karel Vinck KBC KBC Securities LOA MR-kamerlid MR-kamerlid MR-kopstuk NMBS-baas
ABX ABX Logistics ABX had to leave the house ABX-chief Bank van de Post Belgacom Coenjaerts Corelio-newspapers De Tijd Dirk Dirk Saelens an IL Elia levy Etienne Schouppe ICS International Carrierservices Lower House Karel Vinck KBC KBC Securities IL MR-representative representative of the MR MR-boss boss of the NMBS
Systran apparently contains an internal morphological analysis component which is meant to analyse compound words which are not recognized as such by the system. Quite remarkably it also tries to analyse terms for which this processing method is absolutely inappropriate, such as proper names; for example: Olivier → oil four . 12
NMBS-bestuurder NMBS-holding NMBS-missie NMBS-top NMBS-waarborgen Olivier Olivier Chastel PS-partijsoldaat Saelens SP.A-voorzitster Stefan Stefan Coenjaerts Telenet Telenet-woordvoerder Vlaanderen
NMBS-governor NMBS-holding NMBS-mission NMBS leadership NMBS-guarantee Oliver Olivier Chastel PS party soldier Saelens SP.A-chairwoman Stefan Stefan Coenjaerts Telenet Telenet-spokesman Vlaanderen
6. Round Two In table six a side-by-side rendering of the seven text fragments as processed in rounds one and two is shown. The round one version was generated without using any supplementary personal dictionaries; all the topical glossaries except Colloquial were activated. The ranking applied was: Business, Industries, Sciences and Life Sciences. After the error analysis the personal dictionaries13 were compiled and the second translation round was launched. This time, however, only two topical glossaries were activated: Business and Industries as the Sciences and Life Sciences glossaries apparently had little, if not adverse, effect.
SOURCE
ROUND ONE
ROUND TWO
ABX: EEN BONAFIDE VERKOOP OF DE GROTE TREINROOF?
ABX: BONAFIDE SELL OR LARGE THE TRAIN TREINROOF?
ABX: BONAFIDE SALE OR THE BIG TRAIN ROBBERY?
Begin augustus 2006 stootte de NMBSholding zijn veelbesproken logistieke bedrijf ABX Logistics af. De overnemer, voor 1 symbolische euro, was de Britse investeringsgroep 3i. Goed anderhalf jaar later wordt ABX Logistics voor ruim 600 miljoen euro opnieuw te koop aangeboden. Nu blijkt dat destijds een ernstige kandidaat-overnemer, het Italiaanse Sofipa-Capitalia, nooit werd gehoord. 'De verkoop van ABX was van meet af aan verdacht', klinkt het vandaag in financiële kringen.
Beginning august 2006 stootte NMBSholding have been much-discussed logistical company ABX Logistics finished. The take over, for 1 symbolic euro, was the British investment group 3i. Goods a year and a half latter ABX Logistics for more than 600 million euro are again offered buy. Now appears that then a serious candidate purchaser, Italian Sofipa-Capitalia were never heard. It sells today in financial rings of ABX was from the beginning suspected, sounds.
Beginning august 2006 stootte NMBSholding has been much-discussed logistical company ABX Logistics finished. The buyer, for 1 symbolic euro, was the British investment group 3i. About a year and a half later ABX Logistics for more than 600 million euro is again offered buy. Now appears that then a serious candidate purchaser, Italian Sofipa-Capitalia were never heard. Sale of ABX was from the beginning suspected, sounds it today in financial rings.
