Working from most recent backward in this thread:
Entries in the ERG’s “lexicon.tdl” file must each provide at minimum the name of the entry, its lexical type, its orthography, its semantic predicate (except in rare instances), and its PHON phonological onset (vocalic or consonantal), as Mike mentioned, so the grammar knows whether this word might be preceded by “a” or “an” (as in “an honest mistake”). For verbs, adjectives, and nouns that take a complement, the lexical entry may also specify some idiosyncratic property of that complement, via the --COMPKEY or --OCOMPKEY (oblique complement) attribute, typically for verbs that select for a specific particle or prepositional phrase; for example, the verb “look” used in “look up the answer” is of type v_p-np_le and requires its first complement to be “up” via its --COMPKEY value.
Users of the ERG are of course welcome to add their own entries to the lexicon provided, but need to remember that if one adds a new entry, then all words with that same stem must also be defined, since the POS-tag-based handling of unknown words is disabled for any stem that has at least one entry in the manually-defined lexicon.
Now on the main question about the lexical ambiguity for “finished” and similar words: I agree with Alexandre and Guy that the ERG is currently missing an entry for “finished” as an adjective. Apart from often fuzzy intuitions about word senses, there are a few syntactic tests to distinguish adjectives from passivized verbs, and for at least one of these tests (combining with “… enough to…”), “finished” can be an adjective: “This article is finished enough to submit as a short paper”, though it for some reason does not like the degree specifier “very”, another usually reliable test for adjectives: “?This article is not very finished yet”.
It would nice to have some systematic way of discovering a relatively complete inventory of adjectives that have the surface form of a passivized verb; the list of ones included in the ERG lexicon is, I fear, far from complete.