NAV Navbar
Python

NLU as a Service gRPC API

Nuance NLU extracts meaning from what your users say or write NLUaaS

The Nuance NLU (Natural Language Understanding) service turns text into meaning, extracting the underlying meaning of what your users say or write, in a form that an application can understand. NLU as a Service provides a semantic interpretation of the user’s input.

NLU works with a data pack for your language and locale, and a semantic language model customized for your application. It can also use interpretation aids such as additional language models, dictionaries, and wordsets to improve understanding in specific environments or businesses.

The gRPC protocol provided by NLU allows a client application to request interpretation services in all the programming languages supported by gRPC.

gRPC is an open source RPC (remote procedure call) software used to create services. It uses HTTP/2 for transport, and protocol buffers to define the structure of the application. NLU supports Protocol Buffers version 3, also know as proto3.

Version: v1

This release supports the v1 version of the gRPC protocol.

Upgrading to v1

To upgrade to the v1 protocol from v1beta1, you need to regenerate your programming-language stub files from the new proto files, then make small adjustments to your client application.

First regenerate your client stubs from the new proto files, as described in gRPC setup.

  1. Download zip files containing the gRPC proto files for the Runtime service and the Wordset service. We recommend you make a new directory for the v1 files.
  2. Use gRPC tools to generate the client stubs from the proto file.
  3. Notice the new client stub files.

Then adjust your client application for the changes made to the protocol in v1. See gRPC API for details of each item.

NLU essentials

Natural language understanding (NLU) is one of the components of a rich conversational voice experience for your end users. NLUaaS uses engines hosted by Nuance and is accessible from a single gRPC interface.

How it works

NLUaaS accepts input from the user. This input can be text written by the user or the result of user speech transcribed into text by automatic speech recognition (ASR).

NLU applies transformation rules to the text of user input and performs formatting of output for display or further processing.

It derives domain-specific meaning from text using speech technology based on artificial intelligence (AI) and machine learning.

It interprets the text of the input with the aid of a Nuance data pack and a semantic model created in Mix.nlu, returning a semantic interpretation.

Your client application can use this result to drive the next human-machine turn.

Intents and entities

NLUaaS interpretation results consist of one or more hypotheses of the meaning of the user input. Each hypothesis contains intents and entities, along with NLU's confidence in the hypothesis.

An intent is the overall meaning of the user input in a form an application can understand, for example PAY_BILL, PLACE_ORDER, BOOK_FLIGHT, or GET_INFO. See Interpretation results: Intents for some examples.

Entities (also known as concepts) define the meaning of individual words within the input. They represent categories of things that are important to the intent. For example, the PLACE_ORDER intent might have entities such as PRODUCT, FLAVOR, and QTY. Each entity contains a set of values, so the FLAVOR entity could have values such as Chocolate, Strawberry, Blueberry, Vanilla, and so on.

At runtime, NLU interprets the sentence I’d like to order a dozen blueberry pies as:

List entities have specific values, while other types of entity have values defined in a grammar file and/or regular expression. See Interpretation results: Entities for examples.

Extending the model

For more flexibility, you can extend your semantic model with wordsets containing additional terms for dynamic list entities. For entities with small numbers of terms (less than 100), an inline wordset can be included with the interpretation request at runtime. The wordset gets compiled at runtime and used as a resource to improve interpretation. This is convenient and simple, but for large wordsets (hundreds of items and above), this can lead to issues with latency.

As an alternative for larger wordsets, wordset files can be uploaded to Mix ahead of time to be compiled using a separate Wordset API. The compiled wordsets are saved in Mix and can later be accessed at runtime as an external reference. For large wordsets, this can significantly reduce latency in interpretation requests. If in doubt about which approach to take, test the latency with inline wordsets.

Prerequisites from Mix

Before developing your gRPC application, you need a Mix project that provides an NLU model as well as authorization credentials.

  1. Create a Mix project and model: see Mix.nlu workflow to:

    • Create a Mix project.

    • Create, train, and build a model in the project.

    • Create and deploy an application configuration for the project.

  2. Learn how to reference the semantic model in your application. You may only reference models created in your Mix project. See Mix.dashboard URN.

  3. Generate a "secret" and client ID of your Mix project: see Mix.dashboard Authorize your client application. Later you will use these credentials to request an access token to run your application.

  4. Learn the URL to call the NLU service: see Mix.dashboard Accessing a runtime service.

gRPC setup

Download and unzip Runtime proto files 

├── nuance
│   └── nlu
│       └── v1
│           ├── runtime.proto
│           ├── result.proto
│           ├── interpretation-common.proto
│           ├── single-intent-interpretation.proto
│           └── multi-intent-interpretation.proto
│
└── nuance_runtime_proto_files.zip

Download and unzip Wordset proto files 

├── nuance
│   ├── nlu
│   │   ├── wordset
│   │   │ └── v1beta1
│   │   │     └── wordset.proto
│   │   └── common
│   │       └── v1beta1 
│   │           ├── job.proto
│   │           └── resource.proto
│   └── rpc
│       ├── error_details.proto
│       ├── status_code.proto
│       └── status.proto
└── nuance_wordset_proto_files.zip

Install gRPC for Python

$ python3 -m venv env
$ source env/bin/activate
$ pip install grpcio
$ pip install grpcio-tools
$ mkdir -p google/api
$ curl https://raw.githubusercontent.com/googleapis/googleapis/master/google/api/annotations.proto \
    google/api/annotations.proto
$ curl https://raw.githubusercontent.com/googleapis/googleapis/master/google/api/http.proto \
    google/api/http.proto

Generate client stubs from proto files

$ python -m grpc_tools.protoc --proto_path=./ --python_out=./ \
    --grpc_python_out=./ nuance/nlu/v1/runtime.proto
$ python -m grpc_tools.protoc --proto_path=./ --python_out=./ \
    nuance/nlu/v1/result.proto
$ python -m grpc_tools.protoc --proto_path=./ --python_out=./ \
    nuance/nlu/v1/interpretation-common.proto
$ python -m grpc_tools.protoc --proto_path=./ --python_out=./ \
    nuance/nlu/v1/single-intent-interpretation.proto
$ python -m grpc_tools.protoc --proto_path=./ --python_out=./ \
    nuance/nlu/v1/multi-intent-interpretation.proto
$ python -m grpc_tools.protoc --proto_path=./ --python_out=./ \
    --grpc_python_out=./ nuance/nlu/wordset/v1beta1/wordset.proto
$ python -m grpc_tools.protoc --proto_path=./ --python_out=./ \
    nuance/nlu/common/v1beta1/job.proto
$ python -m grpc_tools.protoc --proto_path=./ --python_out=./ \
    nuance/nlu/common/v1beta1/resource.proto
$ python -m grpc_tools.protoc --proto_path=./ --python_out=./ \
    nuance/rpc/status.proto
$ python -m grpc_tools.protoc --proto_path=./ --python_out=./ \
    nuance/rpc/status_code.proto
$ python -m grpc_tools.protoc --proto_path=./ --python_out=./ \
    nuance/rpc/error_details.proto

Final structure of stub files

├── nuance
    ├── nlu
    │   ├── v1
    │   │   ├── runtime_pb2.py
    │   │   ├── runtime_pb2_grpc.py
    │   │   ├── result_pb2.py
    │   │   ├── interpretation-common_pb2.py
    │   │   ├── single-intent-interpretation_pb2.py
    │   │   └── multi-intent-interpretation_pb2.py
    │   ├── wordset
    │ │ └── v1beta1
    │ │     ├── wordset_pb2.py
    │ │     └── wordset_grpc_pb2.py
    │   └── common
    │       └── v1beta1 
    │         ├── job_pb2.py
    │         └── resource_pb2.py
    └── rpc
        ├── error_details_pb2.py
        ├── status_code_pb2.py
        └── status_pb2.py

The basic steps in using the NLUaaS gRPC API are:

  1. Install gRPC for the programming language of your choice, including C++, Java, Python, Go, Ruby, C#, Node.js, and others. See gRPC Documentation for a complete list and instructions on using gRPC with each language.

  2. Download the zip files containing the gRPC proto files for the Runtime service and the Wordset service and related messages. Together, these zip files contain several sets of proto files:

    • nuance/nlu/v1/ contains protos related to runtime NLU interpretation: runtime.proto, result.proto, interpretation-common.proto, single-intent-interpretation.proto, and multi-intent-interpretation.proto.
    • nuance/nlu/wordset/v1beta1/ contains wordset.proto with RPCs and messages related to working with compiled wordsets.
    • nuance/nlu/common/v1beta1/ contains generic classes for referencing external resources and for job status updates: job.proto and resource.proto
    • nuance/rpc/ contains generic classes for status and error codes: status.proto, status_code.proto, and error_details.proto.

  3. Unzip the files in a location that your applications can access, for example in a directory containing your client apps. If you have proto files and applications from an earlier version of NLUaaS, we recommend you create a new directory to hold the new files. This will make it easier to identify the Runtime and Wordset files in their new path structure.

  4. Generate client stub files in your programming language from the proto files using gRPC protoc. Depending on your programming language, the stubs may consist of one file or multiple files per proto file. These stub files contain the methods and fields from the proto files as implemented in your programming language. You will consult the stubs with the proto files.

  5. Write your client application, referencing the functions or classes in the client stub files. See Client runtime app development for details.

  6. Run your client app to request text interpretation, optionally passing in NLU models and wordsets.

Runtime app development

The gRPC protocol for NLU lets you create a client application for requesting and receiving semantic interpretation from input text. This section describes how to implement the basic functionality of NLU in the context of a Python application. For the complete application, see Sample Python runtime client.

The essential tasks are shown in the following high-level sequence flow:

Sequence flow

Step 1: Generate token

The run-nlu-client.sh script requests the token then runs the application

#!/bin/bash

CLIENT_ID=“appID%3ANMDPTRIAL_your_name_company_com_20201102T144327123022%3Ageo%3Aus%3AclientName%3Adefault”
SECRET="5JEAu0YSAjV97oV3BWy2PRofy6V8FGmywiUbc0UfkGE"
export TOKEN="`curl -s -u "$CLIENT_ID:$SECRET" "https://auth.crt.nuance.com/oauth2/token" \
-d "grant_type=client_credentials" -d "scope=tts nlu asr" \
| python -c 'import sys, json; print(json.load(sys.stdin)["access_token"])'`"

./nlu_client.py --serverUrl nlu.api.nuance.com:443 --secure --token $TOKEN 
--modelUrn "urn:nuance-mix:tag:model/bakery/mix.nlu?=language=eng-USA" 
--textInput "$1" --wordsetUrn "urn:nuance-mix:tag:wordset:lang/bakery/item_names/eng-USA/mix.nlu"

Nuance Mix uses the OAuth 2.0 protocol for authorization. Your client application must provide an access token to be able to access the NLU runtime service. The token expires after a short period of time so must be regenerated frequently.

Your client application uses the client ID and secret from the Mix Dashboard (see Prerequisites from Mix) to generate an access token from the Nuance authorization server.

The client ID starts with appID: followed by a unique identifier. If you are using the curl command, replace the colon with %3A so the value can be parsed correctly:

appID:NMDPTRIAL_your_name_company_com_2020...  
-->     
appID%3ANMDPTRIAL_your_name_company_com_2020...

You can generate the token in one of several ways, either as part of the client application or as a script file. This Python sample uses a Linux script to generate a token and store it in an environment variable. The token is then passed as an argument to the application, where it is used to create a secure connection to the NLU service.

Note that the same token should be reused until it expires rather than requested for each interpretation. Generating a new token every time adds to latency. Also, token request has more strict rate limits in place.

