KeywordTemplateMatcher

KeywordTemplateMatcher is a Python library for natural language templates. A template can hold slots, optional phrases, and alternatives. The library expands a template into every sentence it describes, and matches an utterance back to the slot values it filled. Use it to prototype NLU systems, voice assistants, or rule-based query matching.

Features

• Template expansion with:
  • Optional phrases ([optional])
  • Alternatives ((choice1|choice2))
  • Slots ({slot_name})
• Slot substitution from a supplied dictionary
• Fuzzy matching and confidence scoring with rapidfuzz
• A simple template structure that extends to any language or grammar rule
• Built-in integration with simplematch for fuzzy slot matching

Installation

pip install keyword-template-matcher

Usage

1. Expand a template with slots

from kw_template_matcher import expand_slots

template = "change [the ]brightness to {brightness_level} and color to {color_name}"
slots = {
    "brightness_level": ["low", "medium", "high"],
    "color_name": ["red", "green", "blue"]
}

for sentence in expand_slots(template, slots):
    print(sentence)

Output:

change the brightness to low and color to red
change brightness to high and color to blue
... (all combinations)

2. Match a template

from kw_template_matcher import TemplateMatcher

matcher = TemplateMatcher()
matcher.add_templates([
    "[hello,] (call me|my name is) {name}",
    "tell me a [{joke_type}] joke"
])

query = "hello, my name is Alice"
results = matcher.match(query)

for match in results:
    print(match)

How it works

Template syntax

Test example templates

templates = [
    "[hello,] (call me|my name is) {name}",
    "tell me a [{joke_type}] joke",
    "play {query} [in ({device_name}|{skill_name}|{zone_name})]"
]

These templates expand to sentences like:

- call me {name}
- hello, my name is {name}
- tell me a {joke_type} joke
- play {query}
- play {query} in {device_name}

Related projects

• OpenVoiceOS/simplematch - the lightweight template parser this library builds on

Contributions

Open an issue or submit a pull request to report a bug or propose an improvement.

Acknowledgements

• rapidfuzz for fast fuzzy matching
• simplematch for lightweight template parsing