This version of the Speechly documentation is no longer actively maintained. For up-to-date documentation, see the latest version.

Imports and Lookups

Imports are an advanced SAL feature that enables you to define long lists outside the SAL syntax. The lookup entity type allows you to define a normalized return value for entities that have several synonyms.

Importing a List from file

The imports feature enables you to define lists outside the SAL configuration, and load them to variables that you can use as part of your SAL templates.

Normally, a List is assigned to a variable in SAL as follows:

fruit = [apple | orange | blood orange | banana | lemon | apricot | peach | pineapple]

However, if your list contains hundreds of items, the above becomes cumbersome to write and maintain.

Using the import functionality, you can import the list from a CSV file. The CSV file must have the list items in one of its columns, and the file must be located in the same directory as your Configuration YAML file.

A simple import example

In the simplest case the CSV file to be imported contains only one column, with one list item per row. Suppose the file is called fruits.csv, and it has the following content:

apple
orange
blood orange
banana
lemon
apricot
peach
pineapple

To import this list from fruits.csv and store it in a variable called fruit, add the following lines to your Configuration YAML:

imports: 
  - name: fruit
    source: fruits.csv
    field: 1

Now, you can use $fruit in your templates, exactly as you would after having defined it in a SAL expression. The imported variable is like any other variable in SAL.

You can specify as many imports as you like. If you had a similar list in another file called vegetables.csv, you can import both lists to their respective variables (fruit and vegetable) by defining:

imports: 
  - name: fruit
    source: fruits.csv
    field: 1
  - name: vegetable
    source: vegetables.csv
    field: 1

Import Reference

Imports are defined by adding the imports key to your Configuration YAML. The value of imports must be a list of dictionaries that each have the keys name, source and field:

imports:
  ...
  - name: name of the variable to which the list is stored
    source: name of the CSV file that contains the list items
    field: the column index (1-based) in the source file that contains the list items
  ...

The lookup Entity Data Type

The returned values of entities having Entity Data Type lookup are normalized according to a simple lookup mechanism. This is useful for mapping synonyms to a normalized value, such as always returning the entity value TV even if the user said “television”, “TV”, or “telly”.

A simple Lookup example

To define a lookup entity, you must specify an explicit mapping that provides for every synonym its normalized value that is returned by the API. This is done by preparing a CSV file with two columns, one of which contains a list of synonyms, and another that contains the normalized values.

For example, suppose the file devices.csv has the following content:

light,light
lights,light
lamp,light
tv,tv
telly,tv
television,tv
radio,radio
stereo,radio

Above, each row defines synonym in the 1st column, and its normalised value in the 2nd column. That is, the synonyms “light”, “lights”, and “lamp” are all mapped to light, the synonyms “tv”, “telly”, and “television” are mapped to tv, and “radio” and “stereo” are both mapped to radio.

Defining a Lookup entity is done by first importing both columns of devices.csv into two variables using the list import mechanism described above:

imports:
  - name: device_as_spoken
    source: devices.csv
    field: 1
  - name: device_normalized
    source: devices.csv
    field: 2

Now $device_as_spoken is a list that contains the 1st column of devices.csv, and $device_normalized contains the 2nd column.

The lookup entity itself is defined using these two variables in the entities section of your Configuration YAML, by specifying the type of the entity as lookup, and defining two additional keys: input_items and output_items as follows:

entities:
  - name: device
    type: lookup
    input_items: $device_as_spoken
    output_items: $device_normalized

Unlike other Entity Data Types that are defined simply by specifying the name of the entity together with its type, the lookup type requires two additional parameters: input_items and output_items, which here are assigned the $device_as_spoken and $device_as_normalized lists, respectively.

When defining your SAL templates, you should only use $device_as_spoken, because remember that the Example utterances you define must accurately reflect how your users talk. For example:

templates: |
  *turn_on [turn | switch] on the $device_as_spoken(device)
  *turn_on [turn | switch] the $device_as_spoken(device) on  

Now, when the user says “turn on the stereo”, the returned device entity has the value radio.

Some remarks about lookup entities

There are a couple of things to remember when using lookups:

  1. If the user says something that the entity detector can identify as a device, but the corresponding synonym is not present in the lookup mapping, the system returns verbatim the user’s expression. For example, if the user says “turn the music player on”, the entity detector probably identifies “music player” as a device, but since “music player” is not defined in the lookup, the returned device entity would have the value music player.

Lookup reference

Lookups are defined by

  1. importing two lists of equal length, called input list and output list, using the list import mechanism,
  2. defining one of these lists as the input_items, and the other as output_items of the lookup entity.

The lookup entity definition has the fields:

entities:
  ...
  - name: name of the entity
    type: lookup
    input_items: reference to the variable with the imported input list
    output_items: reference to the variable with the imported output list
  ...

While we recommend to import both lists from the same CSV file, this is not a requirement. But it is important that the two lists are of equal length, and that the item in the ith position of the output list is the normalized value of the item in the ith position of the input list.


Last updated by Mathias Lindholm on July 20, 2022 at 16:43 +0300

Found an error on our documentation? Please file an issue or make a pull request