Step 2: Import functions

Import functions from stubs

from nuance.nlu.v1.runtime_pb2 import *
from nuance.nlu.v1.runtime_pb2_grpc import *
from nuance.nlu.v1.result_pb2 import *

In your client application, import all functions from the client stubs that you generated in gRPC setup.

Do not edit these stub files.

Step 3: Authorize and connect

The Python app uses the token as it creates the secure connection to the NLU service

def create_channel(args):
    channel = None
    call_credentials = None

    if args.token:
        log.debug("Adding CallCredentials with token %s" % args.token)
        call_credentials = grpc.access_token_call_credentials(args.token)

    log.debug("Creating secure gRPC channel")
    channel_credentials = grpc.ssl_channel_credentials()
    channel_credentials = grpc.composite_channel_credentials(channel_credentials, call_credentials)
    channel = grpc.secure_channel(args.serverUrl, credentials=channel_credentials)

    return channel

You create a secure gRPC channel and authorize your application to the NLU service by providing the URL of the hosted NLU service and an access token.

In the Python example, the URL of the NLU service is passed to the application as a command line argument. The URL is passed as the first argument: nlu.api.nuance.co.uk:443

Step 4: Configure interpret request

Configure interpret request

# Single intent, plain text logging
params = InterpretationParameters(
 Interpretation_result_type=EnumInterpretationResultType.SINGLE_INTENT, interpretation_input_logging_mode=EnumInterpretationInputLoggingMode.PLAINTEXT)
# Reference the model 
model = ResourceReference(
    type=EnumResourceType.SEMANTIC_MODEL,
    uri=args.modelUrn)
# Reference compiled wordset
wordset_reference = ResourceReference(
        type=EnumResourceType.COMPILED_WORDSET,
        uri = args.wordsetUrn,
        request_timeout_ms = 5000)

resource = InterpretationResource(external_reference = wordset_reference)
resources = [resource]
# Describe the text to perform interpretation on
input = InterpretationInput(
    text=args.textInput)

# Build the request
interpret_req = InterpretRequest(
        parameters=params,
        model=model,
        resources = resources,
        input=input)

An interpretation request includes InterpretationParameters that define the type of interpretation you want. Consult your generated stubs for the precise parameter names. Some parameters are:

It also includes resources referenced to help with the interpretation. This includes:

The input to interpret must also be provided, either plain text or the result from a call to ASRaaS.

For more details, see InterpretRequest.

Step 5: Call Runtime client stub

Call Runtime client stub

with create_channel(args) as channel:
    stub = RuntimeStub(channel)
    response = stub.Interpret(construct_interpret_request(args))
    print(MessageToJson(response))
print("Done")

With a communication channel established, the app calls a client stub function or class, which makes use of the channel. This stub is based on the main service name and is defined in the generated client files. In Python it is named RuntimeStub.

Step 6: Request interpretation

Request interpretation 

def main():
    args = parse_args()
    log_level = logging.DEBUG
    logging.basicConfig(
        format='%(lineno)d %(asctime)s %(levelname)-5s: %(message)s', level=log_level)
    with create_channel(args) as channel:
        stub = RuntimeStub(channel)
        response = stub.Interpret(construct_interpret_request(args))
        print(MessageToJson(response))
    print("Done")

After configuring the InterpretRequest and calling the client stub, you can send the InterpretRequest.

Step 7: Process results

Receive results

def process_result(response):
    print(MessageToJson(response))

Finally the app returns the results received from the NLU engine. These applications format the interpretation result as a JSON object, similar to the Try panel in Mix.nlu.

For details about the structure of the result, see InterpretResult.

Wordset app development

The previous section gives high-level guidance on how to develop a basic NLU client application that can interpret intents and entities from text inputs. Accuracy on interpretation of values for some types of entities can often be greatly improved by providing wordsets that define values to expect.

With the gRPC Wordset API for NLU, you can create a client application to perform requested wordset operations:

This section describes how to implement the basic wordsets functionality in the context of a Python application. For the complete application, see Sample Python wordsets client.

Step 1: Generate token

This is similar to what is done for the client runtime app. Note that when generating the token, you must specify the nlu.wordset scope.

export TOKEN="... -d "scope=tts nlu asr nlu.wordset" ..."

Step 2: Import functions

Import functions from stubs

from nuance.nlu.wordset.v1beta1.wordset_pb2 import *
from nuance.nlu.wordset.v1beta1.wordset_pb2_grpc import *
from nuance.nlu.common.v1beta1.resource_pb2 import *

In your client application, import all functions from the wordset client stubs that you generated in gRPC setup.

Do not edit these stub files.

Step 3: Authorize and connect

This is the same as for the client runtime app.

Step 4: Configure wordset request

Configure compile wordset request

def create_compile_wordset_request(args):
  target_artifact_reference = ResourceReference(uri=args.artifactUrn)
  companion_artifact_reference = ResourceReference(uri=args.modelUrn)
  wordset = json.dumps(json.load(args.wordsetFile))
  return CompileWordsetRequest(
    target_artifact_reference=target_artifact_reference,
    companion_artifact_reference=companion_artifact_reference,
    wordset=wordset,
    metadata=list_to_dict(args.metadata),
    client_data=list_to_dict(args.clientData))

Configure get metadata request

def create_get_wordset_metadata_request(artifactUrn):
    artifact_reference = ResourceReference(uri=artifactUrn)
    return GetWordsetMetadataRequest(artifact_reference=artifact_reference)

Configure delete wordset request

def create_delete_wordset_request(artifactUrn):
    artifact_reference = ResourceReference(uri=artifactUrn)
    return DeleteWordsetRequest(artifact_reference=artifact_reference)

With authorization complete, you can configure your request to:

Step 5: Call Wordset client stub

Call Wordset client stub

with create_channel(args) as channel:
    stub = WordsetStub(channel)

With a communication channel established, the app calls a client stub function or class, which makes use of the channel. This stub is based on the main service name and is defined in the generated client files. In Python it is named WordsetStub.

Step 6: Perform wordset operation

Compile a wordset 

def compile_wordset(args):
  with create_channel(args) as channel:
    stub = WordsetStub(channel)
    compiled_wordset_request = create_compile_wordset_request(args)
    for message in stub.CompileWordsetAndWatch(compiled_wordset_request):
      print(MessageToJson(message))

Retrieve wordset metadata

def get_wordset_metadata(args):
  with create_channel(args) as channel:
    stub = WordsetStub(channel)
    response = stub.GetWordsetMetadata(create_get_wordset_metadata_request(args.artifactUrn))
    print(MessageToJson(response))

Delete a wordset 

def delete_wordset(args):
  with create_channel(args) as channel:
    stub = WordsetStub(channel)
    response = stub.DeleteWordset(create_delete_wordset_request(args.artifactUrn))
    print(MessageToJson(response))

After configuring the wordset request and calling the client stub, you can access the appropriate Wordset API method through the client stub, sending the request.

Sample Python runtime client

Runtime client overview

This basic Python app, nlu_client.py, requests and receives an NLU interpretation. 

import argparse
import sys
import logging
import os
import grpc
import wave
from time import sleep

from google.protobuf.json_format import MessageToJson

from nuance.nlu.v1.runtime_pb2 import *
from nuance.nlu.v1.runtime_pb2_grpc import *
from nuance.nlu.v1.result_pb2 import *

log = logging.getLogger(__name__)

def parse_args():
    parser = argparse.ArgumentParser(
        prog="nlu_client.py",
        usage="%(prog)s [-options]",
        add_help=False,
        formatter_class=lambda prog: argparse.HelpFormatter(
            prog, max_help_position=45, width=100)
    )

    options = parser.add_argument_group("options")
    options.add_argument("-h", "--help", action="help",
                         help="Show this help message and exit")
    options.add_argument("--nmaid", nargs="?", help=argparse.SUPPRESS)
    options.add_argument("--token", nargs="?", help=argparse.SUPPRESS)
    options.add_argument("-s", "--serverUrl", metavar="url", nargs="?",
                         help="NLU server URL, default=localhost:8080", default='localhost:8080')
    options.add_argument('--modelUrn', nargs="?", 
                         help="NLU Model URN")
    options.add_argument('--wordsetUrn', nargs="?", 
                         help="compiled wordset URN")
    options.add_argument("--textInput", metavar="file", nargs="?",
                         help="Text to perform interpretation on")
    return parser.parse_args()

def create_channel(args):
    channel = None
    call_credentials = None

    if args.token:
        log.debug("Adding CallCredentials with token %s" % args.token)
        call_credentials = grpc.access_token_call_credentials(args.token)

    log.debug("Creating secure gRPC channel")
    channel_credentials = grpc.ssl_channel_credentials()
    channel_credentials = grpc.composite_channel_credentials(channel_credentials, call_credentials)
    channel = grpc.secure_channel(args.serverUrl, credentials=channel_credentials)

    return channel


def construct_interpret_request(args):
    # Single intent, plain text logging
    params = InterpretationParameters(interpretation_result_type=EnumInterpretationResultType.SINGLE_INTENT, interpretation_input_logging_mode=EnumInterpretationInputLoggingMode.PLAINTEXT)
    # Reference the model via the app config
    model = ResourceReference(type=EnumResourceType.SEMANTIC_MODEL, uri=args.modelUrn)
    # Reference compiled wordset
    wordset_reference = ResourceReference(type=EnumResourceType.COMPILED_WORDSET, uri = args.wordsetUrn)
    resource = InterpretationResource(external_reference = wordset_reference)
    resources = [resource]
    # Describe the text to perform interpretation on
    input = InterpretationInput(text=args.textInput)
    # Build the request
    interpret_req = InterpretRequest(parameters=params, model=model, resources = resources, input=input)
    return interpret_req

def main():
    args = parse_args()
    log_level = logging.DEBUG
    logging.basicConfig(
        format='%(lineno)d %(asctime)s %(levelname)-5s: %(message)s', level=log_level)
    with create_channel(args) as channel:
        stub = RuntimeStub(channel)
        response =       stub.Interpret(construct_interpret_request(args))
        print(MessageToJson(response))
    print("Done")

if __name__ == '__main__':
  main()

This section contains a fully-functional sample Python client application consisting of the following files:

This Python client requests and receives an NLU interpretation, and also performs these tasks:

Runtime client requirements

To run this sample app, you need:

Runtime client procedure

To run this simple application:

Step 1. Download the sample runtime app here and unzip it in a working directory (for example, /home/userA/nlu-runtime-sample-python-app).

Step 2. Download the gRPC .proto files here and unzip the files in the sample app working directory.

Step 3. Navigate to the sample app working directory and install the required dependencies:

$ python3 -m venv env
$ source env/bin/activate    
$ pip install --upgrade pip
$ pip install grpcio
$ pip install grpcio-tools
$ mkdir -p google/api
$ curl https://raw.githubusercontent.com/googleapis/googleapis/master/google/api/annotations.proto \
google/api/annotations.proto
$ curl https://raw.githubusercontent.com/googleapis/googleapis/master/google/api/http.proto \
google/api/http.proto

Step 4. Generate the client stubs:

$ python -m grpc_tools.protoc --proto_path=./ --python_out=./ \
--grpc_python_out=./ nuance/nlu/v1/runtime.proto
$ python -m grpc_tools.protoc --proto_path=./ --python_out=./ \
nuance/nlu/v1/result.proto
$ python -m grpc_tools.protoc --proto_path=./ --python_out=./ \
nuance/nlu/v1/interpretation-common.proto
$ python -m grpc_tools.protoc --proto_path=./ --python_out=./ \
nuance/nlu/v1/single-intent-interpretation.proto
$ python -m grpc_tools.protoc --proto_path=./ --python_out=./ \
nuance/nlu/v1/multi-intent-interpretation.proto
$ python -m grpc_tools.protoc --proto_path=./ --python_out=./ \
--grpc_python_out=./ nuance/nlu/wordset/v1beta1/wordset.proto
$ python -m grpc_tools.protoc --proto_path=./ --python_out=./ \
nuance/nlu/common/v1beta1/job.proto
$ python -m grpc_tools.protoc --proto_path=./ --python_out=./ \
nuance/nlu/common/v1beta1/resource.proto
$ python -m grpc_tools.protoc --proto_path=./ --python_out=./ \
nuance/rpc/status.proto
$ python -m grpc_tools.protoc --proto_path=./ --python_out=./ \
nuance/rpc/status_code.proto
$ python -m grpc_tools.protoc --proto_path=./ --python_out=./ \
nuance/rpc/error_details.proto

Step 5. Edit the script run-nlu-client.sh, to add your CLIENT_ID and SECRET. These are your Mix credentials as described in Generate token.

