Two readings of the same words

1

A study begins with a question

It asks one question of a place or of a people, through a few prompts anyone can answer aloud. Living with the Bassin asks what the Bassin d’Arcachon means to the people who live around it all year.

Before anyone takes part, its prompts and its questions are sealed: registered with a hash, so what was asked cannot drift once the answers start coming in.

2

Each person answers in their own voice

They open a link of their own and talk to an interviewer that speaks. It asks every prompt exactly as written, and waits until they have finished: a pause to think is not an end.

Then it listens. A model notes what opened in the answer (here, 3 things) and may ask about one of them, most often for one particular time. Code keeps the question only if it is a single open question made of their own words: no word that carries meaning they did not say, nothing they said turned around.

After the last prompt it may return once to what was left most open, and then asks what they would call what they told, in a few words of their own: their names for their answers, given before they have heard any of the study’s terms.

3

The recording becomes passages

What was said is written down, and the recording is cut where each answer began and ended. Every passage knows which prompt it answers, and plays from its own first second.

What the machines heard is kept as heard: in this persona’s second answer, the herons became “errands”.

4

The method reads every passage

The study’s method is written from its questions. Each is put to a passage as a yes or no, by Jev, a literal reader, which answers with how sure it is.

Does the speaker in passage say that they belong to the Bassin, or that it is part of who they are?

Everything anyone recorded is read this way: here, 32 passages from 8 people, 3 questions each.

5

Few moments are asked about

Each dot is a passage, as the method read it for one question. Nobody is asked about everything: each person hears back 4 of their own passages, chosen where the method is least sure, one question each.

The last is chosen at random, so the study also sees moments the method found easy.

6

Heard back, and answered blind

They listen to each moment and answer the study’s own question about it: yes, no, or not sure. They never see the method’s reading first. Theirs is the reading the method will be held to.

“Not sure” is kept, and never counted as either.

7

Two readings in one figure

Each moment is now the teller’s: above the line if they said yes, below if no. Across, how sure the method was; right of its line, it says yes.

A dot above the line and right of it, or below and left, is a moment both read alike. On belonging, that is 4 in 16.

8

The tellers move the method’s line

Where the method starts to say yes is a setting, one per question. Fitted to what people said of their own words, the line moves to where it agrees with them most.

On nature at risk, from 0.50 to 0.15: 7 of 8 become 8 of 8. Kept, that is the method’s next version: its questions unchanged, nothing read again.

9

Some gaps no line can close

The tellers said yes to belonging in 13 of 16 moments; the method, in 1. Fitted anyway, its line drops to 0.05, and it hears belonging in 72% of every passage anyone recorded: agreeing by saying yes to nearly everything.

So the gap is the finding. The method asks of each passage, “Does the speaker in passage say that they belong to the Bassin, or that it is part of who they are?”; the tellers mean more than that question hears. Asked what it missed, one said:

“It’s freedom. Nobody owns the water.”

P-07, oyster farmer at Larros, back at thirty (a persona)

10

More than yes or no

After answering blind, people can say more. They place each moment among the study’s three questions, nearer the ones it is more about. They say which of two of their own passages is more about one. They put a moment in their own words, and then see the method’s reading and say what it missed.

None of it is counted as their answer; all of it is kept beside the method’s reading. Here, the method ordered 14 of 15 pairs as the personas did, and 7 more they judged about the same.

11

The dial

Some hear their whole recording again, a fader under their thumb, and push it up whenever what they hear is about one question. Every second gets their own reading, not only the few moments the method was unsure of.

Here P-07, oyster farmer at Larros, back at thirty, pushes it up on belonging wherever they meant it, and lets it fall where they did not. The method’s reading of the same passages stays below its line in every answer: the gap again, second by second.

A dial is kept beside the method’s reading, never counted as an answer: a level is not a yes or a no.

12

Everyone sees where they land

Before the method’s reading of them is shown, each person says where they would put themselves: for each question, how much of everything they told was about it. Their last page then draws everyone who took part, each a dot, on any two of the study’s questions, or along the two directions people differ most.

They are there twice: where the method hears them in all they said, and where they put themselves. P-07 is heard on nature at risk in 50% of what they told, and put themselves at 43%. Across the 8 personas the two put people in most nearly the same order on nature at risk (a rank correlation of 0.85), and least on taken over (0.31).

Places only, never anyone’s words; and a place, like a dial, is kept beside the method’s reading, never counted as an answer.

13

Their words become questions

What people said their moments were about went to a model, which grouped 4 of them into one candidate, asked only in words they used: “Is this moment about whether the people who work here can live here?”

It was tried on all 32 passages, and passed a gate of code: its own people’s passages read higher than the rest, it is not a question the study already asks, and it is heard in more than one passage but not in all. So it was added, logged as an amendment: the method’s version 0.2.0, every passage read again with it.