Long background story, not directly on topic at first, but brought about by the Whorf boondoggle and so at the heart of much of this talk. Early in the Twentieth Century, anthropological linguists started describing native languages "in their own terms" (i.e., not in term of Latin grammar) and they discovered many strange (to them) things, most of which they could absorb fairly easily, however (Finnish uses accumulated suffixes rather than declensions and prepositions, say, for the same thing). But some weren't so readily interpretable, in particular, cases where whole sentences seemed to be just single words. These words had to be verbs, of course (Latin was not completely dead and besides they had tense markers). So, since they described the world using only verbs and since verbs refer to processes, the speakers must see the world as processes (since they describe it using only verbs). Then a number of other language patterns found their place: speakers of these languages saw the world as embodies abstractions, or overlapping qualities or ...., because that was how their languages worked. So, there were many different observed worlds and no way to pick out reality and no way to really translate from one language to another, etc. etc. -- the whole linguistic relativism bit. In the latter half of the twentieth century, psychologists tested subjects from many languages and found that they all reacted pretty much the same in various situations where their languages would predict they would behave differently. Further, more sophisticated linguistic theories reexamined the exotic languages and found that, although their surfaces expressions were widely dissimilar, at their heart, they embodied the same basic structures and moved from these to the surface using moves from a relatively small toolkit of devices. That is, people who speak different languages don't really see the world differently (except in small ways -- they notice things that their languages make them notice, like different shades of blue in Russian, which others don't notice) and they don't put together the stuff of the world as observed in different ways either, except at very surface levels. To be sure, the usual representations of this common level -- as predicates and arguments, say -- are a product of Aristotelian imperialism, but the level and its contents would be the same even if represented in some other tradition (Chinese, say, or Hopi).
So, against that background, here is the problem. tp, in moving from this common core, uses a variety of transformations such that very different core claims end up being represented by superficially the same utterance (exactly the same words in exactly the same order). We can clarify a given case of this sort by going back a few steps and coming down to utterance by a different path. But it would be handy to have slightly different developments as the norm. We tend to single out -- because of notorious examples -- certain patterns as being especially unclear, but there are others as well. So, in the discussion over on another forum, I have focused on noun, verb, and adjective usage. These are not very precise but they make the point: 'moku' as "food", as "eating" and as "edible" (we miss "capable of eating", for example, which is also possible). And we notice that these three meanings (more or less) turn up in all possible positions: head noun, modifier, and predicate head. Let's take these three positions as the grammatical classes of noun, modifier and verb and the usages mentioned above as the semantic ones. Add to these the lexical classes (we'll figure out what to do with minor classes later), which are assigned mainly to ease figuring out the semantic classes in various lexical roles. So the problem is, how to make clear on the basis of the known lexical class and the observed syntactic role, what the semantic POS should be. And, at the moment, all tp has to offer is "context", which never works in isolated examples and often not in longer cases.
So, to take a classic case: 'jan li moku'. Barring some sneakiness about 'jan', which I will ignore for now (and which doesn't affect matters much in the beginning), this is a straightforward descent from noun + verb sentence ('li' obligatorily inserted and not obligatorily deleted later): "Human eats". But, of course, the same form of words, with the grammar noun + noun is "Human is food", which may come from some different structure like 'ijo li (ken) moku e jan' or from an original noun + noun (inclusion) pattern 'moku li jo e jan' or maybe something else altogether. Similar considerations lie behind "Human is edible", either involving different grammatical structures or different extractions of content from the generic "having to do with ingestion" that is the semantic core of 'moku', where the natural extractions (for us, at least) are the ingester, the thing ingested and the act of ingesting, with the act being primary. (Note that the ingester is never in tp 'moku' alone, for whatever that is worth. But the potential of being an object ingested is, for some reason. Go figure! Maybe that is just because it is the nearest thing to a relevant modifier.)
So, that is where I am coming from and I don't really quite see how it and your concerns come together for a meaningful discussion, but perhaps you have a better idea on that.