#!/bin/bash

# Remember to change the colon (:) in your CLIENT_ID to code %3A
CLIENT_ID="appID%3A...ENTER MIX CLIENT_ID..."
SECRET="...ENTER MIX SECRET..."
export TOKEN="`curl -s -u "$CLIENT_ID:$SECRET" "https://auth.crt.nuance.co.uk/oauth2/token" \
-d "grant_type=client_credentials" -d "scope=tts nlu asr" \
| python -c 'import sys, json; print(json.load(sys.stdin)["access_token"])'`"

./nlu_client.py --serverUrl nlu.api.nuance.co.uk:443 --token $TOKEN
--modelUrn "$1"  \
--textInput "$2"  \
--wordsetUrn "$3"

Step 6. Run the application using the script file, passing it the URN and a text to interpret:

./run-nlu-client.sh modelUrn textInput wordsetUrn

Where:

For example:

$ ./run-nlu-client.sh "urn:nuance-mix:tag:model/bakery/mix.nlu?=language=eng-USA" "I'd like to order a strawberry latte" "mix:tag:wordset:lang/bakery/item_names/eng-USA/mix.nlu"

The NLU engine returns the results:

$ ./run-nlu-client.sh "I'd like to order a strawberry latte"
{
  "status": {
    "code": 200,
    "message": "OK"
  },
  "result": {
    "literal": "I'd like to order a strawberry latte",
    "interpretations": [
      {
        "single_intent_interpretation": {
          "intent": "PLACE_ORDER",
          "confidence": 0.9961925148963928,
          "origin": "STATISTICAL",
          "entities": {
            "PRODUCT": {
              "entities": [
                {
                  "text_range": {
                    "start_index": 31,
                    "end_index": 36
                  },
                  "confidence": 0.9177429676055908,
                  "origin": "STATISTICAL",
                  "string_value": "latte"
                }
              ]
            },
            "FLAVOR": {
              "entities": [
                {
                  "text_range": {
                    "start_index": 20,
                    "end_index": 30
                  },
                  "confidence": 0.9367110133171082,
                  "origin": "STATISTICAL",
                  "string_value": "strawberry"
                }
              ]
            }
          }
        }
      }
    ]
  }
}

Sample Python wordsets client

Wordset client overview

This basic Python app, wordset_client.py allows you to compile wordsets, retrieve wordset metadata, and delete wordsets. The operation to perform is selected with a command line argument. 

import argparse
import grpc
import json
import logging

from nuance.nlu.wordset.v1beta1.wordset_pb2 import *
from nuance.nlu.wordset.v1beta1.wordset_pb2_grpc import *
from nuance.nlu.common.v1beta1.resource_pb2 import *

from google.protobuf.json_format import MessageToJson

log = logging.getLogger(__name__)

def parse_args():
    parser = argparse.ArgumentParser(
        prog="client.py",
        usage="%(prog)s <command> [-options]",
        add_help=False,
        formatter_class=lambda prog: argparse.HelpFormatter(
            prog, max_help_position=45, width=100)
    )

    parser.add_argument('command', metavar='command', nargs='?',
        choices=['compile', 'get-metadata', 'delete'], default='compile',
        help="Command to execute [values: compile, get-metadata, delete] (default: compile)")
    options = parser.add_argument_group("options")
    options.add_argument("-h", "--help", action="help",
                         help="Show this help message and exit.")
    options.add_argument("--token", nargs="?", help=argparse.SUPPRESS)
    options.add_argument("-s", "--serverUrl", metavar="url", nargs="?",
                         help="NLU server URL, default=localhost:9090.", default='localhost:9090')
    options.add_argument("--wordsetFile", type=argparse.FileType("r"), metavar="file",
                         nargs="?", help="Wordset JSON file.")
    options.add_argument("--artifactUrn", nargs="?", metavar="urn", help="Compiled Wordset URN.")
    options.add_argument("--modelUrn", nargs="?", metavar="urn", help="NLU Model URN.")
    options.add_argument("--metadata", metavar="metadata", nargs="+", default=[],
                         help="Wordset metadata defined as one or more key:value pairs.")
    options.add_argument("--clientData", metavar="clientData", nargs="+", default=[],
                         help="Client data defined as one or more key=value pairs.")
    return parser.parse_args()

def create_channel(args):
    channel = None
    call_credentials = None

    if args.token:
        log.debug("Adding CallCredentials with token %s" % args.token)
        call_credentials = grpc.access_token_call_credentials(args.token)

    log.debug("Creating secure gRPC channel")
    channel_credentials = grpc.ssl_channel_credentials()
    channel_credentials = grpc.composite_channel_credentials(channel_credentials, call_credentials)
    channel = grpc.secure_channel(args.serverUrl, credentials=channel_credentials)

    return channel

def create_get_wordset_metadata_request(artifactUrn):
    artifact_reference = ResourceReference(uri=artifactUrn)
    return GetWordsetMetadataRequest(artifact_reference=artifact_reference)

def create_delete_wordset_request(artifactUrn):
    artifact_reference = ResourceReference(uri=artifactUrn)
    return DeleteWordsetRequest(artifact_reference=artifact_reference)

def list_to_dict(list, separator = ':'):
  return dict(entry.split(separator) for entry in list)

def create_compile_wordset_request(args):
  target_artifact_reference = ResourceReference(uri=args.artifactUrn)
  companion_artifact_reference = ResourceReference(uri=args.modelUrn)
  wordset = json.dumps(json.load(args.wordsetFile))
  return CompileWordsetRequest(
    target_artifact_reference=target_artifact_reference,
    companion_artifact_reference=companion_artifact_reference,
    wordset=wordset,
    metadata=list_to_dict(args.metadata),
    client_data=list_to_dict(args.clientData))

def compile_wordset(args):
  with create_channel(args) as channel:
    stub = WordsetStub(channel)
    compiled_wordset_request = create_compile_wordset_request(args)
    for message in stub.CompileWordsetAndWatch(compiled_wordset_request):
      print(MessageToJson(message))

def get_wordset_metadata(args):
  with create_channel(args) as channel:
    stub = WordsetStub(channel)
    response = stub.GetWordsetMetadata(create_get_wordset_metadata_request(args.artifactUrn))
    print(MessageToJson(response))

def delete_wordset(args):
  with create_channel(args) as channel:
    stub = WordsetStub(channel)
    response = stub.DeleteWordset(create_delete_wordset_request(args.artifactUrn))
    print(MessageToJson(response))

def main():
  args = parse_args()
  log_level = logging.DEBUG
  logging.basicConfig(
    format='%(lineno)d %(asctime)s %(levelname)-5s: %(message)s', level=log_level)
  switcher = {
    'compile' : compile_wordset,
    'get-metadata' : get_wordset_metadata,
    'delete' : delete_wordset
  }
  switcher.get(args.command)(args)
  print("Done")

if __name__ == '__main__':
  main()

This section contains a fully-functional sample Python client application consisting of the following files:

This sample Python wordsets app performs the following tasks:

Wordset client requirements

To run this sample app, you need:

Wordset client procedure

Note: Steps 2-4 are the same as for the runtime client, so if you have already carried this out for the runtime client, you can skip.

To run this simple application:

Step 1. Download the sample app here and unzip it in a working directory (for example, /home/userA/nlu-wordset-sample-python-app).

Step 2. Download the gRPC .proto files here and unzip the files in the sample app working directory.

Step 3. Navigate to the sample app working directory and install the required dependencies:

$ python3 -m venv env
$ source env/bin/activate    
$ pip install --upgrade pip
$ pip install grpcio
$ pip install grpcio-tools
$ mkdir -p google/api
$ curl https://raw.githubusercontent.com/googleapis/googleapis/master/google/api/annotations.proto \
google/api/annotations.proto
$ curl https://raw.githubusercontent.com/googleapis/googleapis/master/google/api/http.proto \
google/api/http.proto

Step 4. (If not done previously) Generate the client stubs:

$ python -m grpc_tools.protoc --proto_path=./ --python_out=./ \
--grpc_python_out=./ nuance/nlu/v1/runtime.proto
$ python -m grpc_tools.protoc --proto_path=./ --python_out=./ \
nuance/nlu/v1/result.proto
$ python -m grpc_tools.protoc --proto_path=./ --python_out=./ \
nuance/nlu/v1/interpretation-common.proto
$ python -m grpc_tools.protoc --proto_path=./ --python_out=./ \
nuance/nlu/v1/single-intent-interpretation.proto
$ python -m grpc_tools.protoc --proto_path=./ --python_out=./ \
nuance/nlu/v1/multi-intent-interpretation.proto
$ python -m grpc_tools.protoc --proto_path=./ --python_out=./ \
--grpc_python_out=./ nuance/nlu/wordset/v1beta1/wordset.proto
$ python -m grpc_tools.protoc --proto_path=./ --python_out=./ \
nuance/nlu/common/v1beta1/job.proto
$ python -m grpc_tools.protoc --proto_path=./ --python_out=./ \
nuance/nlu/common/v1beta1/resource.proto
$ python -m grpc_tools.protoc --proto_path=./ --python_out=./ \
nuance/rpc/status.proto
$ python -m grpc_tools.protoc --proto_path=./ --python_out=./ \
nuance/rpc/status_code.proto
$ python -m grpc_tools.protoc --proto_path=./ --python_out=./ \
nuance/rpc/error_details.proto

Step 5. Edit the script wordset-nlu-client.sh, to add your CLIENT_ID and SECRET. These are your Mix credentials as described in Generate token.

#!/bin/bash

# Remember to change the colon (:) in your CLIENT_ID to code %3A
CLIENT_ID="appID%3A...ENTER MIX CLIENT_ID..."
SECRET="...ENTER MIX SECRET..."
export TOKEN="`curl -s -u "$CLIENT_ID:$SECRET" "https://auth.crt.nuance.co.uk/oauth2/token" \
-d "grant_type=client_credentials" -d "scope=tts nlu asr nlu.wordset" \
| python -c 'import sys, json; print(json.load(sys.stdin)["access_token"])'`"

python ./wordset_client.py compile \
    --serverUrl nlu.api.nuance.co.uk:443 --token $TOKEN \
    --wordsetFile "$1" \
    --artifactUrn "$2" \
    --modelUrn "$3" \
    --metadata "$4"

Step 6. Run the application using the script file, passing it:

./wordset-nlu-client.sh wordsetFile artifactUrn modelUrn wordsetUrn

Where:

For example:

$ ./wordset-nlu-client.sh "wordset.json" "mix:tag:wordset:lang/bakery/item_names/eng-USA/mix.nlu" "urn:nuance-mix:tag:model/bakery/mix.nlu?=language=eng-USA"

Here is an example of the script output:

     
55 2021-03-30 14:18:18,855 DEBUG: Adding CallCredentials with token $TOKEN
59 2021-03-30 14:18:18,857 DEBUG: Creating secure gRPC channel
{
  "job_status_update": {
    "status": "JOB_STATUS_PROCESSING"
  },
  "request_status": {
    "status_code": "OK"
  }
}
{
  "job_status_update": {
    "status": "JOB_STATUS_COMPLETE"
  },
  "request_status": {
    "status_code": "OK"
  }
}

Reference topics

This section provides more information about topics in the gRPC API.

