Course Introduction
In the Advanced Computer Assisted Translation (CAT) course, we explored multiple types of software that language professionals use today in machine translation engine training, MT corpora cleaning, QA conducting, etc.
Here is a summary of what we have learned (please find the software we used in bold):
- Filter Configuration (XML in Trados)
- Introduction of an SMT Training Engine (Microsoft Translator Hub)
- MT Quality Evaluation & BLEU Scores
- Maintaining TMs and MT Corpora (Okapi Olifant)
- Introduction of a QA Tool (Okapi Checkmate)
- Financial Aspects of MT
- NMT: Neural Machine Translation
- Introduction of a Customizable SMT Engine (KantanMT)
- Trends in Translation TechnologyTools for localization-related tasks
- Please find a demo video for a file renaming tool: 1-4a Rename
Final Project Overview
In this semester, I joined a group of four other students to apply what we learned to an MT engine training project. We had a kick-off meeting with the client (the professor), trained the MT engine, and held a meeting to discuss the lessons learned.
Our team (Boba Milk 2.0 Return of the TeaM) trained a machine translation engine in Microsoft Translator Hub for Zojirushi Rice Cooker Manuals from English into Traditional Chinese, with the objectives to increase efficiency, decrease cost and improve quality of machine translation. We post-edited machine translations (PEMT) from this engine and compared the cost and time spent to that of human translation. After 14 rounds of training, our BLEU score (a measurement of MT quality) increased from 2.84 to 6.63, which was relatively lower than average. The quality of machine translation was poor compared to human translation. We do not recommend training this engine with the existing data. More details of the training process can be found in the project assets.
Final project Assets:
