Think of a complex document such as an artificial intelligence book, it can include:

• neural networks
• Fuzzy logic
• expert systems
• Image processing principles for vision
• NLP
• IR

All this in what can be considered one document. IR systems must be able to discriminate a text like this in a query such as 'artificial intelligence' but it might also need to discriminate relevant sections in a query such as 'N-grams in NLP'

Getting a single system to do both types of discrimination is a complex task.

Document retrieval and passage retrieval are the part of discourse processing in language processing. But both have different techniques. Documents are more complex and lengthy as compared to passages.

When you want to give ranking to passages of a document then passage level processing has to be done.

I hope this will help you.

The idea behind the two tasks is to explore how document length affects the effectiveness of the different retrieval models.

by Ellen Voorhees, TREC project manager NIST