Status messages and codes

gRPC includes error and exception handling facilities. Use these gRPC features to confirm the success or failure of a request.

NLU also returns a status message confirming the outcome of an InterpretRequest call. The status field in InterpretResponse contains an HTTP status code, a brief description of this status, and, possibly, a longer description.

An HTTP status code of 200 means that NLU successfully interpreted the input. Values in the 400 range indicate an error in the request that your client app sent. Values in the 500 range indicate an internal error with NLUaaS.

Code Indicates
200 Success
400 Bad request. Your client app sent a malformed or unsupported request.
401 Unauthorized. Your client has not authorized properly.
403 Forbidden. Permission was denied.
404 Not found. A resource, such as a model or word list, does not exist or could not be accessed.
415 Unsupported resource type.
500-511 Error: Internal error.

Wordsets

A wordset is a collection of words and short phrases that extends vocabulary by providing additional values for dynamic list entities in a model. For example, a wordset might provide the names in a user’s contact list or local place names. Like models, wordsets are declared with InterpretRequest - resources.

Defining wordsets

The wordset is defined in JSON format as one or more arrays. Each array is named after a dynamic list entity defined within a semantic model. Wordsets allow you to add values and literals to such entities at runtime.

For example, you might have an entity, CONTACTS, containing personal names, or CITY, with place names used by the application. The wordset adds to the existing terms in the entity, but applies only to the current session. The terms in the wordset are not added permanently to the entity.

All entities must be defined in the semantic model, which are loaded and activated along with the wordset.

The wordset includes additional values for one or more entities. The syntax is:

{
   "entity-1" : [
      { "canonical": "value"
        "literal": "written form"
      },
      { "canonical": "value"
        "literal": "written form"
      },
      ...
   ],
   "entity-n": [ ... ]
}

Syntax

Field Type Description
entity String The name of a dynamic list entity defined in a model. The name is case-sensitive. Consult the model for entity names.
canonical String (Optional) The canonical value for the entity to be returned by interpretation by NLU. If not provided, the literal is used.
literal String A written form by which a user could realistically refer to the value.

For ASR purposes, wordsets can also include a field "spoken" to indicate pronunciations of the value. If you happen to reuse the same wordset for both ASR and NLU, this field is ignored by NLU.

You can provide the wordset either inline in the request or reference a compiled wordset using its URN in the Mix environment.

Inline wordsets

Wordset defined inline

# Define semantic model
semantic_model = ResourceReference(
        type = 'SEMANTIC_MODEL', 
        uri = 'urn:nuance-mix:tag:model/bank-app/mix.nlu?=language=eng-USA'
        )

# Define the wordset inline 
payees_wordset = InterpretationResource(
    inline_wordset = '{"PAYEE":[{"canonical": AMEX","literal":"amex"},{"canonical":"AMEX","literal":"american express"},{"canonical":"VISA","literal":"visa"},{"canonical":"SOCALGAS","literal":"southern california gas"},{"canonical":"SOCALGAS","literal":"southern california gas company"},{"canonical":"SOCALGAS","literal":"the gas company"},{"canonical":"SOCALGAS","literal":"socal gas"}]}')

# Include the semantic model and wordset in InterpretRequest
interpret_request = InterpretRequest(
    parameters = InterpretationParameters(...),
    model = semantic_model
    resources = [ payees_wordset ]
)

Wordset read from a local file using Python function

# Define semantic model
semantic_model = ResourceReference(
        type = 'SEMANTIC_MODEL', 
        uri = 'urn:nuance-mix:tag:model/bank-app/mix.nlu?=language=eng-USA'
        )

# Read wordset from local file 
payees_wordset_content = None
with open('payees-wordset.json', 'r') as f:
    payees_wordset_content = f.read()
payees_wordset = InterpretationResource(
    inline_wordset = payees_wordset_content)

# Include the semantic model and wordset in InterpretRequest
interpret_request = InterpretRequest(
    parameters = InterpretationParameters(...),
    model = semantic_model
    resources = [ payees_wordset ]
)

For inline wordsets, the contents are provided as a string as part of the request object. You can either include the string literal directly in the request or read the string in from a local file programmatically.

The example wordset below extends a PAYEE entity in the model with additional payees. 

{
  "PAYEE" : [
    {
      "canonical" : "AMEX",
      "literal" : "amex"
    },
    {
      "canonical" : "AMEX",
      "literal" : "american express"
    },
    {
      "canonical" : "VISA",
      "literal" : "visa"
    },
    {
      "canonical" : "SOCALGAS",
      "literal" : "southern california gas"
    },
    {
      "canonical" : "SOCALGAS",
      "literal" : "southern california gas company"
    },
    {
      "canonical" : "SOCALGAS",
      "literal" : "the gas company"
    },  
    {
      "canonical" : "SOCALGAS",
      "literal" : "socal gas"
    },  
    …
  ]
}

To use a source wordset, specify it as inline_wordset in InterpretationResource:

Compiled wordsets

Compiled wordset

# Define semantic model as before
semantic_model = ResourceReference(
        type = 'SEMANTIC_MODEL', 
        uri = 'urn:nuance-mix:tag:model/bank-app/mix.nlu?=language=eng-USA'
        )

# Define a compiled wordset (here its context is the same as the semantic model)
payees_compiled_ws = InterpretationResource(
    external_reference = ResourceReference(
        type = 'COMPILED_WORDSET',
        uri = 'urn:nuance-mix:tag:wordset:lang/bank-app/payees-compiled-ws/eng-USA/mix.nlu')
)

# Include the semantic model and wordset in InterpretRequest
interpret_request = InterpretRequest(
    parameters = InterpretationParameters(...),
    model = semantic_model
    resources = [ payees_compiled_ws ]
)

Alternatively, you may reference a compiled wordset that was created with the Wordset API. To use a compiled wordset, specify it in ResourceReference as COMPILED_WORDSET and provide its URN in the Mix environment.

This wordset extends the PAYEE entity in the model with travel locations.

Inline or compiled?

Wordsets can be brought in to aid in interpretation in one of two ways depending on the size of the wordset:

If you are unsure of which approach to take, test the latency when using wordsets inline.

Wordset URNs

To compile a wordset, the following need to be provided:

Wordsets can be either:

The URN for the wordset needs to have one of the following structures, depending on the level of the wordset:

Where:

Once the wordset is compiled, it is stored on Mix and can be referenced at runtime by a client application using the same model and wordset URNs.

Scope of compiled wordsets

Wordsets are specific to a Mix App ID and can be used by any Mix applications under the same App ID.

The context tag used for the wordset does not have to match the context tag of the companion model but it is good practice to do so for easier wordset management.

The wordset must be compatible with the companion model. The companion model needs to contain the dynamic list entity that the wordset relates to.

Note that while the Mix ASRaaS provides a similar API for wordsets, wordsets are nevertheless compiled and stored separately for NLUaaS and ASRaaS. If your application uses both services and requires large wordsets for both, the compilation step must be done for each service.

Wordset lifecycle

Wordsets are available for 28 days after compilation, after which they will automatically be deleted and will need to be compiled again.

After a wordset is compiled, the compiled wordset can later be updated.

Compiling a wordset using an existing wordset URN will replace the existing wordset with the newer version if:

Otherwise, the wordset compilation request may return a status ALREADY_EXISTS. In this case, the wordset remains usable at runtime.

Wordsets can also be manually deleted if no longer needed. Once deleted, a wordset is completely removed and can not be restored.

A compiled wordset can only be accessed at runtime within the AppId under which it was compiled. It must be used with a compatible model, ideally the one used at compilation time.

If an NLUaaS runtime request references an incompatible or missing wordset, the request will still succeed, but an error message will be included to indicate that the wordset was incompatible or not found.

For more details about managing compiled wordsets, see the Wordset API documentation.

Wordset metadata

Wordsets have associated metadata. Some metadata keys are available by default. Optionally, you can provide a list of custom metadata to associate with the compiled wordset. One or more metadata entries can be provided as key-value pairs along with the CompileWordsetRequest.

Both default and custom metadata can be retrieved using the Wordset API.

Note: key names must be lower case.

Interpretation results: Intents

This input exactly matches a training sentence for the ASK_JOB intent (note "origin": GRAMMAR). An alternative intent is proposed but with much lower confidence.

"result": {
  "literal": "Do you have any openings for a pastry chef ?",
  "formatted_literal": "Do you have any openings for a pastry chef ?",
  "interpretations": [
    {
      "single_intent_interpretation": {
        "intent": "ASK_JOB",
        "confidence": 1.0,
        "origin": "GRAMMAR"
      }
    },
    {
      "single_intent_interpretation": {
        "intent": "PLACE_ORDER",
        "confidence": 0.00010213560017291456,
        "origin": "STATISTICAL"
      }
    }
  ]
}

This input is similar to the PLACE_ORDER training sentences (note "origin": STATISTICAL) 

"result": {
  "literal": "I'd like to make an order please",
  "formatted_literal": "I'd like to make an order please",
  "interpretations": [
    {
      "single_intent_interpretation": {
        "intent": "PLACE_ORDER",
        "confidence": 0.9779196381568909,
        "origin": "STATISTICAL"
      }
    }
  ]
}

This result returns the intent PLACE_ORDER along with several entities. See Interpretation results: Entities for more results with entities

  "result": {
    "literal": "I'd like to order a blueberry pie",
    "formatted_literal": "I'd like to order a blueberry pie",
    "interpretations": [
      {
        "single_intent_interpretation": {
          "intent": "PLACE_ORDER",
          "confidence": 0.9913266897201538,
          "origin": "STATISTICAL",
          "entities": {
            "FLAVOR": {
              "entities": [
                {
                  "text_range": {
                    "start_index": 20,
                    "end_index": 29
                  },
                  "formatted_text_range": {
                    "start_index": 20,
                    "end_index": 29
                  },
                  "confidence": 0.8997141718864441,
                  "origin": "STATISTICAL",
                  "string_value": "blueberry",
                  "literal": "blueberry",
                  "formatted_literal": "blueberry"
                }
              ]
            },
            "PRODUCT": {
              "entities": [
                {
                  "text_range": {
                    "start_index": 30,
                    "end_index": 33
                  },
                  "formatted_text_range": {
                    "start_index": 30,
                    "end_index": 33
                  },
                  "confidence": 0.8770073652267456,
                  "origin": "STATISTICAL",
                  "string_value": "pie",
                  "literal": "pie",
                  "formatted_literal": "pie"
                }
              ]
            }
          }
        }
      }
    ]
  }

NLU returns two alternative intents for this input: GET_INFO or PLACE_ORDER, both with medium confidence

"result": {
  "literal": "Can I see a price list and place an order",
  "formatted_literal":"Can I see a price list and place an order"
  "interpretations": [
    {
      "single_intent_interpretation": {
        "intent": "GET_INFO",
        "confidence": 0.563047468662262,
        "origin": "STATISTICAL"
      }
    },
    {
      "single_intent_interpretation": {
        "intent": "PLACE_ORDER",
        "confidence": 0.40654945373535156,
        "origin": "STATISTICAL"
      }
    }
  ]
}

Multi-intent interpretation currently returns information similar to single-intent

"result": {
  "literal": "Can I see a price list and place an order",
  "formatted_literal": "Can I see a price list and place an order",
  "interpretations": [
    {
      "multi_intent_intepretation": {
        "root": {
          "intent": {
            "name": "GET_INFO",
            "text_range": {
              "end_index": 41
            },
            "formatted_text_range": {
              "end_index": 41
            },
            "confidence": 0.563047468662262,
            "origin": "STATISTICAL"
          }
        }
      }
    },
    {
      "multi_intent_intepretation": {
        "root": {
          "intent": {
            "name": "PLACE_ORDER",
            "text_range": {
              "end_index": 41
            },
            "formatted_text_range": {
              "end_index": 41
            },
            "confidence": 0.40654945373535156,
            "origin": "STATISTICAL"
          }
        }
      }
    }
  ]
}

The results returned by NLU include one or more candidate intents that identify the underlying meaning of the user's input. (They can also include entities and values, described in Interpretation results: Entities.) You may request either single-intent interpretation or multi-intent interpretation with InterpretationParameters - interpretation_result_type: SINGLE_INTENT or MULTI_INTENT

Single-intent interpretation means that NLU returns one intent for the user's input: the intent that best describes the user's underlying meaning. NLU may return several candidate intents, but they are listed as alternatives rather than as complementary intents.

