In the context of epistemology and pramāṇa theory, we often translate śabda as “testimony.” It is reasonable to do so, since using “word” in sentences like “Word is a genuine source of knowledge” is unpleasant to the ear and confusing to modern readers. That said, it strikes me that the Indian thinkers use of śabda makes good sense. Primarily, this is because “testimony” seems to imply situations where one person is self-consciously teaching another person; yet these are really a subset of the more general way that we gain cognition from the denotative power of language. Much of our day is spent in conversation where we give and take knowledge, typically unselfconsciously, and largely without either side approaching the situation in the way we would approach being taught by an expert. When we do learn from experts, this is just a special case of learning from language. I’d opine that this is why some medieval Sanskrit teaching texts discuss philosophy of language as nested within discussion of śabda as a pramāṇa, something that may strike modern thinkers as mixing together two separate domains of philosophical inquiry. What is central is the denotative power of language and how it can generate true cognition. It is what underwrites all instances of learning “from testimony.”
Secondly we can learn from words without them being an expression of another’s true beliefs in the way that “testimony” seems to presuppose. Admittedly this point is argued and debated both classically and in modern philosophy; but to use an example I find compelling: I can look up the distance between two places in Google maps, and for the sake of argument let’s say no one has ever asked or thought about the exact distance between the two places before. The program then algorithmically generates words from which I learn something true; words which in no way express a belief that was previously held by anyone. Here, I learn from words, but not exactly from an assertion or from testimony unless we want to significantly adjust what we mean by the latter two terms. The reason I learn from Google maps in this case is because the words that are being generated do in fact denote certain places and accurately express the distance between them. That’s it.