.. toctree:: :maxdepth: 2 :caption: Contents: using_adam
.. autosummary:: :toctree: _autosummary
ADAM is a demonstration of "grounded language acquisition," which is to say learning (some amount of) language from observing how language is used in concrete situations, like infants (presumably) do.
In the next section, we present an overview of the system architecture. That is followed by low-level API documentation. For installation instructions and information about contributing, please see the Markdown files in the project's Git repository.
ADAM is ISI's effort under DARPA's Grounded Artificial Intelligence Language Acquisition (GAILA) program. Background for the GAILA program is given in DARPA's call for proposals and here is a video of a talk giving an overview of our plans for ADAM (aimed at an audience familiar with the GAILA program).
The goal of ADAM is to build a language learning module which can
- learn (parts of) human language by observing a sequence of situations paired with situationally-appropriate language (usually a description of the situation).
- be presented with novel situations and give reasonable linguistic descriptions of them.
For example, if the learner has been shown instances of various objects on a table and has also been shown instances of a toy truck, then when given the situation of a toy truck on a table, it should be able to describe it as truck on a table even if it has never seen that particular situation before.
A few aspects of ADAM's approach worth noting are:
- there are several ways one could represent a "situation." Rather than using, for example, videos of real-life situations, we represent situations by a data structure in the computer's memory.
- it is implausible for the learner to observe the data structure representing the situation directly. Instead, the learner will observe a perceptual representation derived from the situation and based on those perceptual primitives that there is evidence an infant has access to.
- in the GAILA program, researchers are permitted to control the type and sequence of example situations which are given to the learner (the curriculum). Rather than specifying a curriculum manually, we provide ways to procedurally generate curricula.
A particular "run" of the ADAM system is described by an Experiment. Every Experiment needs to know
- what TopLevelLanguageLearner to use,
- what Situations to present to the TopLevelLanguageLearner to teach it,
- what Situations to present to the TopLevelLanguageLearner to evaluate it,
- what sort of analyses to perform on the result.
There are a variety of ways to specify the situations for training and testing, but this is prototypically done by generating them procedurally using a SituationTemplateProcessor. The way a Situation is presented to the TopLevelLanguageLearner is controlled by a LanguageGenerator and a PerceptualRepresentationGenerator.
The analyses to perform on results are given by DescriptionObservers.
The ADAM system's main entry point is adam.experiment.log_experiment.
For details on this see :file:`adam/experiment/README.md`.
In addition, there may eventually be an interactive demonstration.
.. automodule:: adam.situation
.. automodule:: adam.language
.. automodule:: adam.perception
.. automodule:: adam.learner
.. automodule:: adam.semantics
.. automodule:: adam.situation.templates
.. automodule:: adam.situation.templates.phase1_templates
.. automodule:: adam.experiment
.. automodule:: adam.learner.alignments
.. automodule:: adam.learner.surface_templates
.. automodule:: adam.relation
.. automodule:: adam.geon
.. automodule:: adam.axes
.. automodule:: adam.axis
.. automodule:: adam.ontology.structural_schema
.. automodule:: adam.ontology.during
.. automodule:: adam.ontology.action_description
.. automodule:: adam.ontology.phase1_spatial_relations
.. automodule:: adam.ontology.phase1_size_relationships
.. automodule:: adam.situation.high_level_semantics_situation
.. automodule:: adam.ontology
.. automodule:: adam.ontology.ontology
.. automodule:: adam.language.ontology_dictionary
.. automodule:: adam.language.language_generator
.. automodule:: adam.language.lexicon
.. automodule:: adam.language.dependency
.. automodule:: adam.language.dependency.universal_dependencies
.. automodule:: adam.perception.developmental_primitive_perception
..automodule:: adam.perception.high_level_semantics_situation_to_developmental_primitive_perception
.. automodule:: adam.perception.perception_graph
.. automodule:: adam.perception.marr
.. automodule:: adam.curriculum
.. automodule:: adam.curriculum.curriculum_utils
.. automodule:: adam.experiment.observer
.. automodule:: adam.ui
.. automodule:: adam.math_3d
.. automodule:: adam.random_utils
.. automodule:: adam.ontology.phase1_ontology
.. automodule:: adam.language_specific.english.english_phase_1_lexicon
.. automodule:: adam.curriculum.phase1_curriculum
.. automodule:: adam.language_specific
.. automodule:: adam.language_specific.english
.. automodule:: adam.language_specific.english.english_language_generator
.. automodule:: adam.language_specific.english.english_syntax