Prosodic Structure in Chinese Characters, Revisited

Home , Small seal script

Prosodic structure Thanks! Overview • Grants from Taiwan’s National Science Council / • Character “prosody” Ministry of Science and Technology: • in Chinese characters, • NSC 97‐2410‐H‐194‐067‐MY3, NSC101‐2410‐H‐194‐115‐ Radical position and reduction MY3, MOST 103‐2410‐H‐194‐119‐MY3 • Reduplication revisited • Many lab assistants over the years: • Stroke shape and position • Ko Yuguang, Yang Chentsung, Hsu Chiung‐Wen, Hong • Guo‐Ming, Hsu Zi‐Ping, Du Pei‐Fen, Su Yu‐Ting, Liu Yu‐ Lengthened strokes James Myers Jay, Chuang Wei‐Chiang, Hsieh Yu‐Yi, Pan Hsiao‐Yin, Su • Curved strokes National Chung Cheng University Si‐Qi • Hooked strokes [email protected] • Plus: • Chen Tsung‐Ying, Zev Handel, Wolfgang Behr, Jane Tsay, Niina Zhang, and Lin Yu‐Hsuan

2 3 ILAS Workshop on Phonetics and Phonology – October 2017

Overview Chinese characters are productive Chinese character grammar • Duality of patterning • Character “prosody” Exponential growth • Radical position and reduction Productivity “Morphology” (model fit > 80%) • • Reduplication implies Recursion (Sproat, 2000) • Stroke shape and position grammar…. 鱗 • Lengthened strokes • • Curved strokes Shape regularities (Wang, 1983) • Hooked strokes 牛 → 特 cf. 牢 “Phonology” • “Prosody”: global shape constraints • Myers (1996, 2016) Data from Yip (2000, p. 19) 4 5 6

Character prosody Overview Radicals (部首) as affixes • Chinese characters tend to show: • Character “prosody” • Morphological properties • Binarity • Closed class • Asymmetrical edge prominence (right and bottom) • Radical position and reduction • Bound • These have wide‐ranging effects on components & strokes • Reduplication • Semantically bleached • Similar to metrical feet in spoken & signed languages • Stroke shape and position • Phonological properties • Two‐dimensional “feet”: • Lengthened strokes • Favor prosodically “weak” positions • Curved strokes 1 2 • Suppletive allomorphs • Hooked strokes • General reduction rules 1

7 8 9 Most characters have left/top radicals Most radicals favor the left or top The left/bottom alternation

9000 8000 • Radicals that most favor the left also sometimes appear on the bottom (see previous slide) 7000 6000 詞～警 Informal writing also favors 5000 left‐side radicals (Wang, 1983): • A prosodic explanation 4000 Left‐sided radicals are “light” horizontally but “heavy” 3000 vertically: Distinct characters 1 2000 1 2 2

1000 1 1 0 Left Right Top Bottom Other 2 2

Computed from 13,060 traditional characters in Tsai (2006) 10 11 12

Radicals may reduce in “weak” position Grammaticalization of radicals Frequency affects grammaticalization • Some radicals have suppletive forms at left/top • Vietnamese (Chữ Nôm) radicals acted more like roots 心～忙人～位水～泊手～拾 • Often “synonymic”, not “taxonymic” (Handel, 2016; 艸～花竹～筆 cf. ancient logographic systems: Rude, 1986) Frequency decreases reduction • But not at right or bottom (generally) • Also often lacked formal reduction: 𢬣 tay ‘arm’ in “strong” positions, increases reduction in “weak” positions 忙～忘泊～泉拾～拿狗～狀 (cf. 分～刻燙～熱) Traditional characters go here • PRC simplification: “Weak” reduction is extended • Other radicals reduce in a more regular way 词～警铅～鉴红～累 • Wang (1983): Diagonalization and stroke reduction 金：鉛～鑒土：地～型牛：物～牽 (cf. 詞～警鉛～鑒紅～累) 木：村～果米：精～粱衣：被～裝 (using Tsai, 2006, database) • May even apply to non‐radicals: 且：助～宜 (Mixed‐effects logistic regression by radical: Left/top vs. right/bottom : p < .0001; 13 14 interaction between frequency and position only marginal: p = .06) 15

