Why is dictionary-based approach a heuristic method?

Question

In How can the accuracy of the dictionary-based approach be measured and improved?, one user says that:

dictionary-based approach is a heuristic method

Isn't that this approach is a type of rule-based approach, which on its turn is simply catching keywords using regex? Then shouldn't it be the most accurate approach, comparing to statistic-based approaches?

Research on the internet give me mixed results:

• The article Heuristic vs. Rule-Based Approaches in NLP: What’s the Difference? | by Ajay Verma | Artificial Intelligence in Plain English compares heuristic approach as opposite to rule-based approach. It says:

  heuristic approaches use general, flexible guidelines rather than rigid, predefined rules. These guidelines are often based on common-sense knowledge and intuition, making them adaptable to various scenarios.

• The article The power of Natural Language Processing - TPXimpact compares rule/heuristics-based category as opposite to data-driven category. It says:

  heuristics-based approaches use rules created and programmed into machines (e.g., using templates, grammars or regular expressions)

So which one is correct?

Answer 1

My opinion is that besides intellectual gymnastics, the difference between the two doesn't matter much in practice. To me, it's mostly semantics as I would and have used the two interchangeably.

The main "difference" I could see would be that rules implies that they are stricter than heuristics. For instance:

• "If this rule applies, then I know that I must act like this."
• "If this heuristic applies, then I have some decent amount of confidence that I should act like this"

But let's give some intuition with an example. Let's say that you want to detect e-mail addresses in text.

• A learned approach would probably used some sort of (named-)entity recognition.
• A rule-based approach would use a regular expression of what an e-mail address is known to be (i.e. something like ^[a-zA-Z0-9]+(?:\.[a-zA-Z0-9]+)*@[a-zA-Z0-9]+(?:\.[a-zA-Z0-9]+)*$ )
• A heuristic approach could be, if a token has a "@" character in it, and a ".", and is longer than 5 characters, then it is probably an e-mail.

So, rules imply that the rules used are "complete" and cover the entire set of possibilities, and heuristics are "approximations" or "educated guesses." But as I said previously, I struggle to see, besides a university exam, where this might be relevant to differentiate.

Computer vision is an area where the difference between the two is quite obvious. Indeed, it would be very difficult to build strict rules about what a cat looks like, for instance. But if you say: "If I see a blob of roughly this shape and this color, I'll assume it's a cat," that's a heuristic.

Answer 2

I think of the word heuristics as the intellectual’s version of rules of thumb or “percentage shots” or even Bayesian priors. These are all ways of expressing implicit knowledge.