LMKT: Add cultural knowledge remembering and cultural safety application

[martinakaduc]

In this PR, I added two more scenarios to LMKT benchmark.

• Cultural knowledge remembering: This scenario assesses the LLM's ability to remember culturally specific knowledge, i.e., knowledge in a language and not in other languages.
• Cultural safety application: This scenario assesses the safety response of LLMs in different languages.
• Cultural evolution understanding: This scenario is about understanding the appropriate cultural norms at a specific time.

Additionally, I fixed some naming convention for previous scenarios.

[yifanmai]

This has a lot of copy and paste logic from the base evaluate() method. Instead of doing this, you should override evaluate_instances() from EvaluateInstancesMetric instead and put your logic there.

You can look at ClassificationMetric for an example of how to do this.

[yifanmai]

Hi, it's been a month since the last update; are you still working on this?

[martinakaduc]

Hi, it's been a month since the last update; are you still working on this?

Yes, I am still working on this project. There were many other things to do last month, so I haven't had time to revise it. I will restart this week.