Automatic Nominalization of Clauses through Textual Entailment

Proceedings of the 29th International Conference on Computational Linguistics, pages 6002–6006 October 12–17, 2022. 6002 Automatic Nominalization of Clauses through Textual Entailment John S. Y. Lee1, Ho Hung Lim1, Carol Webster2, Anton Melser1 1 Department of Linguistics and Translation, City University of Hong Kong 2 College of Professional and Continuing Education, The Hong Kong Polytechnic University Hong Kong SAR, China {jsylee,hhlim3}@cityu.edu.hk, carol.webster@cpce-polyu.edu.hk, anton.melser@my.cityu.edu.hk Abstract Nominalization re-writes a clause as a noun phrase. It requires the transformation of the head verb of the clause into a deverbal noun, and the verb’s modifiers into nominal modifiers. Past research has focused on the selection of deverbal nouns, but has paid less attention to the word order and word forms for the nominal modifiers. We propose using a textual entail- ment model for clause nominalization. Exper- imental results show that a textual entailment model fine-tuned on this task outperforms a number of unsupervised approaches using lan- guage model scores. 1 Introduction Many textbooks on academic writing devote sig- nificant attention to nominalization, which lends a more abstract, concise and objective tone to a text (Kamler and Thomson, 2006; Bailey, 2011). Since nominalization requires careful lexical selec- tion and clause restructuring, it demands advanced vocabulary knowledge and grammatical skills, mak- ing it challenging even for many human writers. Table 1 outlines the steps in nominalizing a sen- tence: extraction of target clauses; nominalization of a target clause through word re-ordering and re-generating the POS and prepositions; and the re-writing of the sentence to incorporate the nomi- nalized clauses. This paper focuses on the clause nominalization step. Past research on clause nominalization has con- centrated on replacement of the head verb with a deverbal noun (e.g. ‘omits’ → ‘omission’) and resource development to support the task (Mey- ers et al., 1998; Habash and Dorr, 2003; Saberi and Lee, 2019). Less attention has been paid to the clause restructuring that is required for trans- forming the verb’s modifiers in the clause to nom- inal arguments, including word reordering (e.g., postposing the subject ‘Arabic’, Table 1i); POS re- generation of the verb’s modifiers (‘frequently’ → Extract Since [Arabic frequently omits clause vowels], the surname is often spelt ... Nomin- (i) Word-order edits: alize [frequently vowels omits Arabic] clause (ii) POS Re-generation: [frequent vowels omission Arabic] (iii) Preposition generation: [frequent vowel omissions in Arabic] Re-write Due to [frequent vowel omissions sentence in Arabic], the surname is often spelt ... Table 1: Main steps in nominalizing a sentence, focusing on the target clause enclosed in [...]. ‘frequent’, Table 1ii); and preposition generation (Table 1iii). This paper investigates methods for clause nomi- nalization that optimize the position and POS of the nominal arguments. We focus on clauses headed by a verb with up to three syntactic arguments, includ- ing subjects, objects, adjectival phrases or preposi- tional phrases. The output is a nominalized form of the clause that preserves its original meaning. Experimental results show that a textual entailment model that is fine-tuned on this task can outperform unsupervised approaches based on …

Content truncated to 500 words