Multi-intent interpretation requires that your semantic model support this type of interpretation. Currently you cannot create these models in Mix, so the feature is not fully supported, but you may still request multi-interpretation without error. Like single-intent results, multi-intent results contain one best candidate for the user's input, optionally with alternatives.

True multi-intent results show all the intents contained within the user's input. These results will be available in an upcoming release.

Interpretation results: Entities

List: The FLAVOR and PRODUCT entities identify what the user wants to order

  "result": {
    "literal": "I'd like to order a butterscotch cake",
    "formatted_literal": "I'd like to order a butterscotch cake",
    "interpretations": [
      {
        "single_intent_interpretation": {
          "intent": "PLACE_ORDER",
          "confidence": 0.9917341470718384,
          "origin": "STATISTICAL",
          "entities": {
            "FLAVOR": {
              "entities": [
                {
                  "text_range": {
                    "start_index": 20,
                    "end_index": 32
                  },
                  "formatted_text_range": {
                    "start_index": 20,
                    "end_index": 32
                  },
                  "confidence": 0.9559149146080017,
                  "origin": "STATISTICAL",
                  "string_value": "caramel"
                  "literal": "caramel",
                  "formatted_literal": "caramel",
                }
              ]
            },
            "PRODUCT": {
              "entities": [
                {
                  "text_range": {
                    "start_index": 33,
                    "end_index": 37
                  },
                  "formatted_text_range": {
                    "start_index": 33,
                    "end_index": 37
                  },
                  "confidence": 0.9386003613471985,
                  "origin": "STATISTICAL",
                  "string_value": "cake",
                  "literal": "cake",
                  "formatted_literal": "cake"
                }
              ]
            }
          }
        }
      }
    ]
  }

Freeform: The MESSAGE entity matches anything prefixed with "Call someone" or "Ask someone" or "Send this message to someone." An additional list entity, NAMES, captures the "someone."

  "result": {
    "literal": "Ask Jenny When should we arrive",
    "formatted_literal": "Ask Jenny When should we arrive",
    "interpretations": [
      {
        "single_intent_interpretation": {
          "intent": "ASK_RANDOM",
          "confidence": 1.0,
          "origin": "GRAMMAR",
          "entities": {
            "NAMES": {
              "entities": [
                {
                  "text_range": {
                    "start_index": 4,
                    "end_index": 9
                  },
                  "formatted_text_range": {
                    "start_index": 4,
                    "end_index": 9
                  },
                  "confidence": 1.0,
                  "origin": "GRAMMAR",
                  "string_value": "Jenny",
                  "literal": "Jenny",
                  "formatted_literal": "Jenny"
                }
              ]
            },
            "MESSAGE": {
              "entities": [
                {
                  "text_range": {
                    "start_index": 10,
                    "end_index": 31
                  },

                  "formatted_text_range": {
                    "start_index": 10,
                    "end_index": 31
                  },
                  "confidence": 1.0,
                  "origin": "GRAMMAR",
                  "string_value": "",
                  "literal": "when should we arrive",
                  "formatted_literal": "when should we arrive"
                }
              ]
            }
          }
        }
      }
    ]
  }

  "result": {
    "literal": "Send this message to Chris Can you pick me up from the five forty train",
    "formatted_literal": "Send this message to Chris Can you pick me up from the 5:40 train",
    "interpretations": [
      {
        "single_intent_interpretation": {
          "intent": "ASK_RANDOM",
          "confidence": 1.0,
          "origin": "GRAMMAR",
          "entities": {
            "MESSAGE": {
              "entities": [
                {
                  "text_range": {
                    "start_index": 27,
                    "end_index": 65
                  },
                  "formatted_text_range": {
                    "start_index": 27,
                    "end_index": 59
                  },
                  "confidence": 1.0,
                  "origin": "GRAMMAR",
                  "string_value": "",
                  "literal": "can you pick me up from the five forty train",
                  "formatted_literal": "can you pick me up from the 5:40 train"
                }
              ]
            },
            "NAMES": {
              "entities": [
                {
                  "text_range": {
                    "start_index": 21,
                    "end_index": 26
                  },
                  "formatted_text_range": {
                    "start_index": 21,
                    "end_index": 26
                  },
                  "confidence": 1.0,
                  "origin": "GRAMMAR",
                  "string_value": "Chris",
                  "literal": "Chris",
                  "formatted_literal": "Chris"
                }
              ]
            }
          }
        }
      }
    ]
  }

Relationship: DATE is an isA entity that wraps nuance_CALENDARX

  "result": {
    "literal": "I want to pay my Visa bill on February twenty eight",
    "formatted_literal": "I want to pay my Visa bill on February 28",
    "interpretations": [
      {
        "single_intent_interpretation": {
          "intent": "PAY_BILL",
          "confidence": 1.0,
          "origin": "GRAMMAR",
          "entities": {
            "DATE": {
              "entities": [
                {
                  "text_range": {
                    "start_index": 30,
                    "end_index": 51
                  },
                  "formatted_text_range": {
                    "start_index": 30,
                    "end_index": 41
                  },
                  "confidence": 1.0,
                  "origin": "GRAMMAR",
                  "entities": {
                    "nuance_CALENDARX": {
                      "entities": [
                        {
                          "text_range": {
                            "start_index": 30,
                            "end_index": 51
                          },
                          "formatted_text_range": {
                            "start_index": 30,
                            "end_index": 41
                          },
                          "confidence": 1.0,
                          "origin": "GRAMMAR",
                          "struct_value": {
                            "nuance_CALENDAR": {
                              "nuance_DATE": {
                                "nuance_DATE_ABS": {
                                  "nuance_MONTH": 2.0,
                                  "nuance_DAY": 28.0
                                }
                              }
                            }
                          }
                        }
                      ]
                    }
                  }
                }
              ]
            },
            "BILL_TYPE": {
              "entities": [
                {
                  "text_range": {
                    "start_index": 17,
                    "end_index": 21
                  },
                  "formatted_text_range": {
                    "start_index": 17,
                    "end_index": 21
                  },
                  "confidence": 1.0,
                  "origin": "GRAMMAR",
                  "string_value": "Visa",
                  "literal": "Visa",
                  "formatted_literal": "Visa"
                }
              ]
            }
          }
        }
      }
    ]
  }

nuance_CALENDARX is a hasA entity containing several date and time entities

  "result": {
    "literal": "I want to pay my AMEX bill on February twenty eight",
    "formatted_literal": "I want to pay my AMEX bill on February 28",
    "interpretations": [
      {
        "single_intent_interpretation": {
          "intent": "PAY_BILL",
          "confidence": 1.0,
          "origin": "GRAMMAR",
          "entities": {
            "nuance_CALENDARX": {
              "entities": [
                {
                  "text_range": {
                    "start_index": 30,
                    "end_index": 51
                  },
                  "formatted_text_range": {
                    "start_index": 30,
                    "end_index": 41
                  },
                  "confidence": 1.0,
                  "origin": "GRAMMAR",
                  "struct_value": {
                    "nuance_CALENDAR": {
                      "nuance_DATE": {
                        "nuance_DATE_ABS": {
                          "nuance_MONTH": 2.0,
                          "nuance_DAY": 28.0
                        }
                      }
                    }
                  }
                }
              ]
            },
            "BILL_TYPE": {
              "entities": [
                {
                  "text_range": {
                    "start_index": 17,
                    "end_index": 21
                  },
                  "formatted_text_range": {
                    "start_index": 17,
                    "end_index": 21
                  },
                  "confidence": 1.0,
                  "origin": "GRAMMAR",
                  "string_value": "American Express",
                  "literal": "AMEX",
                  "formatted_literal": "AMEX"
                }
              ]
            }
          }
        }
      }
    ]
  }

nuance_AMOUNT is another hasA entity containing multiple entities

  "result": {
    "literal": "i'd like to pay six hundred and twenty five dollars on my hydro bill",
    "formatted_literal": "i'd like to pay $625 on my hydro bill"
    "interpretations": [
      {
        "single_intent_interpretation": {
          "intent": "PAY_BILL",
          "confidence": 1.0,
          "origin": "GRAMMAR",
          "entities": {
            "BILL_TYPE": {
              "entities": [
                {
                  "text_range": {
                    "start_index": 58,
                    "end_index": 63
                  },
                  "formatted_text_range": {
                    "start_index": 27,
                    "end_index": 32
                  },
                  "confidence": 1.0,
                  "origin": "GRAMMAR",
                  "string_value": "Hydro"
                  "literal": "Hydro",
                  "formatted_literal": "Hydro"
                }
              ]
            },
            "nuance_AMOUNT": {
              "entities": [
                {
                  "text_range": {
                    "start_index": 16,
                    "end_index": 51
                  },
                  "formatted_text_range": {
                    "start_index": 16,
                    "end_index": 20
                  },
                  "confidence": 1.0,
                  "origin": "GRAMMAR",
                  "struct_value": {
                    "nuance_UNIT": "USD",
                    "nuance_NUMBER": 625.0
                  }
                }
              ]
            }
          }
        }
      }
    ]
  }

NLU results may also include individual words and phrases in the user's input, mapped to entities and values. The results differ depending on the type of entity that is interpreted. Entity creation and annotation is done in Mix.nlu or in Mix.dialog, where the entity types include: List, Freeform, and Relationship (isA and/or hasA).

List entity

A list entity is an entity with named values. For example, FLAVOR is a list entity that might contain values such as caramel, blueberry, strawberry, chocolate, and so on. Or BILL_TYPE might contain values such as Visa, American Express, Telephone, and Hydro.

Each value has one or more user literals, or ways that users might express this canonical value. For example, the FLAVOR entity matches sentences containing phrases such as:

And the BILL_TYPE entity matches sentences such as:

Freeform entity

A freeform entity has values that are only vaguely defined. Freeform entities can match lengthy user input, but they do not give precise information about the contents of the input. For example, MESSAGE is a freeform entity that occurs after keywords in the user input such as:

"Call Fred to say..."
"Ask Jenny..."
"Send a message to Chris."

Relationship entity

Relationship entities are isA and/or hasA entities.

Interpretation results: Confidence

Interpretation results report confidence levels for both intent and entity interpretations.

Confidence levels can range from 0.0 to 1.0, and give an estimate of the probability that the candidate interpretation is correct.

Exact matches

The highest confidence levels will be for input that is detected as an exact match to the structure and entities of one of the training samples for one of the intents. For example, suppose you have defined an intent named CHECK_ACCOUNT_BALANCE, and that this intent has a training sample "Check my [ACCOUNT_TYPE]savings[/] account balance." Suppose as well that the entity ACCOUNT_TYPE has other values defined—"checking," "mortgage," "joint," and so on. A user input of "Check my mortgage account balance" would come back as an exact match.

