clauseembedding

Clause embedding, sometimes written as clauseembedding, is the syntactic process by which a clause is embedded within another clause, yielding complex sentences with a hierarchical structure. An embedded clause functions as a constituent of a larger clause and can serve as a subject, object, complement, or modifier. Embedding contrasts with coordination, where multiple clauses occupy a similar syntactic level.

Types of embedded clauses include complement clauses (that I believe), interrogative or content clauses (whether/if I can come), relative clauses (the book that I read), and adverbial clauses (when I arrived). In many languages, a complementizer word such as that or whether marks the boundary between the main clause and the embedded clause; some languages use zero complementizers. Embedded clauses enable the expression of propositions, questions, and modifiers within a single sentence.

Embedding interacts with constraints on movement and extraction, including long-distance dependencies and island constraints. The depth of embedding varies across languages, with some languages showing frequent deep embedding and others favoring shallower structures. Recursion is a key feature in many languages but may be limited or parameterized, influencing the frequency and complexity of embedded clauses.

In computational linguistics, clause embedding is also used to describe the representation of embedded clauses as vector or structured representations in NLP models. Clause embeddings are derived from parsers, recursive neural networks, or transformer-based encoders and are used in tasks such as semantic role labeling, machine translation, question answering, and syntactic parsing. Challenges include ambiguity in scope, coreference resolution, and cross-linguistic variation.