Natural language processing is able to process human documents and do some interesting things, like judge the "sentiment" of the document or note similarities with other documents in a corpus. There have been a number of (attempts) to automagically summarize documents.

However, the quality of those summaries varies from "Wow! That was pretty good!" to "This is complete rubbish." Unfortunately, the curve is biased toward the less-good end of the spectrum. For applications where you have a truly overwhelming number of input documents, and where some amount of erroneous summarization is not a show-stopper, maybe. But with product specifications, you're talking about very technical documents for which missing a few points could be missing the entire point of the spec, and for which getting a few details wrong can mean not just a useless, but a counter-productive, summary.

NLP is best left to applications where errors and omissions carry lower risks.

Natural language processing is able to process human documents and do some interesting things, like judge the "sentiment" of the document or note similarities with other documents in a corpus. There have been a number of attempts to automagically summarize documents.

However, the quality of those summaries varies from "Wow! That was pretty good!" to "This is complete rubbish." Unfortunately, the curve is biased toward the less-good end of the spectrum. For applications where you have a truly overwhelming number of input documents, and where some amount of erroneous summarization is not a show-stopper, maybe. But with product specifications, you're talking about very technical documents for which missing a few points could be missing the entire point of the spec, and for which getting a few details wrong can mean not just a useless, but a counter-productive, summary.

NLP is best left to applications where errors and omissions carry lower risks.