Uh oh! We couldn’t find any match.
Please try other search keywords.
This glossary covers the main technical concepts discussed in the Developer Guide.
An abstract definition of some operation that may be performed. It
can be a search (such as
FindRestaurants) or perform an operation
BookHotel), and contains action
inputs and one action
A description of an input of an action; in particular, it includes a
concept, and a
A description of the output of an action; in particular, it includes a
concept and a
A description of how an action can be used, including whether it has side effects and whether its effects on the world are atomic or continuing.
The output of a
A format for NL training that includes the utterance as well as embedded annotations (or labels). It is used internally by the platform to represent training provided by developers.
A user can provide authorization for a function to access an external provider via a
When extending a
structure concept, a particular property can be bound to a specific value (e.g.
Restaurant can extend a
Business and bind the business category property to be "Restaurant").
A dot-delimited set of property names indicating a property or any depth of sub-property of a
structure concept (e.g.
country.name on a
review.rating.score on a
A Concept Action Network (CAN), consisting of or relating to the graph of CAN nodes (including concepts and actions), as well as functions and a variety of other systems necessary for interpretation, planning, and execution. Primarily used to describe other ideas, such as CAN development, CAN nodes, etc.
A node in the graph which may represent concepts or actions, as well as other operations such as lookups.
A unit of contribution in the Marketplace, consisting of a group of related functionality. It also delineates permissions among developers.
In many cases in modeling (such as describing action inputs), it is necessary to describe how many values may appear. Cardinality is described in terms of "min cardinality" (i.e. optional or required) and "max cardinality" (i.e. one or many).
A property of a structure concept that should never be projected or used in isolation by the planner.
Modifications or additions by developers to code or data that affect operation of the system, including models, functions, dialog, layout, and NL vocabulary and training.
A prompt that asks the user to confirm something, such as making a purchase.
A user interface in which the user exchanges queries and responses with the platform in a dialog, until an answer is reached.
An abstract definition of a class of possible values. Anything that a user may refer to or that a developer wishes to model can be a concept. Concepts include primitive concepts and structure concepts.
Concept Action Network
The assembly of all concepts and actions into a graph, as well as all the associated components that support interpretation, planning, execution, dialog, and supporting features for users.
A data value that is an instance of a concept.
A relationship between two concepts, where the child concept is a more specific version of the parent concept, but is not an extension of it. Represents a bi-directional "is-a" relationship (e.g. a destination airport is an airport, and vice versa).
Certain essential capsules that are "baked into" the platform, including support for geographic models and data, date and time, payments, and credentials.
A feed of data downloaded from a provider and used to provide vocabulary or data for indexed concepts (e.g. geographic data or business listings).
A point in time, consisting of a date and a time. Dates and times are also used in isolation.
Developer-provided customizations of a user interaction with the platform that help the user in surprising or unexpected ways that differ from normal execution dialog and prompts.
A person or organization that can make modifications to capsules.
A description of the responses the platform can make in response to a user, or while fulfilling a request.
A dialog response triggered before, during, or after execution. Examples include progress notification, action confirmation, prompts for more information, error messages, and result messages.
A composable template for dialog, relevant to a particular mode. The modes are
An operation that changes the plan in response to conditions encountered during execution. Effects include replacing or dropping an input, prompting a user for new input, or halting plan execution.
A structure concept that has a fixed set of values. (For example, the
viv.measurement.LengthUnit concept might have values
A description of the conditions for when two concept values may or may not be the same. Equivalence can return "true," "false," or "uncertain".
Expression Language (EL)
A simple language for expressions used in a few places such as dialogs, strategies and layouts.
The process of running a plan to produce a result, and the associated presentation to the user.
A relationship between two concepts, where a child concept may extend a parent concept by adding additional properties. Represents an single-directional is-a relationship (e.g. a restaurant is a business).
An optional expression indicating to the platform preconditions for a function's execution.
See goal signal.
A signal in a plan intent indicating the result type of the plan.
A strategy that suggests default values for a concept for an instantiation prompt.
