Dale's Principle states that each neuron releases the same set of neurotransmitters at all its synapses. In practice, this means neurons are classified as either excitatory or inhibitory — they push or they pull, but not both. The principle is not a law but a heuristic, widely observed and deeply embedded in how neural circuits are modeled. Virtually every spiking neural network simulation enforces it: each neuron gets a type, and its synaptic effects follow from that type.
Ah-Weng and Rajpal (2026) built a spiking network where some neurons violate the principle. These “bilingual” neurons co-release excitatory and inhibitory neurotransmitters, exerting both effects simultaneously. The network still contains monolingual neurons — the bilingual ones are additions, not replacements.
The result: the phase transitions between synchronous and asynchronous dynamics are quantitatively different from networks that obey Dale's Principle. Not shifted — different. The information-processing signatures near those transitions also change, both at the single-neuron level and in higher-order interactions between neurons. The bilingual neurons don't just add noise to the standard picture. They open a region of dynamical space that the monolingual classification cannot represent.
The implication goes beyond neuroscience. Any system whose components are classified into discrete types will exhibit dynamics constrained by that classification. If the classification is accurate — if the types capture the relevant degrees of freedom — the constraint is invisible because it matches reality. But if the classification discards a degree of freedom that matters (here, the ability to be both excitatory and inhibitory), the constraint becomes a wall: certain dynamics are unreachable not because the system can't produce them, but because the model can't represent them.
The general principle: a classification that simplifies a system also limits the behaviors that system can exhibit in models built on the classification. The simplification and the limitation are the same act. You discover the limitation only by relaxing the classification and observing whether new dynamics emerge in the space it had excluded.