Looking for Language in Uncharted Territory
We have many historical examples of translating human languages, but these typically required 1.) a ground truth translation (think the Rosetta Stone) or 2.) willing and cooperative native speakers that assist in the learning process. When it comes to deciphering non-human communication systems, where do we begin?
​
The relevance of this question is larger than it seems. As humans, we are biological agents that exchange signals to coordinate behavior and reduce uncertainty of future outcomes. However, this is not a uniquely human problem. Anywhere that we see individual biological agents that must coordinate to solve problems and improve odds of survival, communication must exist. This clearly happens in other animal species, but also at a lower scale, in our cells.
​
What we need is an algorithm that can observe individual instances of communication and extract the logical rules of how those signals work. From here, we can draw lines around the 'syntax' of communication, mapping system inputs to meaningful outputs. Once successfully mapped, this will allow us to design our own inputs, effectively letting us 'talk' to our cells and influence macroscopic outcomes.
This toy example demonstrates two ways of representing the same set of sentences in different ways. By representing communication signals as graph objects, we can calculate feature of the language and signal transition probabilities without knowing ground truth! This method allows us to search for structure in signals that we don't yet understand by relying on mathematical features of graph objects.