Upgrade to Pro — share decks privately, control downloads, hide ads and more …

Network Approaches to Old Chinese Reconstruction

Network Approaches to Old Chinese Reconstruction

Talk, held at the 30th conference of the Centre des Recherches Linguistiques sur l'Asie Orientale

Johann-Mattis List

June 30, 2017
Tweet

More Decks by Johann-Mattis List

Other Decks in Science

Transcript

  1. Network Approaches to Old Chinese Reconstruction
    Johann-Mattis List
    Department of Linguistic and Cultural Evolution
    Max Planck Institute for the Science of Human History
    Jena
    2017/06/30
    1 / 35

    View Slide

  2. Introduction
    Introduction
    2 / 35

    View Slide

  3. Introduction Old Chinese Reconstruction
    Old Chinese Reconstruction: Overview
    long tradition of linguistic reconstruction in China (starting with Chén
    Dì 陳第, 1541 – 1606),
    3 / 35

    View Slide

  4. Introduction Old Chinese Reconstruction
    Old Chinese Reconstruction: Overview
    long tradition of linguistic reconstruction in China (starting with Chén
    Dì 陳第, 1541 – 1606),
    breakthrough in the early 20th century with Karlgren’s
    reconstructions and impressive work by Wáng Lí 王力 (1980) and Li
    Fang-kuei 李方桂 (1971),
    3 / 35

    View Slide

  5. Introduction Old Chinese Reconstruction
    Old Chinese Reconstruction: Overview
    long tradition of linguistic reconstruction in China (starting with Chén
    Dì 陳第, 1541 – 1606),
    breakthrough in the early 20th century with Karlgren’s
    reconstructions and impressive work by Wáng Lí 王力 (1980) and Li
    Fang-kuei 李方桂 (1971),
    since then more and more improved concrete reconstructions of Old
    Chinese phonology,
    3 / 35

    View Slide

  6. Introduction Old Chinese Reconstruction
    Old Chinese Reconstruction: Overview
    long tradition of linguistic reconstruction in China (starting with Chén
    Dì 陳第, 1541 – 1606),
    breakthrough in the early 20th century with Karlgren’s
    reconstructions and impressive work by Wáng Lí 王力 (1980) and Li
    Fang-kuei 李方桂 (1971),
    since then more and more improved concrete reconstructions of Old
    Chinese phonology,
    another breakthrough in the 1980s, when Baxter (1992), Starostin
    (1989), and Zhèngzhāng Shàngfāng (see Zhèngzhāng 2003)
    presented reconstructions in which they independently proposed
    several similar features (notably six vowels)
    3 / 35

    View Slide

  7. Introduction Old Chinese Reconstruction
    Old Chinese Reconstruction: Overview
    long tradition of linguistic reconstruction in China (starting with Chén
    Dì 陳第, 1541 – 1606),
    breakthrough in the early 20th century with Karlgren’s
    reconstructions and impressive work by Wáng Lí 王力 (1980) and Li
    Fang-kuei 李方桂 (1971),
    since then more and more improved concrete reconstructions of Old
    Chinese phonology,
    another breakthrough in the 1980s, when Baxter (1992), Starostin
    (1989), and Zhèngzhāng Shàngfāng (see Zhèngzhāng 2003)
    presented reconstructions in which they independently proposed
    several similar features (notably six vowels)
    current state-of-the-art is the reconstruction by Baxter & Sagart
    (2014)
    3 / 35

    View Slide

  8. Introduction Old Chinese Reconstruction
    Old Chinese Reconstruction: Problems
    data and judgments are often not made publicly available
    4 / 35

    View Slide

  9. Introduction Old Chinese Reconstruction
    Old Chinese Reconstruction: Problems
    data and judgments are often not made publicly available
    difficult to compare different judgments without additional data
    4 / 35

    View Slide

  10. Introduction Old Chinese Reconstruction
    Old Chinese Reconstruction: Problems
    data and judgments are often not made publicly available
    difficult to compare different judgments without additional data
    no large-scale comparisons of different types of evidence and how
    scholars interpret
    4 / 35

    View Slide

  11. Introduction Old Chinese Reconstruction
    Old Chinese Reconstruction: Problems
    data and judgments are often not made publicly available
    difficult to compare different judgments without additional data
    no large-scale comparisons of different types of evidence and how
    scholars interpret
    although the research is based on very large amounts of data, those
    data are usually not accessible in a digital format, which makes it
    difficult for scholars to evaluate and investigate a given
    reconstruction system, not to speak of comparing different ones
    4 / 35

