Object model (Grammar sub-package)

The core of Dragonfly is a language object model revolving around three object types: grammars, rules, and elements. This section describes that object model.


A grammar is a collection of rules. It manages the rules, loading and unloading them, activating and deactivating them, and it takes care of all communications with the speech recognition engine. When a recognition occurs, the associated grammar receives the recognition event and dispatches it to the appropriate rule.

Normally a grammar is associated with a particular context or functionality. Normally the rules within a grammar are somehow related to each other. However, neither of these is strictly necessary, they are just common use patterns.

The Grammar class and derived classes are described in the Grammar classes section.


Rules represent voice commands or parts of voice commands. Each rule has a single root element, the basis of a tree structure of elements defining how the rule is built up out of speakable parts. The element tree determines what a user must say to cause this rule to be recognized.

The Rule class and derived classes are described in the Rules section.

Exported rules

Rules can be exported or not exported. Whether a rule is exported or not is defined when the rule is created.

Only exported rules can be spoken directly by the user. In other words, they form the entry points into a grammar, causing things to happen (callbacks to be called) when appropriate words are recognized.

WSR distinguishes between top-level rules, which can be recognized directly, and exported rules, which can be referenced from rules in other grammars. NatLink doesn’t allow inter-grammar rule referencing and uses exported to refer to directly recognizable rules. Dragonfly follows NatLink in functionality and terminology on this topic.

Properties of exported rules:

  • Exported rules are known as a top-level rules for WSR (SRATopLevel).

  • Exported rules can be spoken by the user directly.

  • Exported rules can be referenced from other rules within the same grammar.

  • Exported rules can be referenced from rules in other grammars (only possible for WSR).

  • Exported rules can be enabled and disabled to receive recognitions or not (enable(), disable()).

  • Exported rules have callbacks which are called when recognition occurs (process_begin(), process_recognition()).

Non-exported rules cannot be recognized directly but only as parts of other rules that reference them.

Properties of non-exported rules:

  • Non-exported rules can’t be spoken by the user directly.

  • Non-exported rules can be referenced from other rules within the same grammar.

  • Non-exported rules can’t be referenced from rules in other grammars (never possible for DNS).

Imported rules

Rules can be imported, i.e. defined outside the grammar referencing them, or not imported, i.e. defined within the grammar. Whether a rule is imported or not is defined when the rule is created.

NatLink in general doesn’t allow rules from one grammar to be imported into another grammar, i.e. inter-grammar rule referencing. However, it does provide the following three built-in rules which can be imported:

  • dgnletters – All the letters of the alphabet for spelling

  • dgnwords – All words active during dictation

  • dgndictation – A special rule which corresponds to free-form dictation; imported by Dragonfly for its Dictation element


Lists are dynamic language elements which can be updated and modified without reloading a grammar. There are two list types:

  • List – sub-class of Python’s built-in list type

  • DictList – sub-class of Python’s built-in dict type

Lists of either type can be instantiated and used like normal lists / dictionaries. They are meant to be used in Dragonfly rules via the special ListRef and DictListRef elements. The engine is automatically notified when lists are modified.

The ListBase class and derived classes are described in the Lists section. See the Doctests for the List class for usage examples.


Elements are the basic building blocks of the language model. They define exactly what can be said and thereby form the content of rules. The most common elements are:

  • Literal – one or more literal words

  • Sequence – a series of other elements

  • Alternative – a choice of other elements, only one of which can be said within a single recognition

  • Optional – an element container which makes its single child element optional

  • Repetition – an element class representing a repetition of one child element

  • RuleRef – a reference to another rule

  • RuleWrap – an element class used to wrap a Dragonfly element into a new private rule to be referenced by the same element or other RuleRef elements

  • ListRef – a reference to a Dragonfly List

  • DictListRef – a reference to a Dragonfly DictList

  • Dictation – a free-form dictation element which allows the speaker to say one or more natural language words

  • Modifier – a special element that allows direct modification of the output of another element at recognition time

  • Impossible – a special element that cannot be recognized

  • Empty – a special element that is always recognized, similar to children of Optional elements

See the Doctests for the fundamental element classes for element usage examples.

The above mentioned element types are at the heart of Dragonfly’s object model. But of course using them all the time to specify every grammar would be quite tedious. There are therefore also special elements which construct these basic element types from string specifications:

  • Compound – a special element which parses a string spec to create a hierarchy of basic elements.

  • Choice – a special element taking a choice dictionary argument, interpreting keys as Compound string specifications and values for what to return when compound specs are successfully decoded during the recognition process.

    The choice argument may also be a list or tuple of strings, in which case the strings are also interpreted as Compound string specifications. However, the values returned when compound specs are successfully decoded during the recognition process are the recognized words. Note: these values will be matching part(s) of the compound specs.

See the Doctests for the Compound element class for Compound usage examples and see the Element classes section for class references and further documentation on each element class.