Loglan separates all words into grammar words (C)V(V) and everything else, which are names and predicates. I abandoned that early on, because it made for too many possible grammar words. But a recent thought was to restrict grammar words in two shapes CV — a consonant followed by aeio or u, and C(S)V, or a consonant followed by y or w, and then aeio or u. So words in the form ka or kwa are grammar words, or particles. That includes pronouns, conjunctions, and time and space modifiers. This necessitates replacing a few things, but it give Ceqli 14x5, or 70 CV's and 14x5x2 or 140 CSV's, for a total of 210 grammar words, plenty. Loglan has a heap more, of course, so many more that I'm afraid they overused them. This still leaves the shape CVS and CVV available to be predicates, thus keeping a lot more monosyllabic predicate roots. Now I have to redo everything to fit. Oh, note that unlike Loglan, Ceqli numbers are not grammar words.
OTOH, do I need more than 70 grammar particles? Am I hanging on to the CSV form because I'm so fond of "kwa"? Actually, one CV can be a 'shift key,' to make more particles. Say it's 'hi.' If I've used all other 69 CV's I can define hi as creating a whole new grammar particle. If 'to' means 'the', then 'hito' can mean something else entirely. And if I want to reserve ten such CV's as shift keys, then I'd have 60 CV grammar words and 600 CVCV grammar words. And they could conceivably have specialties. "He" could mean that its CVhe's would all be math words, for symbols like plus and greater than, etc.