    View Slide

  12. Introduction Old Chinese Reconstruction
    Old Chinese Reconstruction: Types of Evidence
    rhymes in ancient poems
    5 / 35

    View Slide

  13. Introduction Old Chinese Reconstruction
    Old Chinese Reconstruction: Types of Evidence
    rhymes in ancient poems
    structure of ancient characters
    5 / 35

    View Slide

  14. Introduction Old Chinese Reconstruction
    Old Chinese Reconstruction: Types of Evidence
    rhymes in ancient poems
    structure of ancient characters
    reconstruction of Middle Chinese via fǎnqiè readings, rhyme books,
    and ancient character descriptions (dúruò etc.)
    5 / 35

    View Slide

  15. Introduction Old Chinese Reconstruction
    Old Chinese Reconstruction: Types of Evidence
    rhymes in ancient poems
    structure of ancient characters
    reconstruction of Middle Chinese via fǎnqiè readings, rhyme books,
    and ancient character descriptions (dúruò etc.)
    Sino-Xenic readings
    5 / 35

    View Slide

  16. Introduction Old Chinese Reconstruction
    Old Chinese Reconstruction: Types of Evidence
    rhymes in ancient poems
    structure of ancient characters
    reconstruction of Middle Chinese via fǎnqiè readings, rhyme books,
    and ancient character descriptions (dúruò etc.)
    Sino-Xenic readings
    ...
    5 / 35

    View Slide

  17. Introduction Old Chinese Reconstruction
    Old Chinese Reconstruction: Types of Evidence
    rhymes in ancient poems
    structure of ancient characters
    reconstruction of Middle Chinese via fǎnqiè readings, rhyme books,
    and ancient character descriptions (dúruò etc.)
    Sino-Xenic readings
    ...
    Interestingly, many of the different types of evidence have been
    investigated with help of rudimentary network approaches by classical
    Chinese scholars in the past.
    5 / 35

    View Slide

  18. Introduction Networks
    Networks: Overview
    6 / 35

    View Slide

  19. Introduction Networks
    Networks: Overview
    NODE (VERTEX)
    represents an
    object
    6 / 35

    View Slide

  20. Introduction Networks
    Networks: Overview
    NODE (VERTEX)
    represents an
    object
    EDGE (LINK)
    represents a
    relation between
    objects
    6 / 35

    View Slide

  21. Introduction Networks
    Networks: Overview
    can be
    tagged or
    labelled
    EDGE (LINK)
    represents a
    relation between
    objects
    6 / 35

    View Slide

  22. Introduction Networks
    Networks: Overview
    can be
    tagged or
    labelled
    can be
    labelled and
    weighted
    6 / 35

    View Slide

  23. Introduction Networks
    Networks: Examples
    Many structures in daily life and science can be modeled as
    networks:
    social networks: nodes are persons, edges are relations between
    persons (e.g., friendship on FaceBook, etc.),
    7 / 35

    View Slide

  24. Introduction Networks
    Networks: Examples
    Many structures in daily life and science can be modeled as
    networks:
    social networks: nodes are persons, edges are relations between
    persons (e.g., friendship on FaceBook, etc.),
    phylogenetic networks: nodes are languages or dialect varieties,
    edges represent genetic closeness,
    7 / 35

    View Slide

  25. Introduction Networks
    Networks: Examples
    Many structures in daily life and science can be modeled as
    networks:
    social networks: nodes are persons, edges are relations between
    persons (e.g., friendship on FaceBook, etc.),
    phylogenetic networks: nodes are languages or dialect varieties,
    edges represent genetic closeness,
    network of sound change patterns: nodes are sounds, directed
    edges represent likelihood of sound change during language
    evolution,
    7 / 35

    View Slide

  26. Introduction Networks
    Networks: Examples
    Many structures in daily life and science can be modeled as
    networks:
    social networks: nodes are persons, edges are relations between
    persons (e.g., friendship on FaceBook, etc.),
    phylogenetic networks: nodes are languages or dialect varieties,
    edges represent genetic closeness,
    network of sound change patterns: nodes are sounds, directed
    edges represent likelihood of sound change during language
    evolution,
    ...
    7 / 35

    View Slide

  27. Introduction Networks
    Networks: Approaches
    identify clusters (communities) of nodes in the network
    8 / 35

