| Computer-generated news has become commonplace in recent years,affecting the patterns of both news production and news consumption.While the impact of automated journalism has been recognized,few studies have examined automated news content from a linguistic point of view.This thesis aims to fill this gap by investigating linguistic variation in computer-generated news texts using Multi-Dimensional analysis.By comparing two self-compiled corpora of automated news articles and journalist-written articles under Biber’s(1988)Multi-Dimensional model,the present study found distinctive characteristics of computer-generated news:in comparison with journalist-written news,news articles produced by algorithms are generally more informational,less narrative,more context-dependent,less inclined toward persuasion,and more abstract.When investigated under three specific news genres,Financial,Sports,and Political,computer-generated news showed inconsistent extents of stylistic deviations from man-written news across different genres.Lexico-grammatical comparisons further discovered six significantly overused/underused linguistic features in automated news articles,namely Sentence Relatives,General Nouns,General Adverbs,BE as Main Verb,Split Auxiliaries,and Present Participial Clauses.The findings of this study provide a comprehensive linguistic description of computer-generated news as a sub-register distinct from journalist-written news.The discovered linguistic gaps between automated news articles and journalist-written articles also help developers to improve news-generating algorithms in terms of stylistic production. |
PDF Full Text Request