eta.transducers.base

Contains abstract base transducer classes for each supported transducer.

A transducer is an object that maps some set of arguments to some value; multiple abstract transducers are supported for various types of mappings with different function signatures.

The abstract transducers defined in this module do not actually implement any behavior; it is expected that they will be extended with transducers that use various techniques to perform a mapping (e.g., rule-based tree transduction, statistical models, large language models, etc.).

Note that each transducer type should return a list; this allows multiple transducer implementations to be “stacked” in practice, e.g., using both a tree transduction and an LLM transducer and collating the results.

Classes

AffectTransducer	Maps the word content of an utterance to an appropriate affect, given a conversation history.
AnswerTransducer	Maps a conversation history and relevant knowledge to an answer utterance.
AskTransducer	Maps a conversation history and relevant knowledge to a question utterance.
GistTransducer	Maps an utterance to a list of gist clauses, given a conversation history.
ParaphraseTransducer	Maps a gist clause, given a conversation history and relevant knowledge, to a paraphrased response.
PragmaticTransducer	Maps a gist clause to a list of ULF formulas capturing its pragmatic meaning.
ReactionTransducer	Maps an observed event to a formula for an action that an agent can take.
ReasonBottomUpTransducer	Maps from a list of facts to a list of new facts inferred from those.
ReasonTopDownTransducer	Maps from a plan step event and a list of facts to a list of new facts inferred from those.
ResponseTransducer	Maps a conversation history and relevant knowledge to a response utterance.
SemanticTransducer	Maps a gist clause to a list of ULF formulas capturing its semantic meaning.
SubplanTransducer	Maps a plan step event to a list of subplans, i.e., sequences of events that are substeps of the given step.
Transducer	The base transducer class.

class Transducer

Bases: object

The base transducer class.

cost()

Report cumulative costs of applying this transducer.

class ReasonTopDownTransducer

Bases: Transducer

Maps from a plan step event and a list of facts to a list of new facts inferred from those.

Parameters:

• step (Eventuality) – The event corresponding to the plan step to use as context for inference.

• facts (list[Eventuality]) – The list of facts to make inferences from.

Returns:

A list of new facts inferred from the plan step and given facts.

Return type:

list[Eventuality]

class ReasonBottomUpTransducer

Bases: Transducer

Maps from a list of facts to a list of new facts inferred from those.

Parameters:

facts (list[Eventuality]) – The list of facts to make inferences from.

Returns:

A list of new facts inferred from the given facts.

Return type:

list[Eventuality]

class GistTransducer

Bases: Transducer

Maps an utterance to a list of gist clauses, given a conversation history.

Parameters:

• utt (Utterance) – The utterance to interpret as a gist clause.

• conversation_log (list[DialogueTurn]) – The conversation history to use as context for interpretation.

Returns:

A list of gist clauses.

Return type:

list[str]

class SemanticTransducer

Bases: Transducer

Maps a gist clause to a list of ULF formulas capturing its semantic meaning.

Parameters:

gist (str) – The gist clause to parse to a logical form.

Returns:

A list of ULF S-expressions.

Return type:

list[s-expr]

class PragmaticTransducer

Bases: Transducer

Maps a gist clause to a list of ULF formulas capturing its pragmatic meaning.

Parameters:

gist (str) – The gist clause to parse to a logical form.

Returns:

A list of ULF S-expressions.

Return type:

list[s-expr]

class ReactionTransducer

Bases: Transducer

Maps an observed event to a formula for an action that an agent can take.

Parameters:

observation (Eventuality) – The observed event for the agent to react to.

Returns:

A list of action formulas that the agent can take in reaction to the observation.

Return type:

list[s-expr]

Notes

TODO: this should likely be expanded to support a list of observed events rather than a single event, since in general a reaction may be contingent on multiple combined observations.

class SubplanTransducer

Bases: Transducer

Maps a plan step event to a list of subplans, i.e., sequences of events that are substeps of the given step.

Parameters:

step (Eventuality) – The event corresponding to a plan step to expand.

Returns:

A list of subplans, where each subplan is a sequence of substep events.

Return type:

list[list[Eventuality]]

class ParaphraseTransducer

Bases: Transducer

Maps a gist clause, given a conversation history and relevant knowledge, to a paraphrased response.

Parameters:

• gist (str) – The gist clause to paraphrase.

• conversation_log (list[DialogueTurn]) – The conversation history to use as context for paraphrasing.

• facts_bg (list[Eventuality]) – A list of “background” knowledge to use for paraphrasing, i.e., knowledge that may impact the resulting paraphrase, but may not be directly included in the paraphrased response.

• facts_fg (list[Eventuality]) – A list of “foreground” knowledge to use for paraphrasing, i.e., knowledge that should be directly included in the paraphrased response.

Returns:

A list of paraphrased responses.

Return type:

list[str]

class ResponseTransducer

Bases: Transducer

Maps a conversation history and relevant knowledge to a response utterance.

Parameters:

• conversation_log (list[DialogueTurn]) – The conversation history to use as context for the response.

• facts_bg (list[Eventuality]) – A list of “background” knowledge to use for the response, i.e., knowledge that may impact the resulting response, but may not be directly included in the response.

• facts_fg (list[Eventuality]) – A list of “foreground” knowledge to use for the response, i.e., knowledge that should be directly included in the response.

Returns:

A list of responses.

Return type:

list[str]

class AnswerTransducer

Bases: Transducer

Maps a conversation history and relevant knowledge to an answer utterance.

Parameters:

• conversation_log (list[DialogueTurn]) – The conversation history to use as context for the answer.

• facts_bg (list[Eventuality]) – A list of “background” knowledge to use for the answer, i.e., knowledge that may impact the resulting answer, but may not be directly included in the answer.

• facts_fg (list[Eventuality]) – A list of “foreground” knowledge to use for the answer, i.e., knowledge that should be directly included in the answer.

Returns:

A list of answers.

Return type:

list[str]

class AskTransducer

Bases: Transducer

Maps a conversation history and relevant knowledge to a question utterance.

Parameters:

• conversation_log (list[DialogueTurn]) – The conversation history to use as context for the question.

• facts_bg (list[Eventuality]) – A list of “background” knowledge to use for the question, i.e., knowledge that may impact the resulting question, but may not be directly included in the question.

• facts_fg (list[Eventuality]) – A list of “foreground” knowledge to use for the question, i.e., knowledge that should be directly included in the question.

Returns:

A list of questions.

Return type:

list[str]

class AffectTransducer

Bases: Transducer

Maps the word content of an utterance to an appropriate affect, given a conversation history.

Parameters:

• words (str) – The words to match to an appropriate affect.

• conversation_log (list[DialogueTurn]) – The conversation history to use as context for affect detection.

Returns:

A list of affects.

Return type:

list[str]