Rule-based techniques

Rule-based techniques in NLP are, as the name suggests, based on rules written by human developers, as opposed to machine-learned models derived from data. Rule-based techniques were, for many years, the most common approach to NLP, but as we saw in Chapter 7, rule-based approaches have largely been superseded by numerical, machine-learned approaches for the overall design of most NLP applications. There are many reasons for this; for example, since rules are written by humans, it is possible that they might not cover all situations if the human developer has overlooked something.

However, for practical applications, rules can be very useful, either by themselves or, more likely, along with machine-learned models.

The next section will discuss the motivations for using rules in NLP applications.