Understanding language through vision

We discuss briefly (from a philosophical point of view) why interfacing a linguistic analyzer with an artificial vision system is an important issue. In particular we claim that a linguistic analyzer supported by perception can actually understand, and not just process symbols. This is a possible objection against John Searle's well-known Chinese room argument. Secondly we present REF-MACHINE, a verification system architecture integrating a linguistic analyzer and a vision system. The global system can establish whether a sentence is true or false in a situation given through a symbolic description of a scene or directly in a real scene (given full perception equipment, a camera, and so on). This is a recognition system, partially implemented in LISP, incorporating an algorithm to interpret some locative expressions.