Exact matches for text input come back with a confidence = 1.0 and an origin = "GRAMMAR." Exact matches for recognition text coming from an ASR input will also come back with origin = "GRAMMAR" and with a confidence that is high, but somewhat less than 1.0. The reduced confidence reflects the statistical uncertainty inherent to the speech recognition process.

Statistical matches

The other type of match is a statistical match. A statistical match will always have a confidence less than 1.0, and will be indicated with origin = "STATISTICAL". If multiple statistical intent-level matches come back for a user input, the individual confidence values will be normalized so that the confidence values for the interpretation intents add up to 1.0.

For multiple exact matches, this normalization will not take place, whether the match comes from ASR input or text input.

Interpretation results: Literals

Interpretation results, whether single-intent or multi-intent, include string literals at different levels corresponding to the full user input used for interpretation as well as the meaningful portions thereof, for example, the text corresponding to a detected SingleIntentEntity. There are two types of literals that can be returned, literal and formatted_literal.

literal

For literals returned with field name literal, the contents depend on the origin of the user input.

For text input, this just contains the raw input text.

For input coming from ASR results, the literal is a concatenation of audio tokens, separated by spaces, in minimally formatted text format.

For example, "Call five one four nine zero four seven eight zero zero," or "Pay five hundred dollars to my phone bill."

formatted_literal

The field formatted_literal is a formatted version of the literal for user inputs coming from ASR results.

For inputs from ASR results, this field gives a text string representing the ASR result, but with formatted text that attempts to render the text as it would be written in text input for special types of content such as:

For example, a formatted_literal could look like "Call (514) 904-7800" or "Pay $500 to my phone bill."

In the case of text inputs, the contents of this field will be identical to those of literal.

The formatting performed depends on the language.

For more information on formatted text, see Formatted text in the ASRaaS documentation.

Defaults

The proto files provide the following default values for InterpretRequest sent to NLU. Mandatory fields are in bold.

     
Fields in InterpretRequest Default value
InterpretRequest  
    parameters  
        interpretation_result_type SINGLE_INTENT
        interpretation_input_logging_mode PLAINTEXT
        post_processing_script_parameters Blank
        max_interpretations 0: Use the configured setting in the NLU instance.
    model  
        type UNDEFINED_RESOURCE_TYPE: NLU will use the resource provided by the content server.
        uri Mandatory
        request_timeout_ms 0: Use the configured setting in the NLU instance.
        headers Blank
    resources  
        inline_wordset
 Blank
    client_data Blank
    user_id Blank
    input Mandatory

Runtime gRPC API

NLU as a Service provides these protocol buffer (.proto) files to define the Runtime NLU service for gRPC. These files contain the building blocks for the runtime of your NLU applications:

Once you have transformed the proto files into functions and classes in your programming language using gRPC tools, you can call these functions from your client application to start interpreting plain text or results from ASR as a Service.

See Client app development for a samples using Python, Go, and Java. For other languages, consult the gRPC and Protocol Buffer documentation:

Proto file structure

The proto files define an RPC service with an Interpret method containing an InterpretRequest and InterpretResponse. Details about each component are referenced by name within the proto file.

This is the structure of InterpretRequest:

Proto files: request

And this shows InterpretResponse:

Proto files: response

Runtime service

Runtime interpretation service. Use the Interpret method to request an interpretation.

Name Request Type Response Type Description
Interpret InterpretRequest InterpretResponse Starts an interpretation request and returns a response.

InterpretRequest

InterpretRequest example

interpret_req = InterpretRequest(
    parameters=params,
    model=model,
    input=input)

The input to interpret, with parameters, model, extra resources, and client tags to customize the interpretation. Included in Runtime service.

Field Type Description
parameters InterpretationParameters Parameters for the interpretation.
model ResourceReference Mandatory. Semantic model to perform the interpretation.
resources InterpretationResource Repeated. Resources to customize the interpretation.
client_data string,string Key-value pairs to log.
user_id string Identifies a particular user within an application.
input InterpretationInput Mandatory. Input to interpret.

This message includes:

InterpretRequest
    parameters InterpretationParameters
        interpretation_result_type (EnumInterpretationResultType)
        interpretation_input_logging_mode (EnumInterpretationInputLoggingMode)
        post_processing_script_parameters
        max_interpretations
    model (ResourceReference)
    resources (InterpretationResource)
    client_data
    user_id
    input (InterpretationInput)

InterpretationParameters

InterpretationParameters example

params = InterpretationParameters(
    interpretation_result_type=EnumInterpretationResultType.SINGLE_INTENT,
    interpretation_input_logging_mode=EnumInterpretationInputLoggingMode.PLAINTEXT)

Optional parameters controlling the interpretation. Included in InterpretRequest.

Field Type Description
interpretation_result_type EnumInterpretation ResultType Format of interpretation result. Default is SINGLE_INTENT.
interpretation_input_logging_mode EnumInterpretation InputLoggingMode Format for input in the call logs. Default is PLAINTEXT.
max_interpretations uint32 Maximum interpretations for the result. Default is 0 for the NLU server's configured setting.

This message includes:

InterpretRequest
  InterpretationParameters
    interpretation_result_type (EnumInterpretationResultType)
    interpretation_input_logging_mode (EnumInterpretationInputLoggingMode)
    max_interpretations

EnumInterpretationResultType

Format of interpretations result. Included in InterpretationParameters.

Name Number Description
UNKNOWN 0 Default. Same as SINGLE_INTENT.
SINGLE_INTENT 1 Always return a single-intent interpretation.
MULTI_INTENT 2 Always return multi-intent interpretation.

EnumInterpretationInputLoggingMode

Format for input in the diagnostic logs. Included in InterpretationParameters.

Name Number Description
PLAINTEXT 0 Default. Log the literal text of the input.
SUPPRESSED 9 Input is replaced with "value suppressed."

ResourceReference

ResourceReference example

options.add_argument('--modelUrn', nargs="?", 
                     help="NLU Model URN")
. . .
# Reference the model via the app config
model = ResourceReference(
    type=EnumResourceType.SEMANTIC_MODEL,
    uri=args.modelUrn)

Parameters to fetch an external resource. Included in InterpretRequest and InterpretationResource.

Field Type Description
type EnumResourceType Resource type.
uri string URN for the resource.
request_timeout_ms uint32 Time, in ms, to wait for a response from the hosting server. Default is 0 for the NLU server's configured setting.
headers string,string Map of headers to transmit to the server hosting the resource. May include max_age, max_stale, max_fresh, cookies.

This message includes:

InterpretRequest | InterpretationResource
  ResourceReference
    type (EnumResourceType)
    uri
    request_timeout_ms
    headers

EnumResourceType

Specifies a semantic model or wordset. Included in ResourceReference. Use the default, UNDEFINED_RESOURCE_TYPE, to determine the type from the content-type header returned by the resource's server.

Name Number Description
UNDEFINED_RESOURCE_TYPE 0 Default. Use the content-type header from the resource's server to determine the type.
SEMANTIC_MODEL 1 A semantic model from Mix.nlu.
WORDSET 2 Currently unsupported. Use InterpretationResource - inline_wordset.
COMPILED_WORDSET 3 A compiled wordset associated to a specific semantic model.

InterpretationResource

A resource to customize the interpretation. Included in InterpretRequest.

Field Type Description
external_reference ResourceReference External resource.
inline_wordset string Inline wordset, in JSON. See Wordsets.

This message includes:

InterpretRequest
  InterpretationResource
    external_reference (ResourceReference)
    inline_wordset

InterpretationInput

InterpretationInput example

def parse_args():
    options.add_argument("--textInput", metavar="file", nargs="?",
        help="Text to perform interpretation on")
. . .
input = InterpretationInput(
    text=args.textInput)

Input to interpret. Included in InterpretRequest. Use either text or the result from ASR as a Service.

Field Type Description
text string Text input.
asr_result nuance.asr.v1.RecognitionResponse RecognitionResponse from ASR as a Service.

This message includes:

InterpretRequest
  InterpretationInput
    text
    or
    asr_result

InterpretResponse

The interpretation result. Included in Runtime service.

Field Type Description
status Status Whether the request was successful. The 200 code means success, other values indicate an error.
result InterpretResult The result of the interpretation.

This message includes:

InterpretResponse
  status (Status)
  result (InterpretResult)
    literal
    interpretations (Interpretation)

Status

A Status message indicates whether the request was successful or reports errors that occurred during the request. Included in InterpretResponse.

Field Type Description
code uint32 HTTP status code. The 200 code means success, other values indicate an error.
message string Brief description of the status.
details string Longer description if available.

InterpretResult

For examples, see Interpretation results: Intents and Interpretation results: Entities.

Result of interpretation. Contains the input literal and one or more interpretations. Included in InterpretResponse.

Field Type Description
literal string Input used for interpretation. For text input, this is always the raw input text. For input coming from ASR as a Service results, this gives a concatenation of the audio tokens, separated by spaces, in minimally formatted text format. For example, "Pay five hundred dollars."
formatted_literal string The formatted input literal. For input coming from ASR as a Service results, this is the text representation of the ASR result, but with formatted text. For example, "Pay $500." When the input for interpretation is text, this is the same as literal.
interpretations Interpretation Repeated. Candidate interpretations of the input.
sensitive bool Indicates whether the literal contains entities flagged as sensitive. Sensitive entities are masked in call logs.

This message includes:

InterpretResponse
  InterpretResult
    literal
    formatted_literal
    interpretations (Interpretation)
    sensitive

Interpretation

Candidate interpretation of the input. Included in InterpretResult.

The interpret request specifies the type of interpretation: either single-intent or multi-intent (see InterpretRequest - InterpretationParameters - interpretation_result_type). Multi-intent interpretation requires a semantic model that is enabled for multi-intent (not currently supported in Mix.nlu). See Interpretation results for details and more examples.

Field Type Description
single_intent_interpretation SingleIntentInterpretation The result contains one intent.
multi_intent_interpretation MultiIntentInterpretation The result contains multiple intents. This choice requires a multi-intent semantic model, which is not currently supported in Nuance-hosted NLUaaS.

This message includes:

InterpretResult
  Interpretation
    single_intent_interpretation (SingleIntentInterpretation)
    or
    multi_intent_interpretation (MultiIntentInterpretation)

SingleIntentInterpretation

Single intent interpretation returns the most likely intent, PLACE_ORDER, and an alternative, PAY_BILL, with a much lower confidence rate

  "result": {
    "literal": "I'd like to place an order",
    "formatted_literal": "I'd like to place an order",
    "interpretations": [
      {
        "single_intent_interpretation": {
          "intent": "PLACE_ORDER",
          "confidence": 0.9941431283950806,
          "origin": "STATISTICAL"
        }
      },
      {
        "single_intent_interpretation": {
          "intent": "PAY_BILL",
          "confidence": 0.0019808802753686905,
          "origin": "STATISTICAL"
        }
      }
    ]
  }
}

Single-intent interpretation results. Included in Interpretation. Theses results include one or more alternative intents, complete with entities if they occur in the text. Each intent is shown with a confidence score and whether the match was done from a grammar file or an SSM (statistical) file.

Field Type Description
intent string Intent name as specified in the semantic model.
confidence float Confidence score (between 0.0 and 1.0 inclusive). The higher the score, the likelier the detected intent is correct.
origin EnumOrigin How the intent was detected.
entities string,SingleIntentEntityList Map of entity names to lists of entities: key, entity list.