ABX moest de deur uit, en zo snel mogelijk. Dat was zowat de richtlijn die toenmalig NMBS-baas Karel Vinck meekreeg van zijn voogdijminister Johan Vande Lanotte (SP.A). Dat blijkt uit het boek SOS NMBS van onderzoeksjournalist Paul Huybrechts. ABX werd destijds in alle
ABX had the door from, and as soon as possible. That was zowat the directive which then NMBS-baas got Karel Vinck of its protection minister Johan Vande Lanotte (SP.A). That becomes clear from the book SOS NMBS of research jounalist Paul Huybrechts. ABX were then depicted in all
ABX had to leave the house, and as quickly as possible. That was sort of the directive which former boss of the NMBS got Karel Vinck of its supervising minister Johan Vande Lanotte (SP.A). That becomes clear from the book SOS NMBS of research jounalist Paul Huybrechts. ABX was then
13
NLUK_BUFI.DCT and PROPER NAMES.DCT for names and collocations with names.
rapporten afgeschilderd als het grote zorgenkind van de NMBS, een onding aangekocht door Karel Vincks voorganger, Etienne Schouppe.
reports as large a care child of the NMBS, a onding Karel bought Vincks predecessor, Etienne Schouppe.
depicted in all reports as the large problem child of the NMBS, a trash bought by Karel Vincks predecessor, Etienne Schouppe.
ABX zou uiteindelijk de stok zijn waarmee Schouppe (die een ontslagvergoeding incasseerde die in de media op 2 miljoen euro werd geraamd) door de paarse regering buiten de deur werd geranseld. De NMBS stond ook onder Europese druk om ABX af te stoten.
ABX, finally the stick with which Schouppe (that dismissal compensation would be received which in the mediums on 2 million euro was estimated) by the purple government outside the door was flogged. THE NMBS stood ricochet also under European occupied ABX.
ABX, finally the stick with which Schouppe (that dismissal compensation would be received which in the media on 2 million euro was estimated) by the purple government outside the door was flogged. THE NMBS stood also under European pressure to push off ABX.
In juli 2005 al stelde MR-Kamerlid Olivier Chastel in de Kamer van Volksvertegenwoordigers de verkoop van ABX Logistics aan 3i ter discussie. Chastel had het toen al over de verkoop voor 1 symbolische euro. Jannie Haek, hoofd van de NMBS-holding in opvolging van Karel Vinck, leidde de operatie in met zijn verklaring dat de waarde van ABX Logistics nul was.
In July 2005 already MR-Kamerlid brought up oil four Chastel in for discussion the chamber of members of parliament sell of ABX Logistics to 3i. Chastel had it then already concerning sell for 1 symbolic euro. Jannie Haek, head of NMBS-holding in follow up of Karel Vinck, direct the operation in with its declaration that the value of ABX Logistics zero were.
In July 2005 already MR-representative brought up Olivier Chastel in the Lower House for discussion sale of ABX Logistics to 3i. Chastel then already talked about sale for 1 symbolic euro. Jannie Haek, head of NMBS-holding in follow up of Karel Vinck, initiated the operation with its declaration that the value of ABX Logistics zero were.
Een verklaring die destijds door logistieke experts met verbijstering werd aangehoord. Op papier zou 3i voor 10 miljoen euro aandelen van ABX Logistics kopen en zou het dan nog eens 70 miljoen euro investeren in de groep. Maar daartegenover stond dan weer het engagement van de NMBS om bepaalde kosten op zich te nemen. …
A declaration which was then listened to by logistical loss assessors with bewilderment. On paper 3i for 10 million euro shares of ABX Logistics buy and then once more 70 the million would invest euro in the group. But resisted on the other hand then the commitment of the NMBS determines in itself costs to take….
A declaration which was then listened to by logistical experts with bewilderment. On paper 3i for 10 million euro shares of ABX Logistics buy and then once more 70 the million would invest euro in the group. But on the other hand then again the commmitment of the NMBS stood to take certain costs on themselves.
BELGACOM TV GROEIT SNELLER DAN TELENETTV
BELGACOM TV GROW MORE RAPIDLY THAN TELENET-TV
BELGACOM TV GROWS MORE RAPIDLY THAN TELENET-TV
Belgacom TV trok de laatste drie kwartalen meer nieuwe digitale kijkers aan dan het concurrerend digitaal tv-product van het kabelbedrijf Telenet. Dat merkt analist Dirk Saelens van KBC Securities op in zijn rapport over de jaarresultaten van de telecomgroep en berekeningen van de redactie bevestigen dat.