    View Slide

  28. Introduction Networks
    Networks: Approaches
    identify clusters (communities) of nodes in the network
    check externally proposed groupings against grouping emerging
    from the network itself
    8 / 35

    View Slide

  29. Introduction Networks
    Networks: Approaches
    identify clusters (communities) of nodes in the network
    check externally proposed groupings against grouping emerging
    from the network itself
    investigate the general dynamics of the network
    8 / 35

    View Slide

  30. Introduction Networks
    Networks: Approaches
    9 / 35

    View Slide

  31. Introduction Networks
    Networks: Approaches
    9 / 35

    View Slide

  32. Introduction Networks
    Networks: Approaches
    9 / 35

    View Slide

  33. Introduction Networks
    Networks: Approaches
    9 / 35

    View Slide

  34. Introduction Data
    Data: Overview
    data should be easily accessible
    10 / 35

    View Slide

  35. Introduction Data
    Data: Overview
    data should be easily accessible
    data should be transparent (easy to comprehend for experts)
    10 / 35

    View Slide

  36. Introduction Data
    Data: Overview
    data should be easily accessible
    data should be transparent (easy to comprehend for experts)
    data should be human- and machine-readable (to allow for
    error-checking and automatic analysis as well as qualitative analysis
    10 / 35

    View Slide

  37. Introduction Data
    Data: Overview
    data should be easily accessible
    data should be transparent (easy to comprehend for experts)
    data should be human- and machine-readable (to allow for
    error-checking and automatic analysis as well as qualitative analysis
    data should be shared immediately with publications if publications
    rely on the data
    10 / 35

    View Slide

  38. Introduction Data
    Data: Increasing Comparability
    simple formats should be preferred over complicated ones
    11 / 35

    View Slide

  39. Introduction Data
    Data: Increasing Comparability
    simple formats should be preferred over complicated ones
    tabular data can be edited in Excel but should be shared in form of
    CSV files
    11 / 35

    View Slide

  40. Introduction Data
    Data: Increasing Comparability
    simple formats should be preferred over complicated ones
    tabular data can be edited in Excel but should be shared in form of
    CSV files
    if data structures cannot be immediately understood, authors should
    add metadata to describe them
    11 / 35

    View Slide

  41. Introduction Data
    Data: Increasing Comparability
    simple formats should be preferred over complicated ones
    tabular data can be edited in Excel but should be shared in form of
    CSV files
    if data structures cannot be immediately understood, authors should
    add metadata to describe them
    data can be easily hosted on scientific repositories like Zenodo
    (zenodo.org), and curation is simple with help of services such as
    GitHub or GitBucket
    11 / 35

    View Slide

  42. Introduction Data
    Data: The CHIP Initiative (with N. W. Hill)
    try to digitize rhyme judgments of different authors (like Baxter’s
    1992 Shījīng readings, Wáng’s 1980 Shījīng readings, etc.)
    12 / 35

    View Slide

  43. Introduction Data
    Data: The CHIP Initiative (with N. W. Hill)
    try to digitize rhyme judgments of different authors (like Baxter’s
    1992 Shījīng readings, Wáng’s 1980 Shījīng readings, etc.)
    make the data comparable, both for humans and machines
    12 / 35

    View Slide

  44. Introduction Data
    Data: The CHIP Initiative (with N. W. Hill)
    try to digitize rhyme judgments of different authors (like Baxter’s
    1992 Shījīng readings, Wáng’s 1980 Shījīng readings, etc.)
    make the data comparable, both for humans and machines
    make the data freely available on the web and on Zenodo
    12 / 35

    View Slide

  45. Introduction Data
    Data: The CHIP Initiative (with N. W. Hill)
    try to digitize rhyme judgments of different authors (like Baxter’s
    1992 Shījīng readings, Wáng’s 1980 Shījīng readings, etc.)
    make the data comparable, both for humans and machines
    make the data freely available on the web and on Zenodo
    Current state-of-the-art: http://github.com/digling/rhymes.
    12 / 35

