I don't think this particular problem is related to the language model. "[item] without [attribute]" is trivial to understand even without a sophisticated language model.

The bad results are because they're not positively indexing the absense of the feature by deeply analyzing the images or products beyond the descriptions. "Shirt with stripes" yields almost exclusively striped shirts. Exclude those results from all "shirts" and there are still a lot of striped shirts that the search algorithm doesn't know enough to exclude.

Because there aren't two kinds of shirt, striped or solid coloured. You even acknowledge that by mentioning Hawai'ian shirts. You might want a small check shirt, a mostly plain shirt but not mind logos so not completely monocolour, a plaid shirt, a large check lumberjack style, spotted or dotted or diamond pattern, flowery pattern, you might not even know what the other available options are to search for them one by one - you just know that you don't want stripes.

There is no ambiguity in "not stripes", you can't invert it and write it in the positive form of what you want; the neatest way to describe the category of what you want to browse is "things which are not stripey".

Particular personal bugbear is car websites where you can filter in "petrol engine" or "diesel engine", but there is no support for negative filtering, so you can't choose "not LPG". In so many search-and-filter options you can't exclude your dealbreakers, and it's much more likely that I have a single dealbreaker which rejects a choice overriding all other considerations, than that I have a single dealmaker which makes a choice overriding all else.

Is it that odd though? I'm not a native english speaker, but I would say "shirt without patterns", not solid. Or for example "without visible logo"

Consider the query, "non-glass skyscrapers", which suffers from the same problem.

What do you call a skyscraper like that if you want to refer to it? They exist, but you can't find them using that search term on Google.

Yo seem to think that queries of this form are ambiguous, but they are not. Given the set of shirts, and that the subset of 'shirts with stripes' is well-defined (which you apparently accept, given that you seem to think that 'shirts with stripes' is an unambiguous query (which it is)), then its complement is well-defined; there is no 'Russell's Barber' paradox here, as there is no self-reference.

>If I want to write clearly and without ambiguity, I rewrite sentences that have these problems. Why wouldn't I do the same for search queries?

Ok, so imagine one online retailer follows your advice and expect the users to write clear and unambiguous queries, while another retailer puts extra effort into attribution.

Which one will make more money?