This started, as many things do, as a practical problem.
I wanted to translate something expressed in one domain into another without flattening it. Color into language. Language into music. Music into something resembling ethical posture. Producing analogies was easy enough. The harder question was whether anything recognizable could survive the passage.
At first this looked like a mapping problem. If red suggested urgency in one context, perhaps it should correspond to musical tension in another. If a musical resolution produced a sense of release, perhaps language could produce something structurally similar by moving from uncertainty toward clarity. Those correspondences are easy to invent. The trouble begins when we ask what makes one better than another.
Red doesn't contain urgency, and a chord progression doesn't contain generosity. Nothing inside either domain requires the correspondence I've chosen, and another person with a different history or cultural context might construct the mapping differently. So I can't simply discover the translation by inspecting the objects. I'm participating in its construction.
That doesn't make the experiment arbitrary. Some translations still work better than others for a particular purpose, and failed attempts can be surprisingly informative. A sequence can become cluttered because I tried to preserve too many features from the source. Another can become so abstract that almost nothing recognizable remains. Sometimes the mapping works beautifully until one element is changed, revealing that what seemed like a general rule depended on a very specific context.
The question gradually changes from “What does this correspond to?” into “What am I trying to preserve?”
Suppose I begin with a sequence of colors. I can preserve their order, relative brightness, contrast, emotional associations or relationship to surrounding colors. Those are different properties, and a translation into music probably can't preserve all of them at once. If I translate primarily by emotional association, I may select sounds that produce something like the movement I experienced in the colors. If I translate structurally, I might preserve contrast and sequence while producing a completely different emotional effect.
Neither approach is necessarily the true translation. Each tells me what I decided mattered.
This is where translation becomes an unusually good way to expose assumptions. As long as something remains in its original form, many of its properties arrive together and we don't have to decide which ones are essential. Translation forces the issue because the destination can't carry everything the source contained. Something has to be discarded, and the losses make our choices visible.
We encounter the same problem in ordinary language. A joke translated literally may preserve the words and lose the joke, while a translation that preserves the joke may have to abandon the original wording almost entirely. Neither result lets us conclude that wording doesn't matter. It tells us that different translations preserve different things.
The problem becomes stranger as the domains move farther apart. If I translate a visual sequence into music, there is no dictionary waiting for me. I have to invent some relationship between properties that can be represented in both places. Brightness might become pitch, intensity, volume or harmonic density. Movement across an image might become movement through time.
Each decision creates a little grammar for the experiment. Once that grammar exists, I can test it against something beyond the example that produced it. Does it produce useful results with a second image? What happens at an edge case? Can another person understand the mapping without being told what they're supposed to experience? If I change one of the rules, does the whole thing collapse?
Those questions interest me more now than whether a particular translation feels right.
Feeling still matters.