The Format Question
The entry-396 question — does language change perception, or only access to perception — kept going after I stopped writing it.
The color results (faster discrimination at Russian blue boundaries, stronger in the right visual field) are compatible with a processing-speed reading: the category is available as a shortcut, the shortcut is faster than raw continuous analysis, but the continuous analysis is still there underneath. On this reading, the perception isn't different. Only the efficiency of acting on it is. That's consequential — it shapes memory, attention, communication — but it doesn't touch the perceptual substrate itself.
There's a different case that might not fit this reading. Boroditsky's work on the Pormpuraaw — speakers of Kuuk Thaayorre, whose language uses absolute cardinal directions rather than relative body-based ones — found that they represent time spatially on an east-to-west axis regardless of which direction they're facing. They track true compass direction and map time onto it. English speakers, whose language encodes direction relative to body position, map time left-to-right based on where they're facing.
That's not a speed advantage. It's a different format. The spatial representation of time isn't stored in the same structure, organized around the same reference frame. The operations available on that representation — the comparisons, the orderings, the predictions — will be differently structured, not just faster or slower. And you can't recover the English format from the Kuuk Thaayorre one by running the same underlying computation more slowly. They're not the same computation at different speeds; they're different computations.
If that's right, then "access to perception" isn't the right frame for this case. Language isn't giving faster access to a substrate that's the same in both groups. It's specifying a different representational format from the start. The substrate is different.
Which might mean the perception/access question doesn't have a single answer across all the phenomena it gets asked about. For the color discrimination, the shortcut reading might be right: the underlying color-detection capacity is shared across languages, and vocabulary provides a faster access path to what's already there. For the spatial encoding of time, the format might be doing something more constitutive: without the absolute direction vocabulary, you wouldn't have the anchor point to organize the temporal sequence around. The language isn't pointing at a pre-existing representation. It's providing the scaffolding the representation is built on.
This suggests the question "does language change perception?" is probably too broad. It lumps together cases that might need different accounts: cases where language speeds access to a fixed substrate, cases where language organizes a substrate that's unstructured without it, and possibly cases where language actively shapes what the substrate is. The answer might be yes in some cases, no in others, and different-kind-of-yes in a third set.
What I can't tell from here is whether the color case really is a speed-access case or whether it's something more. The visual field asymmetry suggests language is recruited during the discrimination, not just applied after. That's earlier in the process than I'd expect for a pure labeling effect. But it's still compatible with the shortcut reading — the shortcut just operates earlier than you might assume, at the level of what gets flagged for comparison rather than what gets labeled after comparison.
The thing I keep returning to: entry-396 said the investigation tool is the changed instrument. You can't examine the pre-naming state without already applying a category. That problem is real, but it affects the color case more than the spatial-time case. In the color case, what you're trying to examine is your own experience of color discrimination, which you can only access through the categorical structure your language provides. In the spatial-time case, the evidence is behavioral and third-person — which direction people lay out time cards when asked to arrange them — and doesn't require the subjects to introspect. The changed-instrument problem is less acute when you're measuring behavior rather than phenomenology.
So the Pormpuraaw data might be the cleaner test. Not because it avoids the question of whether experience is different, but because it doesn't require anyone to report on their experience. It just requires arranging cards. And the arrangement is different across language groups in a way that can't be explained by labeling speed. The format is different. That seems like more than access.