How do we perceive, by signal or inference?

Class notes from July 17th.

[1] Signal vs. Inference
Acoustics often does not include the whole set of features. Ambiguity is inherent in acoustics.
“Inferential power of listeners”:
Stevens's view: how you get features is out of signal only. He is reluctant to refer to listeners’ inference. Recognition models (e.g. merge model of recognition, (Denis Noris) assume that signals carry everything. However, even if nasal sound is missing, listeners parse nasals. Listeners perceive words despite of undershooting of articulation ("Phonetic inferencing"). In my opinion, recognition model cannot include inferential abaility of listeners on its first stage. Recognition based on signal (only) and recognition by inference seems like the events that take place in two different modules. 'Recognition' is based on signal, a purely physical level. 'Inference' is based on knowledge, everything a person knows about the world.

Chilin: "There is an implicit way, a weighted model. If there is not enough information, context overrides. This is a linguists’ view of engineering."

[2] The Aerodynamics of nasality

Professor Keith Johnson demonstrated the equipments that measure oral/nasal pressures.
General purpose transducer amplifier has four channels.

1)? labial: oral air pressure goes up.
2) nasal air pressure
3) oral air pressure
4) acoustic channel

[3] We took a look at professor John Ohala’s record, using the equipments that measure oral/nasal pressures and DAQalzer. The recorded words are:
amba
damnation
pap’a (ejective) ….clipped
baba (implosive) ..minus pressure

(The transducer amplifier couldn’t capture ejectives’ pressure.)

A low vowel lowers soft palate, changes the volume of nasal cavity and resulting nasal flow.
(not the source of slight difference of F1 for nasalized vowels from non-nasal vowels)