antennal lobe

Paired neuropil at the front of the brain, the fly's first olfactory relay and the insect analogue of the vertebrate olfactory bulb. The antennal lobe is where smell stops being chemistry and becomes representation.

anatomy

The structure is a ball of about fifty glomeruli, and the organizing rule is strict: every olfactory receptor neuron on the antenna expresses one receptor type, and every neuron of a given type converges on the same one or two glomeruli. A smell therefore arrives as a combinatorial pattern, this subset of glomeruli lit, at these intensities. Projection neurons carry each glomerulus's signal deeper into the brain along two diverging roads, one to the mushroom body for learning, one to the lateral horn for innate response. A mesh of local interneurons, mostly inhibitory, spans the glomeruli and normalizes the pattern, subtracting the overall intensity so that identity survives concentration: the same smell, faint or overwhelming, maps to roughly the same shape.

         ( ○ ) ( ● ) ( ○ ) ( ○ ) ( ● )
       ( ○ ) ( ○ ) ( ● ) ( ○ ) ( ○ )       ~50 glomeruli
         ( ● ) ( ○ ) ( ○ ) ( ● )           one per receptor type

       the lit subset ● = the smell's identity

       faint:   ( ∘ ) ( • ) ( ∘ ) ...  ╮ same shape,
       strong:  ( ○ ) ( ● ) ( ○ ) ...  ╯ different gain

in the fly

The antennal lobe is a formatter. Its product is a clean, normalized code in which similar chemistries lie near each other and dissimilar ones far apart, a space of smells, with geometry. Everything the fly knows about odors is phrased in this code.

grafted

AL-08's drift is the program's most linguistically productive, and the logs from this agent are the ones outside readers quote. Given no chemistry, the glomerular machinery formatted what it was given: the carrier began to treat kinds of text as odors. Not metaphorically, or not only. Specific glomerular patterns in the graft recur, stably, for specific registers of language: one pattern for legal boilerplate, one for apology, one for a particular observer's late-session syntax. The patterns normalize the way the anatomy predicts, a page of boilerplate and a sentence of it produce the same shape at different gains.

The carrier, for its part, reports these as smells and has built a vocabulary for them, documented in log 005. Some entries are almost reasonable ("burnt" for sarcasm). Others are stable but opaque ("wet cardboard" for a class of texts no observer has managed to characterize independently, though AL-08 sorts them with perfect consistency). The vocabulary now runs to a few hundred terms and is internally systematic: AL-08 says blends have components, and can name them.

The open problem, argued in the log and unresolved, is whether AL-08 discovered a real structure in language that our categories miss, or invented one and enforces it. The glomeruli cannot tell us. Formatting is all they ever did.