Belgacom TV drew the last three trimesters more new digital kijkers to than clashing digital tv-product of the cable company the Telenet. That notices analyst Dirk Saelens van KBC Securities in its report concerning the year results of the telecomgroep and calculations of the redactie confirm that.
Belgacom TV drew the last three quarter more new digital viewers to than it clashing digital television product of the cable company Telenet. That notices analyst Dirk Saelens of KBC Securities in its report concerning the year results of the telecom group and calculations of the editorial board confirm that.
(tijd) - Belgacom TV trok het afgelopen jaar bijna 166.000 nieuwe klanten aan, net even veel als Telenet. De opsplitsing over de vier kwartalen toont echter dat Belgacom TV de laatste negen maanden sneller groeide dan zijn concurrent.
(time) - Belgacom TV drew new customers last year almost 166,000 to, network even much as Telenet. The opsplitsing over the four trimesters shows however that Belgacom TV the last grew nine months more rapidly than its competitor.
(time) - Belgacom TV attracted new customers last year almost 166,000, just as much as Telenet. The breakdown concerning four quarter shows however that Belgacom TV the last grew nine months more rapidly than its competitor.
De aanwas bij Belgacom bedroeg respectievelijk 55.800 (4ekw), 58.000 (3ekw) en 42.000 (2ekw) tegen 51.000 (4ekw), 31.000 (3ekw) en 34.000 (2ekw) bij Telenet.
Respectively the growth at Belgacom amounted to 55,800 (4ekw), 58,000 (3ekw) and 42,000 (2ekw) against 51,000 (4ekw), 31,000 (3ekw) and 34,000 (2ekw) at Telenet.
Respectively the growth at Belgacom amounted to 55,800 (4ekw), 58,000 (3ekw) and 42,000 (2ekw) against 51,000 (4ekw), 31,000 (3ekw) and 34,000 (2ekw) at Telenet.
Telenet-woordvoerder Stefan Coenjaerts stelt in een reactie dat Belgacom TV actief is in heel België, terwijl Telenet zijn product slechts in twee derde van Vlaanderen aanbiedt.
Telenet-woordvoerder Stefan Coenjaerts state in a response that Belgacom TV assets are in complete Belgium, whereas Telenet its product offers only in two thirds of Flanders.
Telenet-spokesman Stefan Coenjaerts states in a response that Belgacom TV is active throughout Belgium, whereas Telenet its product offers only in two thirds of Flanders.
Belgacom en Telenet telden eind december respectievelijk 305.300 en 391.000 digitale tv-kijkers. Belgacom TV wil dit jaar minstens even veel nieuwe kijkers winnen als in 2007. Dat is ruim 165.000 abonnees.
Belgacom and Telenet counted end December respectively 305,300 and 391,000 digital tv-kijkers. Belgacom TV want win this year at least even much new kijkers such as in 2007. That is wide
Belgacom and Telenet counted end December respectively 305,300 and 391,000 digital television viewers. Belgacom TV wants win this year at least as many new viewers as in 2007. That is
165,000 abonnees. …
…
about 165,000 subscribers.
BELGIE HINKT ACHTEROP BIJ ONTWIKKELING VAN
BELGIUM LIMPS AT THE BACK AT DEVELOPMENT
BELGIUM FALLS BEHIND AT DEVELOPMENT OF
WINDMOLENS
OF WIND MILLS
WIND MILLS
Er zijn maar weinig Belgische bedrijven die windmolens produceren of andere technologie waarmee groene elektriciteit wordt geproduceerd. In vergelijking met Duitsland, Spanje of Denemarken hinkt onze industrie achterop. Dat zegt Philippe Maystadt vrijdag in De Tijd. De gewezen minister van Financiën is voorzitter van de Europese Investeringsbank (EIB).