    View Slide

  46. Rhyme Networks
    Rhyme Networks
    13 / 35

    View Slide

  47. Rhyme Networks Modeling
    Modeling
    14 / 35

    View Slide

  48. Rhyme Networks Modeling
    Modeling
    14 / 35

    View Slide

  49. Rhyme Networks Modeling
    Modeling
    14 / 35

    View Slide

  50. Rhyme Networks Modeling
    Modeling
    14 / 35

    View Slide

  51. Rhyme Networks Modeling
    Modeling
    14 / 35

    View Slide

  52. Rhyme Networks Modeling
    Modeling
    14 / 35

    View Slide

  53. Rhyme Networks Modeling
    Modeling
    27.3.A
    30.2.A
    33.3.A
    39.1.A
    54.4.B
    58.1.A
    58.6.B
    59.1.A
    66.1.A
    130.1.A
    204.4.A
    227.2.A
    sī 丝
    qī 淇 móu 谋
    qī 淇
    qī 淇
    sī 丝
    sī 丝
    sī 丝
    sī 丝
    sī 丝
    sī 丝
    sī 丝
    sī 丝
    sī 丝
    sī 丝
    sī 丝
    sī 丝
    sī 丝
    sī 丝
    sī 丝
    sī 丝
    sī 丝
    sī 丝
    sī 丝
    sī 丝
    sī 丝
    sī 丝
    sī 丝
    zhī 之
    zhī 之
    qī 期
    qī 期
    méi 梅
    méi 梅
    yóu 尤
    yóu 尤
    lái 来 sī 思
    lái 来
    lái 来
    sī 思
    sī 思
    sī 思
    sī 思
    sī 思
    sī 思
    sī 丝
    sī 丝 sī 丝
    sī 丝
    sī 丝
    sī 丝
    sī 丝
    sī 丝
    sī 丝
    sī 丝 sī 丝
    sī 丝 sī 丝
    sī 丝 sī 丝
    sī 丝
    sī 丝
    sī 丝
    sī 丝
    sī 丝
    sī 丝
    sī 丝
    sī 丝
    sī 丝
    sī 丝
    sī 丝
    sī 丝
    sī 丝
    sī 丝
    móu 谋
    zāi 哉
    zāi 哉
    zāi 哉
    zāi 哉
    14 / 35

    View Slide

  54. Rhyme Networks Modeling
    Modeling












    14 / 35

    View Slide

  55. Rhyme Networks Modeling
    Modeling
    Poem Stanza Verse Sect. Text Rhyme Pattern MCH OCBS
    4 1 1 1 南有樛木、 木 - muwk C.mˤok
    4 1 1 2 葛藟纍之。 纍 A lwij [r]uj
    4 1 2 1 樂只君子、 子 - tsiX tsəʔ
    4 1 2 2 福履綏之。 綏 A swij s.nuj
    4. 樛木
    南有樛木、葛藟纍之。
    樂只君子、福履綏之。
    南有樛木、葛藟荒之。
    樂只君子、福履將之。
    南有樛木、葛藟縈之。
    樂只君子、福履成之。
    15 / 35

    View Slide

  56. Rhyme Networks Modeling
    Example: The Shījīng Rhyme Browser
    interactive web-based application
    displays Shījīng rhymes in digitized form with rhyme annotations
    following Baxter (1992) and rhyme readings following Baxter and
    Sagart (2014) and Pān (2000, as provided in the Thesaurus
    Linguae Sericae).
    offers a quick and transparent way to inspect Baxter’s rhyme
    annotations, as well as a quick way to search through the Shījīng for
    rhyme patterns and brief glosses.
    URL: http://digling.org/shijing
    16 / 35

    View Slide

  57. Rhyme Networks Testing
    Testing: Vowel Purity (List et al. in press)
    Ho (2016) claims that the principle of vowel purity was important in Old
    Chinese rhyming:
    poets would try to avoid rhyming words with different consonants,
    17 / 35

    View Slide

  58. Rhyme Networks Testing
    Testing: Vowel Purity (List et al. in press)
    Ho (2016) claims that the principle of vowel purity was important in Old
    Chinese rhyming:
    poets would try to avoid rhyming words with different consonants,
    while differences in the codas were more often tolerated
    17 / 35

    View Slide

  59. Rhyme Networks Testing
    Testing: Vowel Purity (List et al. in press)
    Ho (2016) claims that the principle of vowel purity was important in Old
    Chinese rhyming:
    poets would try to avoid rhyming words with different consonants,
    while differences in the codas were more often tolerated
    reconstruction systems which contradict this principle, may
    therefore be externally criticized as neglecting the principle of vowel
    purity
    17 / 35

    View Slide

  60. Rhyme Networks Testing
    Testing: Vowel Purity (List et al. in press)
    Ho (2016) claims that the principle of vowel purity was important in Old
    Chinese rhyming:
    poets would try to avoid rhyming words with different consonants,
    while differences in the codas were more often tolerated
    reconstruction systems which contradict this principle, may
    therefore be externally criticized as neglecting the principle of vowel
    purity
    On the other hand, we can compare different reconstruction
    regarding the degree of purity of their vowels compared to the
    rhyme data in the Shījīng.
    17 / 35

