As explained in a previous post, Structural Semantics is the branch of linguistics which deals with the literal meaning of sentences after abstracting away the specific concepts that non-functional terms express. It specifies how the structural meaning of a sentence is related to the syntactic structure of the sentence and the meaning of functional expressions.
Just to make sure we are clear about what Structural Semantics includes and does not include, I survey here some types of meaning that are not included in this body of knowledge.
Lexical Semantics
Structural Semantics does not include lexical semantics, i.e. knowledge about the meaning of content words and their inter-relationships. Such knowledge is required to understand the difference between the following two cases:
John loaded the wagon with hay.
can infer: The wagon became full.
John loaded hay on the wagon.
cannot infer: The wagon became full.
World Knowledge
As already mentioned, structural semantics does not encompass world knowledge. The following inference relies not only on structural semantics but also on the meaning of the words “man” and “human” and on the world knowledge about the connection between those meanings.
Every human likes watermelons. ⇒ Every man likes watermelons.
Note that the issue here is not that this inference can’t be represented formally. We could formalize it as:
Assumptions: *∀x. [Human(x) → Likes(x,watermelons)]* BackgroundKnowledge Conclusion:∀x. [Man(x) → Likes(x,watermelons)]
This inference is logically valid if BackgroundKnowledge includes wold knowledge about the relation between Man and Human, which can be formalized in FOL as:
∀x. [Man(x) → Human(x)] .
Here is a more elaborate example (which is related to the NLU task Recognizing Textual Entailment):
John ate a cake at a restaurant. ⇒ John probably talked to a waiter before he ate the cake.
Pragmatics
There is also a whole set of linguistic issues that pertain to the way humans use language, beyond literal meanings. These issues are covered under the topic of Pragmatics, the subfield of Linguistics which studies how the literal meaning of language is affected by the context in which it is used.
In the real world, understanding even simple sentences requires being aware of the context. For example, it is highly unlikely that a statement such as “Most students liked the class” claims something about the majority of all students in the world. It probably claims something about most students in a particular group, known in the context – probably the class itself that is mentioned in the sentence. Figuring this out is not part of the “pure” truth conditions of the operator “most”, rather it is a contextual issue.
More complex issues of language use which go beyond the literal meaning include:
- Hyperbole: Exaggeration. E.g., the sentence “This is the worst movie I have ever seen!” might be uttered by a speaker even if, in fact, this is not the worst movie, just one of the worse movies, that the speaker has seen.
- (Verbal) Irony: Using words to convey a meaning that is the opposite of their literal meaning. Example: “Wow, that’s really useful!”, said in a sarcastic tone of voice, conveying “that’s really useless”.
- Metonymy: Referring to something by using the name of something else closely associated with it. E.g. in the sentence “UK imposes harsher lockdown on London, citing new version of virus“, it is not the country itself which imposes a lockdown (only people can actually impose a lockdown). “UK” here is a metonymy for “the government of UK”. And the lockdown is not imposed on the city London itself, but on its residents.
- Metaphor: Describing something by using a seemingly unrelated concept. E.g. the literal meaning of the sentence “Our relationship is not going anywhere” does not make sense because a relationship is an abstract concept and so it cannot move to a different physical location. The intended meaning is that the state of our relationship is not developing (like the speaker thinks it should).
- Speech Act: A NL statement could be used to perform an action, or express the speaker’s intention, in a way that goes beyond the literal information of the statement. Depending on the context, a person uttering “I’m cold” may be implying he wants the situation to change and therefore his utterance conveys a request to another person to close the window.
- Implicature: Additional information that the speaker implies beyond the literal meaning of what is said. For example, the sentence “I may or may not go there” is a tautology as far as truth conditions goes (it’s always true). But since it is usually useless to state tautologies (since they do not add information), the hearer can infer that the intended message here is “I do not know whether I will go there”. Figuring out what the intended meaning is requires identifying adherence to or violation of conversational maxims.
- And there are others…
Go back to: Structural Semantics for Natural Language Understanding.