But there are a little belgian companies that produces wind mills or other technology with which green electricity is produced. In comparison with Germany, Spain or Denmark our industry limps at the back. That says Philippe Maystadt Friday in the time. The former minister of finances is President of European the investment banc (EIB).
But there are few Belgian companies that produces wind mills or other technology with which green electricity is produced. In comparison with Germany, Spain or Denmark our industry limps at the back. That says Philippe Maystadt Friday in De Tijd. The former Minister of Finance is President of European the investment banc (EIB).
De EIB heeft vorig jaar voor 47,8 miljard euro nieuwe kredieten verstrekt. De Europese instelling steunt met dat geld de economische ontwikkeling in Europa en erbuiten. Een flink deel van het geld gaat naar milieuprojecten, bijvoorbeeld de bouw van zuiveringsstations.
THE EIB has last year supplied new credits for 47.8 billion euro. European the institution supports with that money economical the development in Europe and. A considerable part of the money goes to environment projects, for example the construction of purification railway stations.
THE EIB has last year supplied new credits for 47.8 billion euro. European the institution supports with that money economical the development in Europe and abroad. A considerable part of the money goes to environment projects, for example the construction of purification stations.
Maar tot zijn teleurstelling ziet Maystadt weinig Belgische aanvragen voor de ontwikkeling van milieutechnologie. In Duitsland en Denemarken heeft men veel sneller fiscale voordelen toegekend aan groene elektriciteit, waardoor daar meer bedrijven de technologie ontwikkelen. "België heeft nochtans een rijke traditie van mechanische en technologische bedrijven."
But to its vexation Maystadt see a little belgian proposals for the development of environment technology. In Germany and Denmark one has granted many more rapidly fiscal advantages to green electricity, through which there more companies develop the technology. Belgium has nochtans a rich tradition of mechanical and technological companies.
Wat Maystadt wel krijgt, zijn aanvragen voor de bouw van kerncentrales. Kernenergie was wat weggezakt, maar verschillende EU-lidstaten overwegen nu opnieuw centrales te bouwen of open te houden. Volgens Maystadt is dat normaal als je de CO2-uitstoot wil beperken op een moment dat de hernieuwbare energiebronnen nog maar een beperkt rendement hebben.
What, however, gets Maystadt, is proposals for the construction of nuclear plants. Nuclear power was what considers sunk, but different EU-member states now again building call centers or open to hold. According to Maystadt are that normally if you want CO2-uitstoot limit at a moment that still but limits the renewable energy sources yield have.
What, however, gets Maystadt, is proposals for the construction of nuclear plants. Nuclear power was somewhat put aside, but different EU-member states consider now again power stations to build or open to hold. According to Maystadt are that normally as you the CO2 emissions want limit that at a moment the renewable energy sources have still but limited yield.
GROENE LENING PAK DUURDER DAN
GREEN LOAN PACKAGE MORE EXPENSIVELY
GREEN LOAN PACKAGE MORE EXPENSIVELY
HYPOTHECAIRE LENING
THAN MORTAGE LOAN
THAN MORTAGE LOAN
Huiseigenaars die beslissen om verbouwingswerken aan te vatten en daarvoor geld lenen, kiezen best voor een hypothecair krediet en niet voor een lening op afbetaling (LOA of de zogenaamde groene lening). Dat meldt immotheker.be.
house owners who decide to seize verbouwingswerken and for that money lend, choose best for a mortgage credit and not for a loan on payment (LOA or the socalled green loan). That communicates immotheker.be.
house owners who decide to seize rebuilding and for that money borrow, choose best for a mortgage credit and not for an installment loan (IL or the so-called green loan). That communicates immotheker.be.
"Een LOA is vlot aangevraagd, het geld krijg je snel ter beschikking en omdat de bank geen hypotheek neemt op je huis zijn er geen notariskosten, geen registratierechten, kortom geen hypotheekinschrijvingskosten", klinkt het. "Anderzijds is de rentevoet hoger dan bij een hypothecair krediet: pakweg 6,0 procent voor een LOA op tien jaar."
LOA have been requested fluent, the money acquires you fast and because the banc records no mortgage your house is there no notary costs, no registration rights, in short no mortgage registration costs, sound it. On the other side the interest rate is superior than at a mortgage credit: pakweg 6.0 percent for LOA on ten years.
An IL is requested smoothly, the money acquires you fast and because the banc records no mortgage your house is there no notary costs, no registration rights, in short no mortgage registration costs, sound it. On the other hand the interest rate is superior than at a mortgage credit: roughly 6.0 percent for an IL on ten years.
Voor die looptijd is volgens Immotheker al een hypothecair krediet te verkrijgen aan minder dan 4,25 procent. Bovendien kan
For that duration is according to Immotheker already a mortgage credit obtain to less than 4.25 percent. Moreover
For that duration is according to Immotheker already a mortgage credit obtain to less than 4.25 percent. Moreover
But to its vexation Maystadt see few Belgian proposals for the development of environment technology. In Germany and Denmark one has granted much more rapidly fiscal advantages to green electricity, through which there more companies develop the technology. Belgium has yet a rich tradition of mechanical and technological companies.
er met een LOA niet geprofiteerd worden van de fiscale voordelen die specifiek voorzien zijn voor het verbouwen gefinancierd met een hypotheeklening.
cannot become with LOA profited there of fiscal the advantages which are foreseen specific for cultivating have been financed with a mortgage loan.
cannot become with an IL profited there of fiscal the advantages which are foreseen specific for rebuilding have been financed with a mortgage loan.
Immotheker berekende dat bij een lening van 50.000 euro op tien jaar voor verbouwingen, de LOA in totaal 30.702 euro duurder uitkomt dan de hypothecaire lening. "Door het belastingvoordeel wordt de hypothecaire lening een winstgevende zaak en heeft vadertje staat aan het eind van de rit 29 procent van de verbouwing gesponsord", besluit Immotheker.
Immotheker calculated that at a loan of 50,000 euro on ten years for verbouwingen, the LOA in total 30,702 euro end up more expensively than the mortage loan. By the tax advantage the mortage loan a profitable business and has to vadertje state at the end of the ride 29 per cent of the verbouwing is sponsored, conclusion Immotheker.
Immotheker calculated that at a loan of 50,000 euro on ten years for rebuilding, IL in total 30,702 euro end up more expensively than the mortage loan. By the tax advantage the mortage loan a profitable business and has to state at the end of the ride 29 per cent of the renovation is sponsored, concludes Immotheker.
TRIODOS ABSOLUUT TOONBEELD DUURZAAM
TRIODOS ABSOLUTELY TONE PICTURE DURABLY
TRIODOS ABSOLUTELY MODEL DURABLE
BANKIEREN
BANKS
BANKING
Triodos is het absolute toonbeeld voor duurzaam bankieren. ING doet het iets beter dan de andere grootbanken en Argenta en Bank van De Post hebben duidelijk geen boodschap aan een ethisch beleid. Dat is de conclusie van een onderzoek van Test-Aankoop naar het ethische beleid van zeven banken.
Triodos are absolutely the tone picture for durable bank. ING do it something improve then the other large bancs and Argenta and banc of the mail has clear no message to ethical policies. That is the conclusion of an investigation of TestAankoop to the ethical policies of seven bancs.
Triodos are absolutely the model for durable banking. ING do it something improve then other large banks and Argenta and Bank van de Post have clear no message to ethical policies. That is the conclusion of an investigation of TestAankoop to the ethical policies of seven banks.
(belga) - De zeven banken zijn Argenta, Bank van De Post, Dexia, Fortis, ING, KBC en Triodos. Test-Aankoop onderzocht vijf thema's: rechten van werknemers, milieu, belasting(paradijzen), bewapening en mensenrechten.