    View Slide

  61. Rhyme Networks Testing
    Testing: Comparative Data
    Reconstruction System No. Rhymes Density a ɑ æ e ə o ɔ u ʊ ɯ i
    Karlgren (1957) 1830 0.0031 0.0026 x x x x x x x x x x x
    Li 李方桂 (1971) 1830 0.0031 0.0026 x x x x
    Wáng 王力 (1980) 1830 0.0031 0.0026 x x x x x
    Zhèngzhāng 鄭張尚芳 (2003) 1830 0.0031 0.0030 x x x x x x
    Starostin (1989) 1358 0.0035 0.0026 x x x x x x
    Pān 潘悟雲 (2000) 1830 0.0031 0.0026 x x x x x
    Baxter and Sagart (2014) 1431 0.0038 0.0033 x x x x x x
    Schuessler (2007) 1224 0.0041 0.0035 x x x x x x
    18 / 35

    View Slide

  62. Rhyme Networks Testing
    Vowel Purity: Methods
    Assortativity tests whether nodes sharing connections in a
    graph are also similar regarding other characteristics (New-
    man 2003). In social network analyses it can, for exam-
    ple, be used to test whether observed patterns in a network,
    like friendship, come along with properties of the individuals,
    such as language or gender (ibid.). Assortativity can be mea-
    sured by calculating the assortativity coefficient of a network
    in which all nodes have a given attribute.
    19 / 35

    View Slide

  63. Rhyme Networks Testing
    Testing Vowel Purity: Methods
    5
    2
    1
    4
    6
    3
    A
    B
    1
    2
    4
    3
    6
    5
    20 / 35

    View Slide

  64. Rhyme Networks Testing
    Testing Vowel Purity: Methods
    A 1 2 3 4 5 6 B 1 2 3 4 5 6
    1 x x 1 x x
    2 x x 2 x x
    3 x x x 3 x x x x
    4 x x x 4 x x x
    5 x x 5 x x x
    6 x x 6 x x x x
    20 / 35

    View Slide

  65. Rhyme Networks Testing
    Testing Vowel Purity: Methods
    A red blue red + blue B red blue red + blue
    red 6/14 = 0.43 1/14/ = 0.07 7/14 = 0.5 red 6/18 = 0.33 4/18/ = 0.22 10/18 = 0.55
    blue 1/14 = 0.07 6/14 = 0.43 7/14 = 0.5 blue 4/18 = 0.33 4/18 = 0.22 8/18 = 0.44
    red + blue 7/14 = 0.5 7/14 = 0.5 14/14 = 1.0 red + blue 10/18 = 0.55 8/18 = 0.44 18/18 = 1.0
    20 / 35

    View Slide

  66. Rhyme Networks Testing
    Testing: Results
    Reconstruction System Assortativity Randomized
    Assortativity (Ø)
    Standard
    Deviation
    Sigma
    Score
    Rank p-
    Value
    Karlgren (1957) 0.5824 -0.0029 0.0091 64 6 < 0.001
    Li 李方桂 (1971) 0.8230 -0.0026 0.0149 56 8 < 0.001
    Wáng 王力 (1980) 0.7709 -0.0026 0.0127 61 7 < 0.001
    Zhèngzhāng 鄭張尚芳 (2003) 0.7435 -0.0021 0.0103 72 3 < 0.001
    Starostin (1989) 0.8444 -0.0025 0.0115 74 2 < 0.001
    Pān 潘悟雲 (2000) 0.7326 -0.0020 0.0103 71 4 < 0.001
    Baxter and Sagart (2014) 0.8765 -0.0025 0.0112 79 1 < 0.001
    Schuessler (2007) 0.7244 -0.0026 0.0111 66 5 < 0.001
    21 / 35