Testing radical position productivity So let’s run a new experiment! Priming to boost position effect? • Chinese readers were shown fake characters and asked to judge if they were like characters (「像中文字嗎？」) • Lexicality: Component is/isn’t a “real” radical Horizontal (left/right) Vertical (top/bottom) 1 1 • Grammaticality: Is/isn’t in “weak” position (left/top) 0.9 0.9 0.8 Prefer radicals at left, 0.8 Likewise for radicals at 0.7 even fake ones 0.7 top, but only real ones 0.6 0.6 0.5 0.5 0.4 0.4 0.3 0.3 Acceptance rate 0.2 Acceptance rate 0.2 0.1 0.1 0 0 Grammatical Ungrammatical Grammatical Ungrammatical Lexical radical Nonlexical radical Lexical radical Nonlexical radical Created by Chen Tsung‐Ying (Chen & Myers, 2017); Mixed‐effects logistic regression on all items: Grammaticality (p < .0001); Lexicality (p < .05); • Problem: Position is confounded with reduction 16 this experiment was run with him too 17 Grammaticality x Orientation (p < .0001); Grammaticality x Lexicality (p < .0001) 18 Interaction with horizontal reduction Overview Reduplication patterns • Binary horizontal reduplication 林比兢朋弱嚇雙選窳替質瑩 • Character “prosody” • Binary vertical reduplication Unreduced Reduced • Radical position and reduction 1 1 昌吕圭戔多炎哥棗僵渁 0.9 Prefer radicals at left, 0.9 But grammaticality • Reduplication • Triangular reduplication (binary both ways) 0.8 even fake ones, even 0.8 effect is stronger for 0.7 0.7 if not reduced reduced ones • Stroke shape and position 品惢鑫蟲晶畾森聶众磊轟犇 0.6 0.6 • 0.5 0.5 Lengthened strokes • Square reduplication is quite rare: 叕 0.4 0.4 • Curved strokes 0.3 0.3 • Not productive as a full template (splittable): 爽爾齒繼器 Acceptance rate 0.2 Acceptance rate 0.2 • Hooked strokes 0.1 0.1 • Exceptions are quite rare and restricted: 0 0 Grammatical Ungrammatical Grammatical Ungrammatical Non‐binarity (mostly single strokes): 三州巡黑靈 Lexical radical Nonlexical radical Lexical radical Nonlexical radical Inverted triangles (not a single template): 熒 Mixed‐effects logistic regression on all items: Grammaticality (p < .0001); Lexicality (p < .05); Splitting of horizontal reduplication (we’ll come back to this): 變 Grammaticality x Lexicality (p < .05); Grammaticality x Reduction (p < .05) 19 20 21

Deriving reduplication Testing reduplication productivity Grammar independent of lexicon • Productive reduplication requires abstract rules or templates (Berent et al., 2014) • Myers, J. (2016). Knowing Chinese character Significant • So 林 is 木 + [XX], not 木+木 grammar. Cognition, 147, 127‐132. • Abstract reduplication in the PRC: 雙:双聶:聂轟:轰

• Regular reduction still applies: No interaction 根林兟甡玨 • Triangles are derived from binarity + symmetry: 1 2

1 Significant 2

22 23 24

Grammar & lexicon at same time Overview Enlargement in strong positions

• Character “prosody” • Lowest component is enlarged, even if it’s not free • Radical position and reduction 昌吕圭戔炎串官二夫土未 • Reduplication • Similarly for rightmost component • Stroke shape and position 林川州井 • Lengthened strokes • Exceptional top prominence triggers adjacent clash • Curved strokes Clash: 士末畢 No clash: 聿重垂事 • Hooked strokes Normal Exceptional + clash Exceptional + no clash

Survival analysis (Baayen & Blanche, 2017): Plots the evolving probability of acceptance vs. rejection (vertical bar marks when probability rises above chance) 25 26 27 The variety of vertical strokes Distribution of plain strokes Distribution of curved strokes • Left: 同甬行高商岡門雨竹 • Left: 非拜介廾川州卅爪瓜并升飛片爿辡月丹舟周角用虎戶底大右力及看爿片 • Middle: 十中木東本不川卅爪丫羊半下雨忄(忙)亻(仁)牜(牠) 串甲申車革幸聿年韋市午平丰 • Middle: ... (does 班 count?) • Vertical stroke shape (plain, curved, hooked) is never (?) lexically distinctive by itself • Right: 非拜介廾川州卅爿 • Right: ... • Yet the contrast is apparently important enough to appear in many fonts:

十介丁十介丁十介丁

28 29 30

Distribution of hooked strokes Some surface generalizations Curves: A deeper generalization • Plain stroke is default • Left: ... • Curves appear in “weak” position at left: • Also allows bottom contact: 口廿上並 • Curved stroke is also shorter than rightmost: 川 • Curved stroke appears on left, • • Middle: 丁亍予亇乎于小水手事爭扌(把)朩 “Left” is defined within a “prosodic” domain: especially in “tall/thin” components (Wang, 1983): 周 vs.同 • Separate components vs. wholes: 門周月用周角 • Hooked stroke appears on right with material above it: • Thin/tall = one prosodic domain: • Right: 寸才可竹刂(到) • Fat/squat = two prosodic domains: 同甬高商可周同 (㇆)月周同角高商雨門句弓 • Examples conforming to this pattern: e.g. : fat/squat free 肉 vs. its tall/thin radical allomorph: 月 • Hook on right but no material above it: 寸才扌(<手)刂(<刀) • A prediction (thanks to Lin Yu‐Hsuan): • Hook with material above it but not on right: 丁爭 • Thin/tall components favor left/right radicals more than • Hook without material above and not on right: 手小水事 fat/squat components, which also like top/bottom radicals • A near minimal pair consistent with this prediction: 角: 唃埆捔桷确斛(0% top‐ or bottom‐side radicals) 甬: 俑埇蛹誦踊悀捅桶涌勇恿筩(25% top/bottom) 31 32 33