(belga) - the seven bancs are Argenta, banc of the mail, Dexia, Fortis, ING, KBC and Triodos. Test-Aankoop examined five topics: rights of employees, environment, tax (paradises), arms and human rights.
(belga) - seven banks are Argenta, Bank van de Post, Dexia, Fortis, ING, KBC and Triodos. Test-Aankoop examined five topics: rights of employees, environment, tax (paradises), arms and human rights.
Argenta en Bank van De Post hebben volgens het onderzoek duidelijk geen boodschap aan een ethisch beleid. Zo antwoordde Bank van De Post als enige niet op de vragenlijst van Test-Aankoop. Ze geeft nergens informatie over haar beleggingen en haar klanten kunnen op geen enkele manier te weten komen wat met hun geld gebeurt.
Argenta and banc of the mail has according to the investigation clear no message to ethical policies. Thus as some on the questionnaire of Test-Aankoop did not answer banc of the mail. She gives nowhere information on its investments and its customers are possible on absolutely no manner to be possible come what with their money takes place.
Argenta and Bank van de Post have according to the investigation clear no message to ethical policies. So as the only one not on the questionnaire of TestAankoop answered Bank van de Post. She gives nowhere information on its investments and its customers can in absolutely no manner finding out what with their money takes place.
"Wat de andere onderzochte banken betreft, toont de enquête aan dat de banken in kwestie voor diverse domeinen wel degelijk over gedragscodes beschikken, maar dat die veelal vaag zijn en niet voor al hun activiteiten gelden." TestAankoop vindt dat vooral jammer wat de mensenrechten en de rechten van werknemers betreft. "Want die behoren net tot de belangrijkste domeinen waar een financiële instelling op zou moeten werken", luidt het. ...
To concerns, the survey shows the other examined bancs what to that the bancs have in question for several fields really codes of conduct, but that which is mostly vague and does not apply to already their activities. Test-Aankoop find that especially moan what human rights and the rights of employees concerns. Because those belong network to most important the fields where a financial institution would have work on, it sounds.
To other examined concerns, the survey shows banks what to that banks has in question for several fields really codes of conduct, but that which is mostly vague and does not apply to all their activities. Test-Aankoop find that especially moan what human rights and the rights of employees concerns. Because those belong precisely to important the fields where a financial institution would have work on, it sounds.
UNIZO VEROORDEELT DOORGEREKENDE ELIA-
UNIZO CONDEMN CALCULATED ELIA-HEFFING ON
UNIZO CONDEMN CALCULATED ELIA LEVY ON
HEFFING OP ENERGIEFACTUUR
ENERGY INVOICE
ENERGY INVOICE
Unizo veroordeelt de doorgerekende Eliaheffing in de energiefacturen voor burgers en bedrijven en meent dat de intercommunales de onterecht extra geïnde bedragen moeten terugbetalen aan de burgers en bedrijven. Dat meldt de ondernemersorganisatie vrijdag.
Unizo condemn calculated Elia-heffing in the energy invoices for civilians and companies and think that the intercommunales amount to undeserved extra collected must pay back to the civilians and companies. That communicates Friday to the entrepreneur organization.
Unizo condemn the calculated Elia levy in the energy invoices for civilians and companies and think that the intermunicipal companies amount to undeserved extra collected must pay back to the civilians and companies. That communicates the employers' association on Friday.
"De intercommunales willen zo de door hen voorgefinancierde Elia-heffing recupereren", luidt het. Ook vindt de zelfstandigenorganisatie dat de federale regering de meerinkomsten uit de gestegen energieprijzen op haar beurt moet aanwenden om de voorfinanciering door de intercommunales te vergoeden.
The intercommunales want recuperate this way Elia-heffing prefinanced by them, sound it. Also the self-employed person organization finds that the federal government the meerinkomsten from the risen energy prices for her part must apply compensating for the intercommunales prefinancing by.
