This paper proposes a formalization of natural language parsing and generation in Categorial Grammar (CG) as instances of the classical planning problem. The formalization is shown to be useful for analysing the semantics of the language recognized by a CG lexicon in a systematic manner, considering the case of lexicon Lex \(_{R,\sqcap }\) designed for the description logic DL-Lite \(_{R,\sqcap }\) . The approach proposed to solve the parsing and generation planning problems is to use a declarative formalization of heuristics for action selection to guide the search in the state space associated with these planning problems. Examples of declarative formalizations of heuristics that can be used for generating English sentences from DL-Lite \(_{R,\sqcap }\) formulas, and for parsing sentences in the English fragment \(L_{Lex_{R, \sqcap }}\) and mapping them into DL-Lite \(_{R,\sqcap }\) formulas are presented. Situated word learning and word formation scenarios illustrating how the formalism allows exploiting syntactic, semantic, domain and contextual knowledge in an integrated manner are discussed.