How linguists determine the sounds a speaker is performing

Question

So I see things like:

Sharanawa has /ɸ/ instead of /β/, and Shanewana has a labiodental fricative /f/ instead of /ɸ/.

where the table shows [β] as the symbol.

That, along with other examples like the dental t [t̪] and such, I'm wondering how the linguists actually figure out the details of how the sound is pronounced. I personally can't hear (yet perhaps) the difference between [t] and [t̪]. I can sort of hear the difference between [β] and [f], but it sounds like sometimes it can be difficult to tell.

So I'm wondering if they just do all this by ear and tape recording, or if they actually sit down with someone and say "show me where your tongue is when you make this sound" type of stuff. Or "are you closing your epiglottis to do that or is your tongue touching your soft palette." Like for the [ʂ] sound found in Hindi, I can sort of hear the difference between that and [ʃ], but you can just do an [ʃ] with a deeper sound without shaping your mouth that way, and the sound is identical to [ʂ] (at least from a hearing perspective, maybe not with a graph, I don't know). So in these types of cases I wonder how they disambiguate, if they are actually examining a speakers anatomy and having them demonstrate, or if it is just an educated guess. Also would like to know how many different people they include in the study, if it is just one or two, or like 10 or 20 or more.

That is, wondering if a linguist sits down, and, in addition to eliciting a bunch of words and stories, they try to elicit all the sounds from the IPA charts, or at least try to get the speaker to perform each of their languages sounds individually somehow, all the while examining their anatomy. Maybe they try [θ] and the speaker shakes their head "No.", "Okay, how about this one: [ʂ], see my tongue here -^", etc.

The figure below also demonstrates how the analysis can lead to different results.

Answer 1

That is the proverbial $64 question. Ideally, there is a standard and rigorous training course that all linguists take where they get ear-training to hear the various sound distinctions made by languages. The reality is that this does not happen, so some degree of skepticism is warranted, especially if a source simply asserts a particular phonetic value. Most of the time, assertions about articulation are based on the premise that a certain sound is caused by a certain articulation, and if you have been well-trained in epiglottals versus pharyngeals, then the claim may be correct, but otherwise you have to take articulatory claims with a grain of salt. Reconstructing articulation based on acoustics can be accurate up to a point, but only to the extent that the linguist is trained in the myriad sub-variants of now retroflex consonants are produced, for example.

Sometimes one can directly measure articulation – you can look to see what the lips are doing. Typically, if a study has an actual articulatory underpinning, the source publishes those results. (Wiki pages are, of course, not scholarly publications and are usually assembled by other people based on a publication). In the case that you mentioned, the Wiki page claims to get the data from a Berkeley inventory database, who in turn claim to have gotten their information from Faust&Loos Gramática del idioma yaminahua, which miraculously is actually online. However, what you get from that source is a list of sounds and very little discussion of phonetics – no articulatory evidence.

Some people do attempt to elicit speaker intuitions about articulation, but I have found that to be usually unreliable.