This message includes:

InterpretResponse
  InterpretResult
    SingleIntentInterpretation
      intent
      confidence
      origin (EnumOrigin)
      entities (key,SingleIntentEntityList)
        entity (SingleIntentEntity)

EnumOrigin

Origin of an intent or entity. Included in SingleIntentInterpretation, SingleIntentEntity, IntentNode, and EntityNode.

Name Number Description
UNKNOWN 0
GRAMMAR 1 Determined from an exact match with a grammar file in the model.
STATISTICAL 2 Determined statistically from the SSM file in the model.

SingleIntentEntityList

List of entities. Included in SingleIntentInterpretation.

Field Type Description
entities SingleIntentEntity Repeated. An entity match in the intent, for single-intent interpretation.

SingleIntentEntity

Single intent, PLACE_ORDER, with entity FLAVOR: strawberry and PRODUCT: cheesecake

  "result": {
    "literal": "I want to order a strawberry cheesecake",
    "formatted_literal": "I want to order a strawberry cheesecake",
    "interpretations": [
      {
        "single_intent_interpretation": {
          "intent": "PLACE_ORDER",
          "confidence": 0.9987062215805054,
          "origin": "STATISTICAL",
          "entities": {
            "FLAVOR": {
              "entities": [
                {
                  "text_range": {
                    "start_index": 18,
                    "end_index": 28
                  },
                  "formatted_text_range": {
                    "start_index": 18,
                    "end_index": 28
                  },
                  "confidence": 0.9648909568786621,
                  "origin": "STATISTICAL",
                  "string_value": "strawberry",
                  "literal": "strawberry",
                  "formatted_literal": "strawberry"
                }
              ]
            },
            "PRODUCT": {
              "entities": [
                {
                  "text_range": {
                    "start_index": 29,
                    "end_index": 39
                  },
                  "formatted_text_range": {
                    "start_index": 29,
                    "end_index": 39
                  },
                  "confidence": 0.9585300087928772,
                  "origin": "STATISTICAL",
                  "string_value": "cheesecake",
                  "literal": "cheesecake",
                  "formatted_literal": "cheesecake",
                }
              ]
            }
          }
        }
      }
    ]
  }
}

Entity in the intent. Included in SingleIntentEntityList.

Field Type Description
text_range TextRange Range of literal text for which this entity applies.
formatted_text_range TextRange Range of the formatted literal text for which this entity applies. When the input for interpretation comes from an ASR result, this may be absent if there is misalignment.
confidence float Confidence score between 0.0 and 1.0 inclusive. The higher the score, the likelier the entity detection is correct.
origin EnumOrigin How the entity was detected.
entities string, SingleIntentEntityList For hierarchical entities, the child entities of the entity: key, entity list.
string_value string The canonical value as a string.
struct_value google.protobuf. Struct The entity value as an object. This object may be directly converted to a JSON representation.
literal string The input literal associated with this entity. For text input, this is the raw input text. For input coming from ASR results, this is in minimally formatted text format.
formatted_literal string The formatted input literal associated with the entity. When the input for interpretation is text, this is the same as literal.
sensitive bool Indicates whether the entity has been flagged as sensitive. Sensitive entities are masked in call logs.
audio_range AudioRange Range of audio input this operator applies to. Available only when interpreting a recognition result from ASR as a Service.

This message includes:

InterpretResponse
  InterpretResult
    SingleIntentInterpretation
      SingleIntentEntityList
        SingleIntentEntity
          text_range (TextRange)
          formatted_text_range (TextRange)
          confidence
          origin (EnumOrigin)
          entities (key, SingleIntentEntityList)
          string_value
          struct_value
          literal
          formatted_literal
          sensitive
          audio_range (AudioRange)

TextRange

Range of text in the input literal. Included in SingleIntentEntity, OperatorNode, IntentNode, and EntityNode.

Field Type Description
start_index uint32 Inclusive, 0-based character position, in relation to literal text.
end_index uint32 Exclusive, 0-based character position, in relation to literal text.

AudioRange

Range of time in the input audio. Included in and SingleIntentEntity, OperatorNode, IntentNode, and EntityNode. Available only when interpreting a recognition result from ASR as a Service.

Field Type Description
start_time_ms uint32 Inclusive start time in milliseconds.
end_time_ms uint32 Exclusive end time in milliseconds.

MultiIntentInterpretation

Multi-intent interpretation against a standard semantic model returns just one intent for each candidate in the "root" object

  "result": {
    "literal": "Price list and place order",
    "formatted_literal": "Price list and place order",
    "interpretations": [
      {
        "multi_intent_intepretation": {
          "root": {
            "intent": {
              "name": "GET_INFO",
              "text_range": {
                "end_index": 26
              },
              "formatted_text_range": {
                "end_index": 26
              },
              "confidence": 0.8738054037094116,
              "origin": "STATISTICAL"
            }
          }
        }
      },
      {
        "multi_intent_intepretation": {
          "root": {
            "intent": {
              "name": "PLACE_ORDER",
              "text_range": {
                "end_index": 26
              },
              "formatted_text_range": {
                "end_index": 26
              },
              "confidence": 0.08993412554264069,
              "origin": "STATISTICAL"
            }
          }
        }
      }
    ]
  }

Multi-intent interpretation. Contains a tree of nodes representing the detected operators, intents, and entities and their associations. Included in Interpretation.

Multi-intent interpretation may be requested without error, but it is not currently supported as it requires a multi-intent semantic model, not yet available in Mix. When requesting multi-intent interpretation against a single-intent model, the results contain the same information as a single-intent interpretation, but formatted slightly differently: the root of the multi-intent interpretation contains the intent.

Field Type Description
root InterpretationNode Root node of the interpretation tree. Can be either OperatorNode or IntentNode.

This message includes:

InterpretResponse
  InterpretResult
    MultiIntentInterpretation
      root (InterpretationNode)

InterpretationNode

Node in the interpretation tree. Included in MultiIntentInterpretation.

Field Type Description
operator OperatorNode The relationship of the intents or entities.
intent IntentNode The intents detected in the user input.
entity EntityNode The entities in the intent.

This message includes:

InterpretResponse
  InterpretResult
    MultiIntentInterpretation
      InterpretationNode
        operator (OperatorNode)
        intent (IntentNode)
        entity (EntityNode)

OperatorNode

Logical operator node. Included in InterpretationNode.

Field Type Description
operator EnumOperator Type of operator.
text_range TextRange Range of the literal text this operator applies to.
formatted_text_range TextRange Range of the formatted literal text this operator applies to. When the input for interpretation is an ASR result, this may be absent if there is misalignment.
children InterpretationNode Repeated. Child nodes for this operator. An operator node always has children.
literal string The input literal associated with this operator. For text input, this is the raw input text. For input coming from ASR as a Service results, this is in minimally formatted text format.
formatted_literal string The formatted input literal associated with this operator. When the input for interpretation is text, this is the same as literal.
sensitive bool Indicates whether a child entity of this operator node has been flagged as sensitive. Sensitive entities are masked in call logs.
audio_range AudioRange Range of audio input this operator applies to. Available only when interpreting a recognition result from ASR as a Service.

This message includes:

InterpretResponse
  InterpretResult
    MultiIntentInterpretation
      InterpretationNode
        OperatorNode
          operator (EnumOperator)
          text_range (TextRange)
          formatted_text_range (TextRange)
          children (InterpretationNode)
          literal
          formatted_literal
          sensitive
          audio_range (AudioRange)

EnumOperator

Logical operator type, AND, OR, or NOT. Included in OperatorNode.

Name Number Description
AND 0 The following item is an additional intent or entity.
OR 1 The following item is an alternative intent or entity.
NOT 2 The following item is not detected.

IntentNode

Node representing an intent. Included in InterpretationNode.

Field Type Description
name string Intent name as specified in the semantic model.
text_range TextRange Range of literal text this intent applies to.
formatted_text_range TextRange Range of the formatted literal text this operator applies to. When the input for interpretation is an ASR result, this may be absent if there is misalignment.
confidence float Confidence score between 0.0 and 1.0 inclusive. The higher the score, the likelier the detected intent is correct.
origin EnumOrigin How the intent was detected.
children InterpretationNode Repeated. Child nodes for this intent. An intent node has zero or more child nodes.
literal string The input literal associated with this intent. For text input, this is the raw input text. For input coming from ASR as a Service results, this is in minimally formatted text format.
formatted_literal string The formatted input literal associated with this intent. When the input for interpretation is text, this is the same as literal.
sensitive bool Indicates whether a child entity of this intent node has been flagged as sensitive. Sensitive entities are masked in call logs.
audio_range AudioRange Range of audio input this operator applies to. Available only when interpreting a recognition result from ASR as a Service.

This message includes:

InterpretResponse
  InterpretResult
    MultiIntentInterpretation
      InterpretationNode
        IntentNode
          name
          text_range (TextRange)
          formatted_text_range (TextRange)
          confidence
          origin (EnumOrigin)
          children (InterpretationNode)
          literal
          formatted_literal
          sensitive
          audio_range (AudioRange)

EntityNode

Node representing an entity. Included in InterpretationNode.

Field Type Description
name string Entity name as specified in the semantic model.
text_range TextRange Range of literal text this intent applies to.
formatted_text_range TextRange Range of the formatted literal text this operator applies to. When the input for interpretation is an ASR result, this may be absent if there is misalignment.
confidence float Confidence score between 0.0 and 1.0 inclusive. The higher the score, the likelier the detected entity is correct.
origin EnumOrigin How the intent was detected.
children InterpretationNode Repeated. Child nodes for this entity. A hierarchical entity node can have child entity and operator nodes. Entity nodes currently never have intent nodes as children.
string_value string The value of the entity as specified in the semantic model.
struct_value google. protobuf.Struct Structured data, ready to convert to a JSON representation.
literal string The input literal associated with this entity. For text input, this is the raw input text. For input coming from ASR results, this is in minimally formatted text format.
formatted_literal string The formatted input literal associated with this entity. When the input for interpretation is text, this is the same as literal.
sensitive bool Indicates whether the entity has been flagged as sensitive. Sensitive entities are masked in call logs.
audio_range AudioRange Range of audio input this operator applies to. Available only when interpreting a recognition result from ASR as a Service.

This message includes:

InterpretResponse
  InterpretResult
    MultiIntentInterpretation
      InterpretationNode
        EntityNode
          name
          text_range (TextRange)
          formatted_text_range (TextRange)
          confidence
          origin (EnumOrigin)
          children (InterpretationNode)
          string_value
          struct_value
          literal
          formatted_literal
          sensitive
          audio_range (AudioRange)

Wordset gRPC API

NLU as a Service provides a set of protocol buffer (.proto) files to define a gRPC Wordset service. These files allow you to use large wordsets with your NLUaaS applications:

Wordset proto file structure

The file wordset.proto defines a Wordet service with several RPC methods for creating and managing wordsets.

Wordset messages

For the nuance.nlu.common.v1beta1 messages, see Common messages.

For the nuance.rpc.Status and related messages, see RPC status messages.

Job status vs. request status

This API includes two types of status:

Wordset service

The Wordset service offers three RPC methods to compile and manage wordsets.

Name Request Type Response Type Description
CompileWordsetAndWatch CompileWordset Request WatchJobStatus Response stream Submit and watch for job completion (server streaming)
GetWordsetMetadata GetWordset MetadataRequest GetWordset MetadataResponse Gets a compiled wordset's metadata (unary)
DeleteWordset DeleteWordset Request DeleteWordset Response Delete the compiled wordset (unary).

CompileWordsetAndWatch

Process flow of CompileWordsetAndWatch method

CompileWordsetAndWatch method

This RPC method submits a request to compile a wordset and returns streaming messages from the server until the job completes. It consists of CompileWordsetRequest and WatchJobStatusResponse.

