Not enough benefit due to triplets

[pradeepiisc]

Hi,
It was a nice article to read about detecting hallucination via generating triplets from the summary and then checking the same against the reference from which the summary was generated. At first the trick looks interesting and useful, but when you actually see the implementation, there's no benefit provided by the approach.

The triplets that are generated are almost like the original sentence itself, specially when it is sent for refchecking to align score or other method, we join the triplets using, say, whitespace or newline, which then resembles the original sentence.

I might as well just call the align score method with the reference and the summary sentence instead of the triplet and get similar results.

Could you provide an argument why using triplets is better than using the original sentence ?