    View Slide

  67. Rhyme Networks Testing
    Testing: Results
    Reconstruction System 100 # 200 # 300 # 400 # 500 # 600 # 700 # 800 #
    Karlgren (1954) 7 5 16 3 23 3 32 3 36 6 45 6 52 5 60 5
    Li 李方桂 (1971) 7 5 13 8 18 8 26 8 32 8 40 8 45 8 51 8
    Wáng 王力 (1980) 7 5 15 7 21 7 30 7 35 7 43 7 49 7 56 7
    Zhèngzhāng 鄭張尚芳 (2003) 9 2 16 3 23 3 32 3 40 3 49 3 55 3 64 3
    Starostin (1989) 9 2 17 2 25 2 35 2 43 2 51 2 59 2 68 2
    Pān 潘悟雲 (2000) 9 2 16 3 23 3 32 3 39 4 48 4 54 4 63 4
    Baxter and Sagart (2014) 10 1 18 1 27 1 37 1 46 1 55 1 63 1 72 1
    Schuessler (2007) 7 5 16 3 22 6 31 6 38 5 46 5 52 5 60 5
    21 / 35

    View Slide

  68. Rhyme Networks Testing
    Testing: Results
    Provided the principle of vowel purity was really dominant during
    time of the creation of the Shījīng, our results indicate that
    reconstruction systems with six vowels outperform those with less
    or more vowels.
    Given that we do not know to which degree vowel purity was
    important in Old Chinese rhyming, this does not allow us to prove or
    disprove any of the reconstruction systems.
    Further research on rhyming practice and pragmatics are needed.
    22 / 35

    View Slide

  69. Rhyme Networks Testing
    Testing: Appendix
    One obvious criticism in the vowel purity paper remains.
    23 / 35

    View Slide

  70. Rhyme Networks Testing
    Testing: Appendix
    One obvious criticism in the vowel purity paper remains.
    We tested on Baxter’s (1992) rhyme judgments, which could have
    easily influenced the results in the favor of the Baxter-Sagart system
    and of six-vowel systems in general!
    23 / 35

    View Slide

  71. Rhyme Networks Testing
    Testing: Appendix
    One obvious criticism in the vowel purity paper remains.
    We tested on Baxter’s (1992) rhyme judgments, which could have
    easily influenced the results in the favor of the Baxter-Sagart system
    and of six-vowel systems in general!
    We defended this with availability of data: when making the initial
    studies, no digital version of alternative rhyme judgments were
    available.
    23 / 35

    View Slide

  72. Rhyme Networks Testing
    Testing: Appendix
    One obvious criticism in the vowel purity paper remains.
    We tested on Baxter’s (1992) rhyme judgments, which could have
    easily influenced the results in the favor of the Baxter-Sagart system
    and of six-vowel systems in general!
    We defended this with availability of data: when making the initial
    studies, no digital version of alternative rhyme judgments were
    available.
    This is different now with the CHIP initiative, which just managed to
    digitize Wáng’s (1980) rhyme judgments along with the
    reconstructions.
    23 / 35

    View Slide

  73. Rhyme Networks Testing
    Testing: Appendix
    We can now compare the difference in rhyme judgments in Baxter
    (1992) and Wáng (1980).
    1I excluded those with multi-morpheme rhymes.
    24 / 35

    View Slide

  74. Rhyme Networks Testing
    Testing: Appendix
    We can now compare the difference in rhyme judgments in Baxter
    (1992) and Wáng (1980).
    A simple measure is to compare, how many stanzas differ. From
    1014 common stanzas1, 131 are different between Wáng and
    Baxter (12.9%).
    1I excluded those with multi-morpheme rhymes.
    24 / 35

    View Slide

  75. Rhyme Networks Testing
    Testing: Appendix
    We can now compare the difference in rhyme judgments in Baxter
    (1992) and Wáng (1980).
    A simple measure is to compare, how many stanzas differ. From
    1014 common stanzas1, 131 are different between Wáng and
    Baxter (12.9%).
    A far more useful measure is to compare how much different
    stanzas differ. Comparing rhyme judgments with a cluster task,
    B-Cubed scores are the perfect measure (Amigo et al. 2009).
    Applying B-Cubed scores to compare the rhyme judgments, we find
    96.8% of similarity between Baxter’s and Wáng’s rhyme judgments.
    1I excluded those with multi-morpheme rhymes.
    24 / 35

    View Slide

  76. Rhyme Networks Testing
    Testing: Appendix
    We can now compare the difference in rhyme judgments in Baxter
    (1992) and Wáng (1980).
    A simple measure is to compare, how many stanzas differ. From
    1014 common stanzas1, 131 are different between Wáng and
    Baxter (12.9%).
    A far more useful measure is to compare how much different
    stanzas differ. Comparing rhyme judgments with a cluster task,
    B-Cubed scores are the perfect measure (Amigo et al. 2009).
    Applying B-Cubed scores to compare the rhyme judgments, we find
    96.8% of similarity between Baxter’s and Wáng’s rhyme judgments.
    A new rhyme browser has now been created which contrasts
    rhymes by Wáng (1980) and Baxter (1992) and is available from
    http://digling.org/shijing/wangli/.
    1I excluded those with multi-morpheme rhymes.
    24 / 35

