From the Lojban web site FAQ (http://www.lojban.org/files/brochures/faq.htm):

lujvo: compound word

lujvo are based on tanru; they encode one conventionalised meaning of the semantically ambiguous tanru. tanru, in turn, are a sequence of brivla. This gives the (much simplified!) grammar:

  • selbri :- brivla | tanru
  • tanru :- brivla brivla+
  • brivla :- lujvo | gismu | fu'ivla | cmavo


And (semantically)

  • lujvo :- tanru


The use of cmavo as brivla, it should be noted, is marginal, and involves primarily anaphora, rather than distinct semantic content words.

This means that, underlyingly, all Lojban concepts are claimed to be expressable as a combination of gismu and fu'ivla. (This leads to the Gismu Deep Structure hypothesis.)

A lujvo is composed of a sequence of rafsi, concatenated according to a rather large set of morphological rules. Two strings of characters that are both a lujvo are considered the same word if and only if their rafsi belong to the same set of words in the right order. This means that mitpavycinglepre, mitpavycinletpre, mi'urpavycinglepre and mi'urpavycinletpre have no difference in meaning, and may alternate freely. However, it is only common to list the first alternative in dictionaries, because of the four, that's the one selected by the Tansky-Lechevalier scoring algorithm.

Of course. Anything we can't express with gismu and combinations thereof, we borrow as fu'ivla.

If it can't be expressed as a fu'ivla from ANY language, it cannot be (or has not been) said by humans.