Original Hypothesis - landart/LanguageEvolution GitHub Wiki

Back to Home

This is the original hypothesis developed, where the agents behave rather simply, it serves as a basis for developing more advanced systems.

Agents give unique names to unknown items, then share their dictionaries until they converge. There is a probability that an agent will accept another's agent dictionary items according if they are unknown to the first agent, or who was the last one that named the object (neologisms). Finally, there is a chaotic element that represents barbarian invasions and that tampers with the agents' dictionaries.

Rules:

• Each turn, each agent puts a name to one unclassified item nearby.
• When naming or renaming an item, the agent is its "owner".
• There is a 20% chance that another agent renames an "owned" item (neologism).
• The agents move randomly every clock tick, no matter what they do.
• When an agent meets another, the former mixes its dictionary with the latter's:
• The receiver's dictionary fills its gaps with the other agent's.
• If both have named an item, there is a 20% (possession) of adoption if the reciver is the "owner", 70% (neologism) if it is the other agent, and 40% (conservative) otherwise.
• Finally, the simulation should converge when all the agents have discovered all the items and agreed upon their names.

The Kraken:

The "Kraken" is a foreigner agent that causes chaos, destabilizing the system. Once all agents have agreed upon their dictionaries and the simulation has converged, you will be presented with a link to release the Kraken. Its behavior is simple:

• Once each turn, take an item randomly and change its name.
• If there are other agents in its range, it confuses them by replacing randomly each item in its dictionary with another random name with a 20% probability.
• The Kraken moves twice as fast as an agent and is range is almost double (5 cells Vs. 3).