Metameric is a simulator for Interactive Activation (IA) networks. Interactive Activation networks are localist connectionist models, which means that their neurons uniquely identify a single concept, e.g. one letter neuron for each letter, or a word neuron for each word. Interactive Activation was first introduced in McClelland & Rumelhart (1981), but has been used widely in the field computational psycholinguistics.
Unlike distributed connectionist models, which have been supported by a variety of useful tools and toolkits, no such toolkit exists for localist connectionist modeling. Metameric intends to fill this gap.
It is first and foremost meant to be a out-of-the-box simulator for the canonical IA model, but is easily extensible to other models, such as TRACE.
Metameric includes several innovations which allows the IA model to simulate stimuli of different lengths out of the box, and is, to our knowledge, the first simulator to do so.
These innovations are:
- Negative input features
- Space padding the input
- Weight adaptation
These three innovations are fully explained in the companion paper, which has been accepted for publication to the Mental Lexicon.
The name Metameric comes from the short story
The White Death by polish author Stanisław Lem:
The hereditary and at the same time perpetual ruler was Metameric, for he possessed a cold, beautiful and many-membered frame, and in the first of these members resided his mind; when that grew old, after thousands of years, when the crystal networks had been worn away from much administrative thinking, its authority was taken over by the next member, and thus it went, for of these he had ten billion.
First make sure all requirements in requirements.txt are installed.
pip install -r requirements.txt
conda install --yes --file requirements.txt
Then install using
python3 setup.py install
Then, you can run metameric with.
python3 -m metameric -i MY_INPUT_FILE -o MY_OUTPUT_FILE
For a quick example, use
You can also try normal preparation by running the
python3 -m metameric.prepare -i example_orth.csv -o example.csv -d orthography --decomposable_names letters -f letters --feature_sets fourteen
This turns a normal word csv with fields for orthography and frequency into data which can be fed into a full IA model. This can be used with the English Lexicon Project out of the box, for example.
python3 -m metameric.prepare -i elp-items.csv -o test.csv -d Word --decomposable_names letters -f letters --feature_sets fourteen --disable_strict
disable_strict is added because not all items in the elp are completely alpha-numeric, and hence can't be featurized by our feature set
You can also use the web interface.
python3 -m metameric.web
example.csv as a quick example.