Linguistically, that's really not a very good question. A word is not a concrete term, but rather a generalization made by speakers (and not linguists) to distinguish speech units on a pretty shady basis, the most solid of which is actually the writing system (i.e. "a word is what comes between two spaces"), thus, the phrase "cannot" will constitute one word, and the phrase "can not" two. Attempts to use a different basis will most likely drag you into the grey areas of the language, which linguists have a hard time with, such as determining what makes a 'set phrase', and whether it should count as one word or two.
That's not all — other than the elusiveness of the term word in the current language state (synchronous application), there's also the issue of language change over time and between dialects, not less elusive: there's no real way to compare words in ancient Greek or Hebrew to one another or to English words and say 'these are the same words'.
Someone above suggested a rather plausible test of recognition, however it probably still wouldn't do for a scientific test. A few reasons are:
- There's no real telling how the words sounded in the ancient languages.
- It's often hard to trace the route of a word between distant languages (e.g. if English got it from Greek, from Proto Indo-European, from Latin, etc.).
- Even if we did know the old pronunciations, it would almost never be exactly the same, and often less resembling than it seems in the first place. Mind that aside from the consonants and the vowels there are also differences in stress and intonation.
So, although from many aspects it is quite plausible to compare language to a biological system in terms of evolution, it's still not possible to determine 'the oldest word in a language'. A word is new every time it is uttered, and a language changes every day — we can call it by any name we please, but English today is not the same English as yesterday.