Symbolic reasoning derives conclusions by manipulating explicit symbols, facts, and logical rules. In an NLP pipeline, it complements probabilistic models by grounding language in facts that can support more reliable inferences. section 11.0 section 11.1
A knowledge graph stores facts about a chosen world, while a natural-language dependency tree exposes logical relations in text. Together, they separate reasoning from language generation: a language model can produce fluent text while symbolic structures constrain whether its claims are correct. section 11.0 section 11.1
The approach connects several ideas:
- Treats a dependency tree as analogous to a compiler parse tree or abstract syntax tree for extracting relations. section 11.0
- Contrasts symbolic reasoning with probabilistic reasoning used by machine-learning models. section 11.1
- Uses first-order logic and GOFAI as established symbolic approaches, now combined with neural methods. section 11.1
Figure 11.1 An example of a knowledge graph
Inspect a knowledge graph to see how explicit facts provide a substrate for symbolic inference. section 11.0
