這份報告首次詳細介紹了三個大型粵語語料庫中的單詞和子詞頻。此三個 語料庫中的詞頻總體上具有相似的結構,但語料庫之間的成對比較顯示詞 彙重疊較低,單個詞的頻率相關性較弱。相比之下,聲音結構的頻率,包 括片段、音節和聲調,都具有良好的相關性,但由於類型/標記、正字法編 碼、詞的位置和口語文本類型的區別,聲音結構的頻率相關性中仍然出現 重要的差異。這些差異提醒廣東話的心理語言學研究學者,實驗條件標準 必須包括頻率。此外,我們記錄了如何從正文中分割單詞、如何對單詞進 行拼寫和語音編碼、以及從大型語料庫中提取標記和類型頻率的方法,從 而提供對數據的進一步分析。最後,我們通過 預測粵語語音錯誤及單詞識 別來驗證詞頻數據。所有這些發現都在開放數據集中進行了總結。

This report gives the first detailed account of word and sub-lexical frequency in three large corpora of Cantonese. Word frequencies across the corpora have a similar structure overall, but pairwise comparisons between corpora showed low lexical overlap and weak correlations in the frequencies of individual words. By contrast, sound structure frequencies, including segment, syllable, and tone, are well-correlated, but nonetheless exhibit important differences due to the type/token distinction, orthographic encoding, word position, and speech genre. These differences inform psycholinguistic studies of Cantonese that include frequency as an experimental condition. In addition, we document the methods used to segment words from running text, encode words orthographically and phonologically, and extract token and type frequencies from large data sets, thereby providing further access to the data. Finally, we validate the word frequency data by using it as a predictor of speech error and word recognition data in Cantonese. All of these generalizations are summarized in public data sets.

粵語、單詞/子詞頻率、聲音結構頻率、語料庫語言學

Cantonese, word frequencies, phonological frequencies, corpus linguistics