I've just made some test recordings of very low-level midrange sinusoids via the line input. It looks to me as if either it's dithering for 16-bit, or the input noise dithers the signal well enough.
ETA: I say that because no matter what level I chose for the test tone, I never saw any spurious harmonics. The classic sign of inadequate dithering is harmonic distortion that decreases and eventually vanishes as you inject increasing amounts of random noise around the level of one LSB. (IOW as you're looking at a spectrum analysis of the output from a poorly dithered recorder without any additional injected noise, the level of each harmonic visibly stands above the noise floor to some degree--but as you add very-low-level random noise and increase the level of that noise gradually, you can see the distortion decrease proportionally.)
(And by the way, contrary to the drivel that one still hears from some audiophile pundits, the decrease in distortion as the dither increases and reaches its optimal level is for real--it's an absolute decrease in distortion, not just relative to the noise level. In fact as you bring up the level of the dither, the levels of the distortion components come down distinctly faster than noise is rising. Dither prevents quantization distortion--it doesn't just "cover" or "conceal" it; and anyway, to do the latter would require a noise level some 10 to 20 dB higher, since people can hear signal components that are below the level of random noise.) [end of ETA]
I see no sign of noise-shaping; the noise floor has a very gentle upward slope throughout the audio range.
Spectrum graphs from Sound Forge are attached (the little blip around 15.7 kHz may be due to interference from a nearby video monitor). Please disregard the "-90" in the filenames; that's not accurate with respect to the signal level at the times shown in these pictures. I recorded at various levels from around -60 dBFS down to around -90 to see whether the typical artifacts of truncated samples would be present at any level, and I didn't find any.
The surprise for me was the small difference between the 16-bit and 24-bit noise floor--maybe 2 or 3 dB, but that's all. I don't get a clear impression that it's worthwhile to record 24-bit with this recorder, even though it supports the format. [ETA: I'm looking in to this some more. In these tests, the record level control on the M10 was set lower than it would probably ever be set in any actual recording situation; I want to do the tests over with a more attenuated signal and a more realistic setting of the record level control.]
--best regards