# ByteDance Interview NLP Engineer (Machine Translation)

## Info
- Team: ByteDance Research AI Foundation - Machine Translation Team
- All 3 rounds conducted in Chinese (after conscent).
- Below are the translations of the questions that I could remember.

## R1 - Technical
Time: 1.5hr
- Self intro + describe your work (papers) one by one.
- If you used encoder only model for translation, wouldn't it be worse than encoder-decoder model?
- Are there any other in your lab working on translation?
- What topics did your lab worked on?
- What are the recent breakthrough in speech?
- Leetcode 130. Surrounded Regions
## R2 - Technical
Time: 46m
- Self intro
- paper: NARST-CTC
- CTC is recognized as monotonic. Why do CTC work for translation?
- What are your findings?
- Besides CTC, what are some of the other NAR methods?
- Out of your works, what is the most novel idea?
- Anticipation-Free Training for Simultaneous Machine Translation
- paper: CIF
- What is CIF, what was it used for?
- CIF can be used for NAR and AR right?
- What is adaptive policy
- Did anyone try to use offline model to do simultaneous translation, if not, why?
- No coding
## R3 - Technical
Time: 1hr
- Self intro + What is your best work.
- What is your baseline?
- How are the results?
- Describe how did you parallelize it?
- How do you come by / use simultaneous translation data?
- What are the gap between current simultaneous translation systems and one applicable to real world?
- Do you think LLM can or should be applied to simultaneous translation? Justify your answer.
- Leetcode 33. Search in Rotated Sorted Array

## R4 - HR round?
Not proceeding.