It's a problem of recognising context. Current technology is not very good at this. What would be nice to have is automatic disambiguation - taking fdej's example, when searching for "Taylor shift", a search engine could ask "Did you mean Taylor Swift, or do you want to shift polynomials?".
Other possible examples: "bat" (flying animals or sports equipment?), "rock" (stones, music, or maybe both? :P), "minute" (60 seconds, report, or something extremely small?). In all of these cases, the software can't tell the meanings apart very well yet. That's a problem with the software.
duckduckgo does exactly that - it lists all the possible interpretations across the top of the results page, plus a list of likely matches (based on statistics I suppose). Try searching for 'polish' which could mean making shoes shiny, or central European people.
Other possible examples: "bat" (flying animals or sports equipment?), "rock" (stones, music, or maybe both? :P), "minute" (60 seconds, report, or something extremely small?). In all of these cases, the software can't tell the meanings apart very well yet. That's a problem with the software.