Dutch model from OPUS-MT

Dutch models converted from OPUS-MT:

https://drive.google.com/drive/folders/1NS6MUrLMk25_Qn6WM4v4ZTU5wy3BbnQn?usp=sharing

1 Like

I may change argospm-index to use this model; our current Dutch models aren’t very good. I’ve heard people say positive things about the performance of the OPUS translation models.

1 Like

Just uploaded Greek to the folder too (only English => Greek direction). The Greek => English model uses BPE encoding.

But seems to work quite well:

Input: When people don't see moose as potentially dangerous, they may approach too closely and put themselves at risk.

OPUS-MT: Όταν οι άνθρωποι δεν βλέπουν την άλκη ως δυνητικά επικίνδυνη, μπορεί να πλησιάσουν πολύ στενά και να θέσουν τον εαυτό τους σε κίνδυνο.

Current: Όταν οι άνθρωποι δεν βλέπουν τη μομφή ως δυνητικά επικίνδυνη, μπορεί να προσεγγίσουν πολύ στενά και να τεθούν σε κίνδυνο.

Backtranslations:

OPUS-MT: When people do not see the moose as potentially dangerous, they may get too close and put themselves in danger.

Current: When people don't see the reproach as potentially dangerous, they may approach too closely and put themselves in danger.

D:\Locomotive>python eval.py --config run\en_el-opus_1.0\config.json --bleu
BLEU score: 74.7987
1 Like

I’ve converted/uploaded Swedish also, although the backtranslations seems pretty close, not sure it’s an improvement.

This is where this (Playing Around with Machine Translation - #4 by pierotofy) would come really handy.

1 Like

I just published this OPUS-MT Dutch model:

Dutch

De raket krijgt impuls naar voren door massa (meestal hete gassen) met hoge snelheid naar achteren uit te werpen. Dit wordt beschreven volgens de raketvergelijking van Tsiolkovski. Omdat naast de brandstof ook de oxidator (zoals zuurstof) voor de motor aan boord is, kan een raket ook in vacuüm werken. Een belangrijke parameter van een raketsysteem is de specifieke impuls: hoe hoger, hoe zuiniger in gebruik van reactiemassa. Daarnaast is bij een lancering essentieel dat de raket meer kracht kan uitoefenen dan zijn eigen gewicht. Eenmaal in de ruimte kan een langdurige werking met beperkte capaciteit ook goed werken. De benodigde energie komt meestal uit een chemische reactie die de reactiemassa ondergaat (chemische raket), maar kan bijvoorbeeld in de ruimte bij langdurige werking met beperkte capaciteit ook van de zon komen.

Een grote chemische raket is vaak een meertrapsraket. Sommige lanceertuigen zijn modulair opgebouwde raketten. Dat houdt in dat afhankelijk van de massa, het formaat en de vereiste snelheid van de lading er extra hulpraketten, een andere of een extra bovenste rakettrap of een ander formaat neuskegel gebruikt kunnen worden.

Source

1.8 English translation

The rocket gets impulse forward by ejecting mass (usually hot gases) at high speed to the rear. This is described according to the rocket equation of Tsiolkovsky. Because next to the fuel is the oxydator (such as oxygen) for the engine on board, a rocket can also work in vacuum. An important parameter of a rocket system is the specific impulse: the higher, the more economical in the use of reaction mass. In addition, a launch is essential that the rocket can exert more power than its own weight. Once in space, prolonged operation with limited capacity can also work well. The required energy usually comes from a chemical reaction that undergoes the reaction mass (chemical rocket), but can also come from the sun in space for long-term operation with limited capacity.

A large chemical missile is often a multi-stage rocket. Some launchers are modular rockets. This means that depending on the mass, the size and the required speed of the charge additional auxiliary rockets, another or an additional upper rocket staircase or another size nose cone can be used.

1 Like

I’m hearing that this Dutch model works well. From Discord:

That new Dutch model is :fire::fire: so far I used it. Will run a few tests and maybe have a look if I can improve it where needed, but so far, so good

1 Like

The translation is ferry good 8.2/10 i have seen some problems with some double word for example foo bar bar and the word Republic is translate to Republikeinse as in the us party (Republican) instate of Republiek (Republic) as in the state form.
But this is easy to do if i Argos/Libre translate have a option to pick the correct word.

1 Like