The method submits a wordset to be compiled and starts a batch compilation job. The response is a server stream of job progress notifications, which stays alive until the end of the compilation job, followed by a final job status.

The WatchJobStatusResponse returns these notifications:

CompileWordsetRequest

Request to compile a wordset related to a dynamic list entity.

Field Type Description
wordset string Mandatory. Inline wordset JSON resource. Note: 4 MB request size limit.
companion_artifact_reference nuance.nlu.common.v1beta1. ResourceReference Mandatory. URN reference to the NLU model in which the entity is defined.
target_artifact_reference nuance.nlu.common.v1beta1. ResourceReference Mandatory. URN reference to use with the compiled wordset. The URN will later be used to reference the wordset to use it as an interpretation resource.
metadata string,string Client-supplied key,value pairs to associate with the artifact. Keys can only contain lower case characters.
client_data string,string Client-supplied key,value pairs to inject into the logs.

This message includes:

CompileWordsetRequest
  wordset
  companion_artifact_reference (nuance.nlu.common.v1beta1.ResourceReference)
  target_artifact_reference (nuance.nlu.common.v1beta1.ResourceReference)
  metadata
  client_data

WatchJobStatusResponse

A series of these are streamed in response to a CompileWordset request. The responses provide status on both the wordset compilation job and the CompileWordsetAndWatch gRPC request.

Field Type Description
job_status_update nuance.nlu.common.v1beta1. JobStatusUpdate Immediate job status
request_status nuance.rpc.Status Request status. Returned once at the end.

This message includes:

WatchJobStatusResponse
  job_status_update (nuance.nlu.common.v1beta1.JobStatusUpdate)
  request_status (nuance.rpc.Status)

GetWordsetMetadata

Process flow for GetWordsetMetadata

GetWordsetMetadata method

This RPC method requests and returns information about a compiled wordset. It does not return the content of the wordset. It provides two types of metadata:

GetWordsetMetadataRequest

Request for a GetWordsetMetadata request.

Field Type Description
artifact_reference nuance.nlu.common.v1beta1. ResourceReference Reference to the wordset artifact.

This message includes:

GetWordsetMetadataRequest
  artifact_reference (nuance.nlu.common.v1beta1.ResourceReference)

GetWordsetMetadataResponse

Response to a GetWordsetMetadata request.

Field Type Description
metadata string,string Default and client-supplied key,value pairs.
request_status nuance.rpc.Status Indicates whether fetching of metadata was done successfully.

This message includes:

GetWordsetMetadataResponse
  metadata
  request_status (nuance.rpc.Status)

DeleteWordset

Process flow for DeleteWordset

DeleteWordset method

This RPC method deletes a specified wordset.

DeleteWordsetRequest

Request to a DeleteWordset request.

Field Type Description
artifact_reference nuance.nlu.common.v1beta1. ResourceReference Reference to the wordset artifact.

This message includes:

DeleteWordsetRequest
  artifact_reference (nuance.nlu.common.v1beta1.ResourceReference)

DeleteWordsetResponse

Response to a DeleteWordset request.

Field Type Description
request_status nuance.rpc.Status Indicates whether deletion was done successfully.

This message includes:

DeleteWordsetResponse
  request_status (nuance.rpc.Status)

Common messages

These messages are part of the nuance.nlu.common.v1beta1 package and are used to reference external files like NLU models and to report job statuses.

Common reference and job status messages

nuance.nlu.common.v1beta1.ResourceReference

Input message for fetching an external resource.

Field Type Description
uri string Location of the resource as a URN reference
headers string, string Field for internal use

nuance.nlu.common.v1beta1.JobStatusUpdate

Job status update to a request.

Field Type Description
job_id string Not used.
status JobStatus Job status
messages JobMessage Repeated. Messages specifying about the job failure or completion.

nuance.nlu.common.v1beta1.JobStatus

Job Status Enum

Name Number Description
JOB_STATUS_UNKNOWN 0 Job status not specified or unknown.
JOB_STATUS_QUEUED 1 Not used.
JOB_STATUS_PROCESSING 2 Job is processing.
JOB_STATUS_COMPLETE 3 Job is complete.
JOB_STATUS_FAILED 4 Job has failed.

nuance.nlu.common.v1beta1.JobMessage

Job Message

Field Type Description
code int32 Code
message string Message
data string, string Additional key/value pairs

RPC status messages

These messages are part of the nuance.rpc package and are used for RPC status and error information.

RPC status and error messages

nuance.rpc.Status

Common status message.

Field Type Description
status_code StatusCode Mandatory. Status code, an enum value
status_sub_code int32 Application-specific status sub-code.
http_trans_code int32 Not used.
request_info RequestInfo Information about the original request
status_message LocalizedMessage Message providing the details of this status
help_info HelpInfo Help message providing the possible user action(s)
field_violations FieldViolation Repeated. Set of field violations.
retry_info RetryInfo Retry information
status_details StatusDetail Repeated. Detailed status messages.

nuance.rpc.StatusCode

Set of global canonical status codes to be commonly used by compliant Nuance APIs.

Name Number Description
UNSPECIFIED 0 Unspecified status
OK 1 Success
BAD_REQUEST 2 Invalid message type: the server cannot understand the request.
INVALID_REQUEST 3 The request has an invalid value, is missing a mandatory field, and so on.
CANCELLED_CLIENT 4 Operation terminated by client. The remote system may have changed.
CANCELLED_SERVER 5 Operation terminated by server. The remote system may have changed.
DEADLINE_EXCEEDED 6 The deadline set for the operation has expired.
NOT_AUTHORIZED 7 The client does not have authorization to perform the operation.
PERMISSION_DENIED 8 The client does not have authorization to perform the operation on the requested entities.
NOT_FOUND 9 The requested entity was not found.
ALREADY_EXISTS 10 Cannot create entity as it already exists.
NOT_IMPLEMENTED 11 Unsupported operation or parameter, for example an unsupported media type.
UNKNOWN 15 Result does not map to any defined status. Other response values may provide request-specific additional information.
The following status codes are less frequently used:
TOO_LARGE 51 A field is too large to be processed due to technical limitations for example a large audio or other binary block. For arbitrary limitations (for example the name must be n characters or less), use INVALID_REQUEST.
BUSY 52 The server understood the request but could not process it due to lack of resources. Retry the request as is later.
OBSOLETE 53 A message type in the request is no longer supported.
RATE_EXCEEDED 54 Similar to BUSY. The client has exceeded the limit of operations per time unit. Retry request as is later.
QUOTA_EXCEEDED 55 The client has exceeded quotas related to licensing or payment. See your client representative for additional quotas.
INTERNAL_ERROR 56 An internal system error experienced while processing the request

nuance.rpc.RequestInfo

A message may use this type to refer to the original request that resulted in an error. This may be particularly useful in the streaming scenarios where the correlation between the request and response may not be so obvious.

Field Type Description
request_id string Identifier of the original request, for example, its OpenTracing id.
request_data string Relevant free format data from the original request, for troubleshooting.
additional_request_data string,string Map of key,value pairs of free format data from the request.

nuance.rpc.LocalizedMessage

A localized message. The choice of the locale is up to the backend. Depending on the use case, it may be the preferred language provided by the browser or the user-specific locale. The server will do its best to determine.

Field Type Description
locale string The locale as xx-XX, for example en-US, fr-CH, es-MX, per the specification bcp47.txt. Default is provided by the server.
message string The message text in the local specified.
message_resource_id string A message identifier, allowing related messages to be provided if needed.

nuance.rpc.HelpInfo

Provides a reference to the help document that may be shown to the end user in order to take an action based on the error/status response. For example, if the original request carried a numerical value that is out of allowed range, this message may be used to point to the documentation that states the valid range.

Field Type Description
links Hyperlink Repeated. Set of links related to the context of the enclosing message
Field Type Description
description LocalizedMessage A description of the link in a specific language.
By default, the server handling the URL manages language selection and detection.
url string The URL to offer to the client, containing help information. If a description is present, this URL should use (or offer) the same locale.

nuance.rpc.FieldViolation

Provides the information about a field violating a rule.

Field Type Description
field string The name of the request field in violation as package.type[.type].field.
rel_field string Repeated. The names of related fields in violation as package.type[.type].field.
user_message LocalizedMessage An error message in a language other than English.
message string An error message in American English.
invalid_value string The invalid value of the field in violation. (Convert non-string data types to string.)
violation ViolationType The reason (enum) a field is invalid. Can be used for automated error handling by the client.

nuance.rpc.ViolationType

Name Number Description
MANDATORY_FIELD_MISSING 0 A required field was not provided.
FIELD_CONFLICT 1 A field is invalid due to the value of another field.
OUT_OF_RANGE 2 A field value is outside the specified range.
INVALID_FORMAT 3 A field value is not in the correct format.
TOO_SHORT 4 A text field value is too short.
TOO_LONG 5 A text field value is too long.
OTHER 64 Violation type is not otherwise listed.
UNSPECIFIED 99 Violation type was not set.

nuance.rpc.RetryInfo

This message is used to control how quickly the client may retry the request if the request is retriable. Failure to respect this retry info may indicate the misbehaving client.

Field Type Description
retry_delay_ms int32 Clients must wait at least this long between retrying the same request.

nuance.rpc.StatusDetail

A status message may have additional details, usually a list of underlying causes of an error. In contrast to field violations, which point to the fields in the original request, status details are not usually directly connected with the request parameters.

Field Type Description
message string The message text in American English.
user_message LocalizedMessage The message text in a language other than English.
extras string, string Map of key,value pairs of additional application-specific information.

Scalar value types

The data types in the proto files are changed to equivalent types in the generated client stub files.

Proto Notes C++ Java Python
double double double float
float float float float
int32 Uses variable-length encoding. Inefficient for encoding negative numbers. If your field is likely to have negative values, use sint32 instead. int32 int int
int64 Uses variable-length encoding. Inefficient for encoding negative numbers. If your field is likely to have negative values, use sint64 instead. int64 long int/long
uint32 Uses variable-length encoding. uint32 int int/long
uint64 Uses variable-length encoding. uint64 long int/long
sint32 Uses variable-length encoding. Signed int value. These encode negative numbers more efficiently than regular int32s. int32 int int
sint64 Uses variable-length encoding. Signed int value. These encode negative numbers more efficiently than regular int64s. int64 long int/long
fixed32 Always four bytes. More efficient than uint32 if values are often greater than 2^28. uint32 int int
fixed64 Always eight bytes. More efficient than uint64 if values are often greater than 2^56. uint64 long int/long
sfixed32 Always four bytes. int32 int int
sfixed64 Always eight bytes. int64 long int/long
bool bool boolean boolean
string A string must always contain UTF-8 encoded or 7-bit ASCII text. string String str/unicode
bytes May contain any arbitrary sequence of bytes. string ByteString str

Change log

2022-08-17

Some minor updates to Status messages and codes.

2022-07-20

Some minor updates to Wordsets.

2022-05-04

Updates to Sample Python runtime client and Sample Python wordsets client to add downloadable files for the sample apps.

2022-03-30

Updates to Sample Python wordsets client.

2022-03-16

Minor updates to sample app run script in Runtime app development.

2021-09-25

2021-08-04

To make use of the new wordset features, follow instructions in gRPC setup to:

  1. Download the latest proto files
  2. Generate the client stubs from the proto files

2021-06-02

2021-05-04

To make use of the new wordset features, follow instructions in gRPC setup to:

  1. Download the latest proto files
  2. Generate the client stubs from the proto files

2021-01-13

2020-12-14

2020-06-24

2020-04-30

2020-03-30

2020-02-19

2020-01-22

2019-12-18

2019-12-11

2019-12-02

2019-11-25

2019-11-15

Below are changes made to the NLUaaS API documentation since the initial Beta release:

Python