MPAI-MMC V2.5 AIMs - Natural Language Understanding

Function
Ref. Model
I/O Data
SubAIMs
JSON MData
Profiles
Ref. Software
Conformance
Performance

1 Functions

The Natural Language Understanding (MMC‑NLU) AIM receives an input text that might have been generated by a keyboard or by an MMC‑ASR AIM and produces a refined text (if the input text was produced by an MMC‑ASR AIM) and the Meaning of the input text. The MMC‑NLU AIM may also receive the descriptors of an audio‑visual scene and the ID of an object:

Receives	Text Object directly input by the Entity.
	Recognised Text from an Automatic Speech Recognition AIM.
	The ID of an Instance.
	The Audio‑Visual Scene Descriptors containing the Instance ID.
Refines	Input Text if coming from an Automatic Speech Recognition AIM.
Extracts	Meaning (Text Descriptors) from Recognised Text or Entity’s Text Object.
Produces	Refined Text.
	Text Descriptors (Meaning).

2 Reference Model

Figure 1 depicts the Reference Model of the Natural Language Understanding (MMC‑NLU) AIM.

Natural Language Understanding MMC-NLU AIM

Figure 1 – The Natural Language Understanding (MMC‑NLU) AIM

3 I/O Data

Table 1 specifies the Input and Output Data of the Natural Language Understanding (MMC‑NLU) AIM.

Table 1 – I/O Data of the Natural Language Understanding (MMC‑NLU) AIM

Input	Description
Text Object	Input Text.
Recognised Text Object	Text from the Automatic Speech Recognition AIM.
Instance Identifier	The Identifier of the specific Audio or Visual Object belonging to a level in the taxonomy.
Audio-Visual Scene Geometry	The digital representation of the spatial arrangement of the Visual Objects of the Scene.
Visual Instance Identifier	The Identifier of the specific Visual Object belonging to a level in the taxonomy.
Output	Description
Meaning	Descriptors of the Refined Text.
Refined Text Object	The refined version of the Recognised Text from the NLU AIM.

4 SubAIMs

No SubAIMs.

5 JSON Metadata

https://schemas.mpai.community/MMC/V2.5/AIMs/NaturalLanguageUnderstanding.json

6 Profiles

The Profiles of the Natural Language Understanding (MMC‑NLU) AIM are specified.

7 Reference Software

Not part of this specification.

8 Conformance Testing

Table 2 provides the Conformance Testing Method for the MMC‑NLU AIM.

If a schema contains references to other schemas, conformance of data for the primary schema implies that any data referencing a secondary schema shall also validate against the relevant schema, if present, and conform with the Qualifier, if present.

Table 2 – Conformance Testing Method for the MMC‑NLU AIM

Input	Text Object	Shall validate against Text Object schema. Text Data shall conform with Text Qualifier.
	Recognised Text	Shall validate against Text Object schema. Text Data shall conform with Text Qualifier.
	Instance ID	Shall validate against Instance ID schema.
	Audio-Visual Scene Geometry	Shall validate against AV Scene Geometry schema.
Output	Refined Text	Shall validate against Text Object schema. Text Data shall conform with Text Qualifier.
	Meaning	Shall validate against Meaning schema.

Table 3 provides an example of MMC‑NLU AIM Conformance Testing.

Table 3 – An example of MMC‑NLU AIM Conformance Testing

Input Data	Data Format	Input Conformance Testing Data
Input Selector	Binary data	All Input Selectors shall conform with Selector.
Text Object	Unicode	All input Text files to be drawn from Text files.
Recognised Text	Unicode	All input Text files to be drawn from Text files.
Output Data	Data Format	Output Conformance Testing Criteria
Meaning	JSON	All JSON files shall validate against Meaning schema.
Refined Text	Unicode	All Text files produced shall conform with Text.

The four taggings: POS_tagging, NE_tagging, dependency_tagging, and SRL_tagging must be present in the output JSON file of Meaning. Any of the four tagging values may be null.

9 Performance Assessment