    View Slide

  77. More on Networks
    More Networks
    25 / 35

    View Slide

  78. More on Networks Fǎnqiè Networks
    Fǎnqiè Networks
    It has long since been known that fǎnqiè 反切 readings can also be
    analyzed by exploiting their network characteristics (see, e.g., Gěng
    Zhènshēng 耿振生 2004 on the fǎnqiè xìliánfǎ 反切系聯法).
    26 / 35

    View Slide

  79. More on Networks Fǎnqiè Networks
    Fǎnqiè Networks
    It has long since been known that fǎnqiè 反切 readings can also be
    analyzed by exploiting their network characteristics (see, e.g., Gěng
    Zhènshēng 耿振生 2004 on the fǎnqiè xìliánfǎ 反切系聯法).
    But with modern network approaches, we can handle the data more
    consistently and transparently.
    26 / 35

    View Slide

  80. More on Networks Fǎnqiè Networks
    Fǎnqiè Networks
    It has long since been known that fǎnqiè 反切 readings can also be
    analyzed by exploiting their network characteristics (see, e.g., Gěng
    Zhènshēng 耿振生 2004 on the fǎnqiè xìliánfǎ 反切系聯法).
    But with modern network approaches, we can handle the data more
    consistently and transparently.
    By extracting, for example, all fǎnqiè shàngzì 反切上字 from the
    Guǎngyùn 廣韻, we can create networks of fǎnqiè connections.
    26 / 35

    View Slide

  81. More on Networks Fǎnqiè Networks
    Fǎnqiè Networks
    It has long since been known that fǎnqiè 反切 readings can also be
    analyzed by exploiting their network characteristics (see, e.g., Gěng
    Zhènshēng 耿振生 2004 on the fǎnqiè xìliánfǎ 反切系聯法).
    But with modern network approaches, we can handle the data more
    consistently and transparently.
    By extracting, for example, all fǎnqiè shàngzì 反切上字 from the
    Guǎngyùn 廣韻, we can create networks of fǎnqiè connections.
    These networks are ideal for teaching Chinese traditional
    phonology, but also for comparison if scholars have different
    opinions.
    26 / 35

    View Slide

  82. More on Networks Fǎnqiè Networks
    Fǎnqiè Networks
    2
    2
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1 1 1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1 1
    1
    2
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1 1
    2 1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    2
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    2
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1 2
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1 1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    2
    2
    1
    1
    1
    1
    1
    2
    1
    1
    2
    1
    1
    1
    1
    1
    1
    2
    1
    1
    1
    1
    1
    1
    1
    2
    1
    1
    1
    1
    1
    1
    1
    2
    1
    1
    1
    1
    1
    1
    1
    2
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    2
    1
    1
    1
    1
    1
    1
    1
    2
    1
    1
    1
    1
    1
    1
    1
    2
    2
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    2
    1
    1
    1
    1
    1
    1
    1
    1
    1 1
    1
    1
    1
    1
    1
    1
    2
    2
    1
    1
    1
    1
    2
    2
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    2
    1
    1
    1 1
    1
    1
    1
    2
    1
    1
    2
    1
    2
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    2
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1 1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    2
    1
    1
    1
    1
    1
    1
    1
    1
    1
    3
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    2
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    2
    1
    1
    1
    1
    2
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    2
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    2
    1
    1
    1
    1
    1
    1
    1
    1
    1
    3
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    2
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    1
    27 / 35

    View Slide

  83. More on Networks Fǎnqiè Networks
    Fǎnqiè Networks
    1
    1
    1
    1
    1
    1
    1
    2
    1
    2
    1
    1
    1
    1
    1
    2
    2
    2
    1
    1
    1
    1
    1
    1










    姑 古






    詭 空


    27 / 35

    View Slide

  84. More on Networks Fǎnqiè Networks
    Fǎnqiè Networks
    1
    1
    1
    1
    1
    1
    1
    2
    1
    2
    1
    1
    2
    1
    2
    2
    1
    1
    1
    1
    1
    1
    1
    1
    kʰɐk
    kʰəu˥
    kʰɑk
    kʰu�
    kʰɑ˥
    kʰien�
    kʰu�
    kʰiem�
    ku˥
    kɑn�
    kwɐi�
    kai�
    ku�
    kɑk
    kiem�
    kʰuŋ�
    kĭwe˥
    kʰɐi˥
    kʰɑŋ�
    kɐk
    kuŋ�
    kuɑ�
    27 / 35

