Menulance has built machine translation engines for the food and beverage language. The F&B language is notoriously famous for hilarious translations published in blogs or in a restaurant menu.
Our engines have been trained with datasets containing 168 million lines. The performance of the engines were measured against external testing sets containing 29,600 sentences and they reached an average of 97.6% of correct translations, all verified manually across all the engines used.
We used Phrase-Based engine, Hierarchical Translation engines and Transformers. They all reached a BLEU score at around 83% and a human measured score of 97.6%. The human verification process took into account the use of synonyms that were ignored by the BLEU score.
Our results show that the time we invested in developing dataset transformation technologies for NLP tasks paid off with higher than expected performance results, no matter what type of engine we used. The dataset precision and properties it possesses (derivation, decomposition, closure...) were keys to this performance.
We are going to update Menulance.com to offer our web-based machine translation engines for the F&B language.
|
|
|
|
|
1-10 employees
View all Menulance.com employees
|
|
|
Software Development
|
|
|
2012
|
|
|
Data Processing, Search, Ai, Machine Learning, Deep Learning, Machine Translation, Advanced Data Structure
|
The decision makers in Menulance.com are Frank Badrines, etc. Click to Find Menulance.com decision makers emails.
Find accurate personal emails, work emails and phone numbers for employees
Accelerate prospecting with instant access to 350M professionals from 40M companies with the right contact details.