Task103 facts to story

[Mihir3009]

Added new task 103: Facts to Story. Please review.

[danyaljj]

Thanks for the PR!

Few very minor suggestions:

• For "Sentence Generation": mention that the concept names are separated by "#".
• Merge "There are 3 to 5 concepts in each concept set. " with the first sentence: "you are given concept set (with 3 to 5 concepts".
• We need to use all the provided concepts in the output generation, right? If so, we can create a negative example that uses k-1 concepts (from a concept set of size k).
• I see that some instances have more than one gold output (which is good!) If the dataset has more instances than 3.5k, I would suggest selecting a subset of instances that a higher number of gold outputs per instance.

Long text generation:

• Merge "In this task, five ordered key facts are given. All the given facts are expressed in natural language." --> more concise
• There are a bunch of <sep><cls> tokens; drop.

Please update the readme of task names as well.

[Mihir3009]

I updated the task 102 file - Sentence Generation. Please review.

I have a question regarding Task 103 - Long text generation:
When you say "There are a bunch of "" tokens; drop.". Can you please explain it more? I didn’t get the point. So that I can review and resubmit.

[danyaljj]

When you say "There are a bunch of "" tokens; drop.". Can you please explain it more? I didn’t get the point. So that I can review and resubmit.

Sorry, I miss-typed. There are a bunch of non-natural(<cls> or other special tokens) that we should drop.

[Mihir3009]

Got it.

I updated both the files according to your suggestions. Please review. Thanks!

[swarooprm]

Looks good to me! Just a minor suggestion:
Change the file names that indicate the source dataset and the category e.g. if source for both tasks are 'commongen', then task name should be taskNNN_.json (e.g. 'task001_quoref_question_generation.json')

[Mihir3009]

I renamed the file and uploaded the final ones.

[swarooprm]

Accept from my side!