    View Slide

  85. More on Networks Fǎnqiè Networks
    Fǎnqiè Networks
    Fǎnqiè networks are still underexplored, both with respect to
    traditional scholarship on Chinese historical phonology and
    with respect to the way they are best handled, and poten-
    tial differences across Chinese rhyme books or other sources
    containing fǎnqiè readings from different epochs or authors.
    However, it seems promising to further exploit and test the ap-
    proaches, as they may drastically increase the transparency
    of current approaches.
    28 / 35

    View Slide

  86. More on Networks Xiéshēng Networks
    Xiéshēng Networks
    Since Duàn Yùcái 段玉裁 detected the strong correlation between
    the phonetic part of xíngshēng 形聲 characters, we know that the
    Chinese system basically reflects a network structure, since a large
    part of the characters can be decomposed into subparts which
    reflect other characters or recur across different characters.
    29 / 35

    View Slide

  87. More on Networks Xiéshēng Networks
    Xiéshēng Networks
    Since Duàn Yùcái 段玉裁 detected the strong correlation between
    the phonetic part of xíngshēng 形聲 characters, we know that the
    Chinese system basically reflects a network structure, since a large
    part of the characters can be decomposed into subparts which
    reflect other characters or recur across different characters.
    Not often really taken into consideration is the historical aspect of
    these connections. Not all phonetic units of xíngshēng characters
    were formed at the same time, and the characters reflect a complex
    evolution of character formation at different steps.
    29 / 35

    View Slide

  88. More on Networks Xiéshēng Networks
    Xiéshēng Networks
    Since Duàn Yùcái 段玉裁 detected the strong correlation between
    the phonetic part of xíngshēng 形聲 characters, we know that the
    Chinese system basically reflects a network structure, since a large
    part of the characters can be decomposed into subparts which
    reflect other characters or recur across different characters.
    Not often really taken into consideration is the historical aspect of
    these connections. Not all phonetic units of xíngshēng characters
    were formed at the same time, and the characters reflect a complex
    evolution of character formation at different steps.
    Instead of listing xíngshēng series in form of lists of characters and
    common component, we should create explicit networks, as they
    are much more transparent to display where scholars disagree in
    their analyses, but also which characters are immediately
    composed of other characters.
    29 / 35

    View Slide

  89. More on Networks Xiéshēng Networks
    Xiéshēng Networks

    滂 訪








    30 / 35

    View Slide

  90. More on Networks Xiéshēng Networks
    Xiéshēng Networks
    Displaying character structures, especially aspects of char-
    acter formation, with help of directed networks could greatly
    benefit not only scientific exchange among scholars, who
    would be encouraged to present their judgments more trans-
    parently, but also other aspects of Chinese writing, such as,
    e.g., pedagogical aspects of teaching the structure of the writ-
    ing system to beginners, or information-theoretic aspects.
    31 / 35

    View Slide

  91. More on Networks Dynamic Networks
    Dynamic Networks
    Not only the data we model in networks can be enhanced,
    also our methods to analyze the networks need to be further
    improved. As an example, consider dynamic networks, which
    would analyze and model network changes in time. By im-
    proving on these methods, we could, for example, compare
    fǎnqiè networks across different epochs, as well as rhyme
    networks from different authors, dialects, and styles. We
    could further try to induce fundamental hierarchies and rel-
    ative time frames from xiéshēng networks.
    32 / 35

    View Slide

  92. Outlook
    Outlook
    Outlook
    33 / 35

    View Slide

  93. Outlook
    One of the major insights that I made during the last years
    in my research is that despite the great achievements schol-
    ars have made in historical linguistics, and especially in Chi-
    nese traditional phonology, we still lack clear-cut frameworks
    that help us to produce our data transparently. Historical lin-
    guistics is a data-driven discipline, but scholars tend to ignore
    this when presenting their incredible insights in an intranspar-
    ent form. Networks can help in two ways here: first, they are
    a transparent way of data-representation; and second, they
    provide an added value in those cases, where data becomes
    too large for scholars to be inspected by eye-balling only.
    34 / 35

    View Slide

  94. Outlook
    Merciàtous!
    35 / 35

    View Slide