I recently acquired a copy of Murrinhpatha Morphology and Phonology, partly because it was relatively cheap on Amazon, and partly because I’ve been interested for a while in the structure of the finite verb stem in similar / related languages. In short overviews (e.g. the chapter of Murrinhpatha in The Oxford Handbook of Polysynthesis), the structure of the finite verb is usually not described in much detail.
And the reason why is basically that it’s a bit of a mess. Murrinhpatha only has 39 finite verbs (with most other verbal notions being expressed by verb+coverb structures), but those verbs are marked for 3 subject persons, 3 subject numbers and four core tense/mood categories by combinations of prefixes, suffixes, and stem mutation/suppletion that produce 34 distinct exponence patterns. This means that there are almost as many ways to inflect a finite verb in Murrinhpatha as there are actual verbs.
Even more confusingly, many of the stem mutations and suppletions are not unique to an individual verb either. The stem consists of at most one syllable and at least a single phoneme, and many of these stems appear in the paradigm of more than one verb. I’ve done some counting of how often each stem appears, to illustrate, and in the worst case the same stem is associated with the paradigms of six different verbs:
na (6): be, use feet, use feet (refl), use hands, say/do, use fire a (6): stand, pierce/use mouth, appear, put together, use fire, watch ɾi (5): sit, lie, stand, go, watch (refl) ɾa (5): stand, go, pierce/use mouth, appear, watch ni (5): be, use feet, use feet (refl), say/do, look u (4): go, coerce, slash, slash (refl) i (4): sit, lie, carry, watch (refl) ɖi (3): be, impel, impel (refl) ŋa (3): put together, pull, pull (refl) nna (3): use feet, use feet (refl), use fire ne (3): use hands, say/do, say/do/use hands (refl) me (3): use hands, say/do, say/do/use hands (refl) ju (3): sit, lie, slash (refl) je (3): go, slash, slash (refl) da (3): affect, lower, lower (refl) bi (3): lie, stand, hear ba (3): affect, lower, lower (refl) ɻi (2): stand, watch (refl) ɻa (2): stand, watch ɖɖi (2): impel, impel (refl) ɖɖe (2): impel, impel (refl) ɖɖa (2): impel, impel (refl) ɖe (2): be, impel ɖa (2): impel, impel (refl) ŋi (2): pull, pull (refl) n̪t̪i (2): perch, carry n̪t̪a (2): perch, crouch nu (2): use feet, coerce nni (2): use feet, use feet (refl) mpa (2): go, slash mi (2): say/do, look ma (2): use hands, say/do lla (2): wipe, eat la (2): wipe, eat j (2): slash, slash (refl) de (2): affect, affect (refl) be (2): affect, affect (refl) ɾu (1): go ɾe (1): pierce (refl) ɻu (1): go ɳi (1): be ɳe (1): be ɳa (1): be ɖɖu (1): impel ɖu (1): impel ŋu (1): pull ŋe (1): pull (refl) n̪t̪e (1): carry nuj (1): use feet nnu (1): use feet nne (1): use feet mu (1): coerce lli (1): wipe (refl) lle (1): eat li (1): wipe (refl) le (1): eat ji (1): go e (1): pierce (refl) duj (1): lower (refl) du (1): lower di (1): hear buj (1): lower (refl) bu (1): lower
And this is where the large number of paradigms comes in. Some of these clashes will be resolved by different forms of the following TAM suffixes, which are typically a single syllable with a stable onset but a variable vowel afterwards. Some will be resolved by the nature of subject agreement: in the singular both prefixation of a CV syllable and stem consonant mutation are used, whereas the plural can trigger either gemination or suppletion. In both cases, the prefix vowel can vary depending on the underlying verb. The end result is that, in many cases, no one component of the word alone identifies which verb an inflected form is associated with, but the form nonetheless is unambiguously associated with only one verb.
What’s also impressive is that this semi-irregularity covers quite large paradigms of 42 forms per verb. It’s not quite as bad as memorising 39*42=1638 verb forms individually, since not only are there only 34 exponence patterns for 39 verbs, but there are repeating structures within some parts of those patterns. But it’s impressively arbitrary enough that I wouldn’t have thought it was that plausible if I didn’t know it already existed.