Learning Mandarin Through Whole-Word Images and Phrase Melodies
Year: 2026
Type: article
Abstract: Author: Kriger, Boris (Information Physics Institute; Institute of Integrative and Interdisciplinary Research, Toronto); ORCID 0009-0001-0034-2903 Language: English Description: The paper describes and analyzes a method for learning Mandarin Chinese vocabulary that departs from the dominant mnemonic traditions in three ways. Visual unit. The learner memorizes the whole written word of two or three characters as a single picture, not the single character broken into components. Auditory unit. The learner memorizes a short formulaic phrase together with the melody it has in connected speech, not syllables with their dictionary tones. The learner writes the pronunciation in the letters of his own language. Six sound contrasts must be kept distinct in that writing; everything else is written by ear. Pinyin serves only as a reference. Who finds the associations. The learner finds every association, visual or by sound. The teacher, human or AI, gives only directions through a four-level ladder of hints. Associations are found in more than one channel and tied into a short story for each lesson. The supports are then withdrawn through reading and dialogue built only from phrases the learner already knows. The paper proves a set of results about the method’s components: when whole-word encoding is cheaper than decomposition, depending on semantic transparency and on how often characters recur; bounds on retrieval through several associations, whether or not they are independent; that giving hints one level at a time, only after failure, gives the strongest memory when self-generation strengthens memory, and when the first hint level should be skipped; how the cost of tone errors falls as the unit gets longer; that writing two sounds with the same letters can only create homophones, never remove them; the expected number of reviews in the support-fading procedure; a (1−1/e) guarantee for greedy phrase selection by communicative coverage, and the value of repair phrases. Computations on a multi-source Mandarin frequency list (wordfreq 3.1.1) with a context-aware pronunciation dictionary (pypinyin) show: about 78% of the 5,000 most frequent word types are multi-character; with tones ignored, about 91% of frequent monosyllables share their sound with another frequent word, against about 13% of disyllables; the pronounced tone pattern differs from the dictionary tones in at least 13% of disyllabic and 27% of trisyllabic word tokens; for Russian-speaking learners, merging aspirated with unaspirated consonants alone creates new homophones for about 17% of frequent words, and six such merges together for about 31%. The paper gives five complete prompts by which large language models can run the method: tutor, card generator, checker of the learner’s associations and stories, reading and dialogue practice, and phrase selector. It reports an informal voice-mode trial of these prompts and the revisions it led to. It discusses the evidence that keyword-based memories decay quickly without review. It proposes a factorial experiment designed so that every outcome is informative. The code that reproduces all figures and statistics is included in the appendix. Keywords: Mandarin Chinese; vocabulary acquisition; keyword method; generation effect; dual coding; formulaic language; lexical tone; tone sandhi; aspiration; spaced retrieval; scaffolding; large language models; computer-assisted language learning; Russian-speaking learners
Author: Kriger, Boris
Language: English
Subfield: Developmental and Educational Psychology
Field: Psychology
Domain: Social Sciences
Open Access status: green
OpenAlex ID: W7225628931
FWCI:
Cites:
Cited by:
Related to: