Parsing for a machine-readable form that is semantically equivalent to Turing machine and PAC learnable

ACL ARR 2025 May Submission2189 Authors

Abstract: Developing a parser as reliable as a human being is a key to map natural language (utterance) to a comprehensive logic form. In this paper, we introduce the Enterprise-Participant (EP) data model and propose a semantic parser that maps utterance to an EP database. Because EP, a recursive language, is semantically equivalent to Turing machine, i.e., an EP database is mathematically capable of inventorying all the properties of a partial recursive function with the hypothesis of infinite space and time, we assume and expect the meaning of natural language can be adequately fit (or precisely approximated) into an EP database having finite objects with infinite properties including self-applicable functions. Instead of using a formal grammar, we accumulate parsing rules from sample sentences, i.e., given a randomly selected sentence, we add the corresponding syntactical structure and meaning into an EP database. Because an EP database is PAC learnable, the accumulation process converges, when sample sentences amount to a foreseeable size, to the ultimate machine readable form of the entire natural language. As a side effect, the collection of the parsing rules will be converged to map arbitrary utterance to their syntactical structures as part of the ultimate machine readable form in an EP database.

Paper Type: Long

Research Area: Syntax: Tagging, Chunking and Parsing

Research Area Keywords: Semantic parsing, computability, PAC learnability (Symbolic machine learning), knowledge representation and reasoning

Contribution Types: Theory

Languages Studied: General natural languages, examples in English, as well as system language: the Enterprise-Participant (EP) data model and Froglingo

Submission Number: 2189