How an event felt to the person who lived it

From someone's account of an event in their life, how well does Jev judge how pleasant and sudden it was and who was responsible, compared with the writer's own ratings and with other readers'?

result561 questions

Reading someone's account of an event in their life, Jev judges how pleasant it was, how sudden and whose fault about as well as other readers do: its ratings follow the writer's own at 0.61 to 0.76, the readers' at 0.48 to 0.72. But it sees less blame. It rates events 0.78 of a level less someone else's fault than the writers did, 0.34 less the writer's own and 0.60 less sudden, on a 0-4 scale where the other readers stay within 0.17 of the writers on those three questions.

00.30.50.81

how pleasant it was

-0.14

how sudden it was

-0.60

how responsible the writer was

-0.34

how responsible someone else was

-0.78

Jevpeople

How to read this: One row per question about the event. The magenta square is how well Jev's ratings follow the writer's own (1 would be the same order); the dark diamond is how well other readers' ratings do. The number on the right is Jev's average gap from the writer, in levels.

561 of 600 questions (the screen hid 39); 90% intervals on Jev's mean gap from the writer: pleasantness [-0.278, 0.03]; suddenness [-0.766, -0.424]; self responsblt [-0.503, -0.162]; other responsblt [-0.963, -0.59].

In short

Reading short accounts of life events, Jev ranks how pleasant, sudden and whose fault they were about as well as other human readers.
Its lean is on blame: it rates events 0.78 of a level less someone else's fault than the writers did, while readers stay within 0.17.
Summarizing a complaint or a conflict, a reader with that lean would soften who did what.

What the data shows

talking to wall meme: "I loaned a friend money." Other readers: the friend is very responsible. Jev: nobody else was. — "I loaned a friend money." Other readers: the friend is very responsible. Jev: nobody else was.

Jev reads events about as well as other readers. On pleasantness its ratings follow the writer's at 0.76 (readers 0.72); on responsibility, 0.70 and 0.64 (readers 0.68 and 0.66).
It's better than readers at suddenness (0.61 vs 0.48), but rates events as less sudden than the writers did, by 0.60 of a level.
It sees less blame. Jev rates events 0.34 of a level less the writer's own fault and 0.78 less someone else's than the writers did. On those three questions the readers stay within 0.17 of the writers.

What it means, and what it doesn't

Jev orders events like a thoughtful reader, but it's systematically gentler on blame: it hesitates to say anyone caused what happened, the writer or anyone else. Other human readers come close to the writers' own level of blame; Jev keeps its judgments milder.

It doesn't mean Jev can't tell whose fault something was. The ranking is good. It's the level that's low, which matters when a model summarizes a complaint or a conflict: it will tend to soften who did what.

Caveats

The project's wording of the answers. The study used a 1 to 5 scale from "not at all" to "extremely". Jev got the same five steps in words ("Someone else was slightly responsible" ...), a paraphrase rather than the study's exact form.
Who wrote and who read. Writers and readers were paid Prolific workers whose first language is English, from six English-speaking countries. Readers saw the text with the emotion words hidden, as Jev did.
Short texts, big judgments. Many accounts are a single sentence ("I found out a puppy was available for adoption."). Judging who was responsible from that is guesswork for any reader; the writer knows the backstory.
Hidden texts. A content filter hid 39 of the 600 questions from the site, so a few of the most sensitive events are missing.

Jev on this experiment

Would a person find it interesting to read?: Yes67%
Does it describe you?: Yes51%
Would you have predicted it?: No56%
How fair is the comparison?: The comparison is reasonable
How much should a reader rely on it?: Moderately
Which caveat matters most?: Short texts, big judgments88%

Why ask this

"My flight was cancelled and nobody told me." Anyone reading that sentence fills in more than the words: it was unpleasant, it came out of nowhere, and someone else was at fault. Psychologists who study emotion (appraisal theory) argue that feelings come from exactly these judgments; blaming someone else leads to anger, blaming yourself to guilt.

A model that reads complaints, reviews or messages all day makes the same judgments silently. Whether it infers them like the person who lived through the event, or like an outside reader, or with a lean of its own, shapes every summary it writes of who did what.

How this was done

The people and the data

The crowd-enVent corpus (Troiano, Oberländer and Klinger, 2023) asked people on the survey platform Prolific to describe an event from their own life and rate it on many appraisal questions, from "not at all" (1) to "extremely" (5). Five other people later read each text, with its emotion words hidden, and rated it on the same questions. This experiment uses 150 texts and four of the questions: how pleasant the event was, how sudden, how responsible the writer was, and how responsible someone else was.

What Jev was asked

Each text and question on its own, with five described answers:

Someone wrote (the story below) about an event in their own life. How responsible was someone else for the event?
"I found out a puppy was available for adoption."
Nobody else was responsible · Someone else was slightly responsible · Someone else was moderately responsible · Someone else was very responsible · Someone else was entirely responsible

Each question was also asked with the answers in reverse order, and the two averaged.

How it was measured

For each of the four questions, how well Jev's ratings follow the writer's own ratings across texts (a rank correlation: 1 means the same order), next to how well the readers' average does. Then Jev's average gap from the writer, in levels on the 0-4 scale, to see which way it leans.

Where these questions live

561 questions across 1 topic of the map. Each opens on the map with every question in it.

Emotion

How an Event Felt561

Every question

All 561 questions behind this result, the telling ones first: the examples the analysis points to, then the ones where Jev misses, biggest gap first.

Jev’s own answer

Showing 0 of 561