The intermunicipal companies want the Elia levy prefinanced by them recuperate so, sound it. Also the independents' association finds that the federal government the surplus revenue from the risen energy prices in turn must apply to compensate for prefinancing by the intermunicipal companies.
Unizo wijst er ook op dat de Vlaamse regering de Elia-heffing voor gezinnen en bedrijven officieel afschafte vanaf 1 januari 2008. Dat betekende een lastenverlaging van 83 miljoen euro. De minderinkomsten voor de Vlaamse gemeenten werden gecompenseerd via het fiscaal pact, aldus de organisatie.
Unizo indicate also that the Flemish government officially abolished Elia-heffing for families and companies from 1 January 2008. That meant a charge lowering of 83 million euro. The less income for the Flemish communities was compensated through fiscal the pact, thus the organization.
Unizo indicate also that the Flemish government officially abolished the Elia levy for families and companies from 1 January 2008. That meant a lowering of the charges of 83 million euro. The income reduction for the Flemish communities was compensated through the fiscal pact, thus the organization.
WIE UMICORE WIL OVERNEMEN ZAL EEN FORSE
`WHICH UMICORE WANT TAKE OVER A STRONG
WHO UMICORE WANT TAKE OVER A LARGE
PREMIE MOETEN BETALEN
PREMIUM MUST PAY
PREMIUM MUST PAY
(tijd) – ‘Wie Umicore wil overnemen zal een forse premie moeten betalen.’ Dat zegt gedelegeerd bestuurder Thomas Leysen in een gesprek met De Tijd. Leysen zal een hoog bod op zijn bedrijf bestuderen, maar lijkt er op te vertrouwen dat de aandeelhouders niet op eender welk bod zullen ingaan.
(time) - `which Umicore want take over a strong premium must pay. That says delegate manager Leysen in an interview with the time to Thomas. Leysen high an offer on its company observate, but seem there on at trust the shareholders to that not on alike which offer will enter.
(time) - `who Umicore want take over a large premium must pay. That says Thomas Leysen in an interview with De Tijd to managing director. Leysen high an offer on its company study, but seem on at trust the shareholders to that on any offer will not enter there.
De forse koersval op 14 februari na de bekendmaking van de schitterende resultaten, baart Leysen niet echt zorgen.
The strong quotation fall on 14 February after the disclosure of the superb results, really does not bear Leysen looks.
The strong quotation fall on 14 February after the disclosure of the superb results, does not give cause for Leysen real concern.
‘Ik heb ’s avonds even goed met mijn vrouw Valentijn gevierd als anders’. Hij stipt aan dat de koers intussen grotendeels hersteld is. Vandaag splitste Umicore overigens zijn aandelen in vijf. Die splitsing kom er na de verzesvoudiging van het aandeel over de voorbije acht jaar
`I at night even goods with my woman have celebrated Valentijn as the other. He exact to that the quotation has meanwhile mainly repaired. Today split up Umicore moreover its shares in five. That junction bowl there after the verzesvoudiging of the share over the past eight years
`I just as well with my wife has at night celebrated Valentijn as usual. He exact to that the quotation is meanwhile for the most part recovered. Today split up Umicore moreover its shares in five. That split bowl there after the sextupling of the share over the past eight years
Umicore heeft over die acht jaar een grote metamorfose achter de rug na het afstoten van de koper- (Cumerio) en zinkactiviteiten (Nyrstar). De transformatie naar een specialist in materiaaltechnologie (voor katalysatoren, oplaadbare batterijen, zonne-energie) is volgens Leysen makkelijker uit te leggen in het buitenland dan in België. …
Umicore have which eight years large a metamorfose behind the back after ricocheting the purchasers (Cumerio) and zinc activities (Nyrstar). The transformation to a specialist in material technology (for catalysts, oplaadbare batteries, zonneenergie) is according to explain Leysen more easily abroad than in Belgium.