A prompt that asks a user to provide or confirm a value of a concept.
The process of converting an utterance to an intent.
A formal expression of what a user's expressed intent, in terms of CAN nodes. It is made of signals.
A description of the way content or forms are presented to the user.
A pattern that matches against a subset of the plan graph. It is expressed as a labeled sub-graph of the plan graph, written as a tree where the root is the current point of execution.
The Marketplace is the multi-device, multi-locale location, where all your capsule submissions can be published to all the Bixby Targets defined in your
A set of abstract definitions of concepts and actions.
The definition of concepts and actions to enable the platform to solve a use case. Good modeling takes into account the planning and execution process.
See CAN node.
Examples contributed by developers to that teach the platform's machine learning system how to map utterances to intents. Internally, NL training is represented as aligned text, though more commonly it is edited via the training tool.
Natural Language (NL)
Anything a human might type or say when interacting with the system. It need not be grammatical, and may be of any language.
A unique identifier for a CAN node. The node ID of a concept or action is typically of the form
restaurant.FindRestaurants (where, in the first example,
air is a module name, and
Flight is a name for the concept, and similarly for the action). Other nodes may have internal or generated IDs that reflect their function in a graph.
A goal signal that labels the entire utterance, and is not attached to particular word.
When extending a concept, it is possible to override a property with modifications. For example a
Restaurant concept may override a
name property to be a
An entity that has ownership of concept values, including both capsules and users.
The core software of Bixby, excluding contributions from developers, including systems that support interpretation, planning, and execution, as well as core models.
A set of steps (technically, a graph) that may be executed to yield a goal.
The component of the platform that produces a plan based on an intent.
The process of producing plans from intents.
The graph that describes a plan, consisting of nodes corresponding to nodes and edges indicating relationships, such as inputs, outputs, projections, and other operations.
The process of displaying results to the user, or accepting or eliciting inputs.
One of the primitive data types, such as numbers or string names.
The permission to access to a specific piece of sensitive data, such as a credit card number or password.
An interruption in execution to confirm with the user, or to elicit a value or selection.
A field within a structure concept. Represents a "has-a" relationship between the structure concept and the property's concept.
An entity or organization that provides data or services to the platform. For example one provider might be a weather data feed; another might be a restaurant booking API.
A type of intermediary goal for which a prompt is prompting. When Bixby fails to reach the original goal and has to prompt the user, that response is used to reach the recovery goal, which in turn helps Bixby complete the original goal.
A concept value that is the output of a function.
A signal in a plan intent indicating a concept or action that is a way-point in the plan.
A way to offer advice about which values of a selection prompt are good and which ones are not.
A prompt that asks the user to make a selection.
A service used by function implementations, such as a REST or SOAP web service.
When an action changes things in the world, such as modifying things remote service, or performing an action in the "real world," we say it has side effects.
A part of an intent representing a node and the role it is expected to play in the desired plan (goal, route, reference, subplan, etc.).
A suggested approach for handling execution flow in a particular situation, such as an instantiation strategy or selection strategy.
A concept that, like a record, holds multiple properties, each itself corresponding to another concept.
A signal that indicates a recursive subplan.
Any string value, when used to define a primitive enum concept.
A device and locale collection of capsules. For example,
bixby-mobile-ko-KR is the Target for all capsules for mobile devices in Korean.
Referring to the sum total of all NL contributions for a capsule. For example, you could say “The training for the Uber capsule is in great shape”. But you shouldn't say “The trainings for the Uber capsule are in great shape”. See also NL Training.
An individual unit of training for a capsule. A set of those units would then be referred to as training examples and should never be called "trainings." You don't create "a training" to add to a capsule's "trainings"; rather, you contribute "training examples." Alternatively, you could call them training entries.
An end-user of the entire system.
A natural language request made actually or potentially made by user.
See concept value.
A signal in a plan intent indicating a concept value.
Words or phrases contributed to a particular node, to allow NL intent interpretation to recognize when that node is referred to.