Testing curving and radical position Testing the acceptability of lengthening and curving Lengthening & curving results 180 16% 160 • Another Worldlikeness experiment Lengthening Curving Straight‐side components 0.7 0.7 140 take more top/bottom (with Chen Tsung‐Ying) Strong > weak: p < .001 Weak > strong: ns 120 radicals: squatter! • Testing 380 combinations of stroke size, shape, position, 0.6 0.6 100 0.5 0.5 6% and orientation:「有中文字的感覺嗎？」 80 Top/bottom radical 0.4 0.4 60 Left/right radical 0.3 0.3 Acceptability Acceptability

Number of characters 40 0.2 0.2 20 0.1 0 0.1 Curved Straight 0 0 Left stroke Horizontal (三)Vertical (川) Leftmost Other Fisher’s test, p < .05, testing 丹月用有舟角周風 vs. 巾內冉冊冋市同向朿肉冏甬兩岡高商喬 Strong position Weak position as non‐radicals in 227 characters with only one level of embedding (mixed‐effects logistic regression confirms that this pattern not due to base component type frequency) 34 35 (Mixed‐effects linear regression on arcsine‐transformed Likert‐scale judgments) 36 Hooking is more lexicalized Hooks: Deeper generalizations? Testing the acceptability of hooking • Hooks tend to be in “strong” positions: • Yet another Worldlikeness experiment • No simple generalization applies to all cases • At right (月寸) or alone in the middle (丁) (again with Chen Tsung‐Ying) • Near minimal pairs: 聿事平乎 • Hooks also tend to have material above: • Testing 24 combinations of shape and position: • Not predictable from small seal script (小篆): • Usually contact at top: 可周同丁爭手乎 (cf. 寸才小水事) 聿十 • Stroke coordination as gesture coordination? • Syllables are the domain of gesture coordination 事丁 • Complex rimes are heavier (more gestures to coordinate) • Complex onsets may also be heavier (see Hsieh, yesterday) 平可乎寸 37 38 39

Hooking judgment results More on lexicalization Summary and conclusions • Left component in horizontal reduplication shows regular 0.7 reduction or curving (single “foot”): 玨林艸辡 • The prosodic hypothesis accommodates many facts: 0.6 • But a lexical subset are still splittable: 班辦 Radicals, reduplication, strokes all show: Right > left: p < .0001 Topped > untopped: Left is weaker than right, top is weaker than bottom, 0.5 • Reduplication also doesn’t trigger suppletive reduction: in binary groupings p < .0001 從搻惢(*亻*扌 *忄) 0.4 • Topped cf. 獄 (reanalysis as affixation) 朒 (different elements) Why does it work? 0.3 Untopped • Universal Grammar applies to any sufficiently large and

Acceptability • In spoken phonology this would imply ordering/opacity: complex communication system (e.g., also sign language)? 0.2 [XX] → Suppletion → Redup → Reduc/Curv → Infixation • Extra‐linguistic constraints on motor control, perception, 0.1 [XX] → [扌巴] →…→ … →…→[把] memory, and cognition? 0 [XX] → … → [木木] → [林] →…→[林] Left side Right side [XX] → … → 王王 → 玨 → 班 [ ]split [ ]split [ ] B 8 (Mixed‐effects linear regression on arcsine‐transformed Likert‐scale judgments) 40 [XX] → … → [辛辛]split → [辡]split → [辦] 41 42

References (1/2) References (2/2)

Baayen, H., & Blanche, P. (2017). Dynamic survival analysis. Talk Myers, J. (2016). Knowing Chinese character grammar. Cognition, 147, presented at the Workshop on New Methods in Statistics, Tübingen, 127‐132. Germany, January 19‐20. Rude, N. (1986). Graphemic classifiers in Egyptian hieroglyphics and Berent, I., Dupuis, A., & Brentari, D. (2014). Phonological reduplication in Mesopotamian cuneiform. In C. Craig (Ed.) Noun classes and sign language: Rules rule. Frontiers in Psychology, 5, Article 560, 1‐15. categorization (pp. 133‐138). John Benjamins. Chen, T.‐Y., & Myers, J. (2017). Worldlikeness: A Web‐based tool for Sproat, R. (2000). A computational theory of writing systems. Cambridge, typological psycholinguistic research. University of Pennsylvania UK: Cambridge University Press. Working Papers in Linguistics, 23(1), Article 4, 20‐30. Tsai, C.‐H. (2006). Frequency and stroke counts of Chinese characters. Handel, Z. (2016). Sinography: A cross‐linguistic study of the borrowing http://technology.chtsai.org/charfreq/. and adaptation of the Chinese script. University of Washington ms. Wang, J. C.‐S. (1983). Toward a generative grammar of Chinese Myers, J. (1996, June). Prosodic structure in Chinese characters. Poster Character structure and stroke order. University of Wisconsin‐Madison, presented at the Fifth International Conference on Chinese Linguistics, Ph.D. dissertation. National Tsing Hua University, Taiwan. Yip, P. C. (2000). The Chinese lexicon: A comprehensive survey. Routledge.

43 44 45