Umicore have over eight years large a metamorphosis behind the back after pushing off of the purchasers (Cumerio) and zinc activities (Nyrstar). Transformation into a specialist in material technology (for catalysts, chargeable batteries, solar power) is according to explain Leysen more easily abroad than in Belgium.
Admittedly the translation obtained in the second round is significantly better than the one generated in the first round. Yet, the burden of translation and syntax errors is considerable. 86 errors were counted; the nature of these errors, however, is quite different from what was encountered in the first round. The results of a detailed error analysis are shown in Table 7. ERROR TYPE TOTAL Vocabulary Word order
ROUND 1 225 99
ROUND 2 86 45
Verb – conjugation Verb – infinitives Translation errors Gender/Number Article Pronouns Miscelaneous
11 5 10 5 2 2 6
Clearly the total number of errors is down from 225 to 86. It should be noticed however, that 99 vocabulary errors were counted in the first round, while in round 2 no obvious vocabulary errors occurred14. As already mentioned, it does not make much sense to examine structural errors in the absence of adequate dictionary information; that is why in round 1 the remaining errors (126) were not analyzed. Quite a number of them should not reappear as soon as the correct dictionary information is made available to the MT-system. In round 2 using the upgraded dictionaries, the expectation was that the resulting (raw) machine translations would give a good idea of the strong and weak points of Systran as a parsing and structural transfer tool. As the error analysis demonstrates the result is rather disappointing since 52.32 % of the remaining errors are word-order errors and 18.60 % are verb errors, i.e. more than 70 % of all that went wrong was due to defective source syntactic analysis, or flawed structural transfer. Here are some typical examples: de onterecht extra geïnde bedragen als je de CO2-uitstoot wil beperken de economische ontwikkeling de Europese Investeringsbank (kunnen) … te weten komen wat met hun geld gebeurt (B… trok) het afgelopen jaar bijna 166.000 nieuwe klanten (aan) van de fiscale voordelen
(door de paarse regering) buiten de deur werd geranseld wat de mensenrechten en de rechten van werknemers betreft terwijl Telenet zijn product (… in … Vlaanderen) aanbiedt heeft vorig jaar (voor 47,8 miljard euro nieuwe kredieten) verstrekt wordt (ABX … voor 600 …) opnieuw te koop aangeboden maar lijkt er op te vertrouwen (waar een financiële instelling) op zou moeten werken
amount to undeserved extra collected as you the CO2 emissions want limit economical the development European the investment banc (can) … finding out what with their money takes place (B… drew) new customers last year almost 166,000 (at) of fiscal the advantages (by the purple government) outside the door was flogged what human rights and the rights of employees concerns whereas Telenet its product offers has last year supplied (ABX … for 600 …) is again offered buy but seem on at trust (where a financial institution) would have work on
There are ways to remedy the errors encountered in the second round; the question remains though, if that is a worthwhile investment in time and energy, particularly as there is no way to turn these corrections into general improvements of the MT-system. Users can intervene on the level of the vocabulary by making personal dictionaries and selecting as well as ordering specific topical glossaries, but they cannot alter syntactic analysis processes. The only options left to the user of Systran include
There is a small number (± 5) of structural errors that might perhaps be avoided by adding better dictionary information to certain items in the personal dictionary. 14
Pre-editing Post-editing Adding phrases and clauses to the translation memory
These are activities implying a heavy workload without any prospect of a general improvement of the MT-system. Moreover, the amount of time needed to formulate these corrections is quite probably a multiple of the time required for the “manual” translation of a given Dutch source text. That means that, at least for translation of Dutch source texts, Systran meets none of the requirements to raise MT as a valuable alternative for human translation: neither its quality of service, nor the overall speed of operation15, nor the price of the translation would justify the adoption of Systran MT for the development of new translation modules. Given the choice between investing in the correction and improvement of Systran MT and the option of starting from scratch to develop a modern MT System relying upon the know-how available these days it is obvious that the odds are not in favour of Systran. [Bibliography]
The course of time from source text to delivery of readable, understandable output in the target language 15