Computer-Assisted Language Comparison

Slide 1

Slide 1 text

Computer-Assisted Language Comparison Bridging the Gap between Traditional and Quantitative Approaches in Historical Linguistics Johann-Mattis List Forschungszentrum Deutscher Sprachatlas Philipps-University Marburg 2014-10-09 1 / 50

Slide 2

Slide 2 text

Traditional Historical Linguistics 2 / 50

Slide 3

Slide 3 text

Traditional Historical Linguistics Characteristics Characteristics FRANZ BOPP VERY, VERY LONG TITLE 3 / 50

Slide 4

Slide 4 text

Traditional Historical Linguistics Characteristics Research Object 4 / 50

Slide 5

Slide 5 text

Traditional Historical Linguistics Characteristics Research Object German ʦ aː n - * Proto-Germanic t a n d English t ʊː θ - ** Proto-Indo-European d o n t Italian d ɛ n t e * Proto-Romance d e n t French d ɑ̃ - - 4 / 50

Slide 6

Slide 6 text

Slide 7

Slide 7 text

Traditional Historical Linguistics Characteristics Research Object German ʦ aː n - - * Proto-Germanic t a n d English t ʊː - θ - ** Proto-Indo-European d o n t Italian d ɛ n t e * Proto-Romance d e n t French d ɑ̃ - - - 4 / 50

Slide 8

Slide 8 text

Traditional Historical Linguistics Characteristics Research Object German ʦ aː n - - Proto-Germanic t a n θ - English t ʊː - θ - ** Proto-Indo-European d o n t Italian d ɛ n t e Proto-Romance d e n t e French d ɑ̃ - - - 4 / 50

Slide 9

Slide 9 text

Traditional Historical Linguistics Characteristics Research Object German ʦ aː n - Proto-Germanic t a n θ - English t ʊː - θ ** Proto-Indo-European d o n t Italian d ɛ n t e Proto-Romance d e n t e French d ɑ̃ - - 4 / 50

Slide 10

Slide 10 text

Traditional Historical Linguistics Characteristics Research Object German ʦ aː n - Proto-Germanic t a n θ - English t ʊː - θ Proto-Indo-European d e n t - Italian d ɛ n t ə Proto-Romance d e n t e French d ɑ̃ - - 4 / 50

Slide 11

Slide 11 text

Traditional Historical Linguistics Characteristics Research Object German ʦ aː n - * Proto-Germanic t a n d English t ʊː - θ Proto-Indo-European d e n t Italian d ɛ n t ə * Proto-Romance d e n t French d ɑ̃ - - 4 / 50

Slide 12

Slide 12 text

Traditional Historical Linguistics Characteristics Research Object German ʦ aː n Proto-Germanic t a n θ English t ʊː θ Proto-Indo-European d e n t Italian d ɛ n t e Proto-Romance d e n t e French d ɑ̃ German ʦ aː n Proto-Germanic t a n θ English t ʊː θ Proto-Indo-European d e n t Italian d ɛ n t e Proto-Romance d e n t e French d ɑ̃ 1 4 / 50

Slide 13

Slide 13 text

Traditional Historical Linguistics Characteristics Research Object History individual events (description( individual processes (description) general processes (modeling, analysis) Language History individual language states (description of sound system, grammar, lexicon) individual instances of language development (description of sound change patterns, grammaticalization, lexical change) general language development (modeling and analysis of sound change processes, grammaticalization, lexical change) 5 / 50

Slide 14

Slide 14 text

Traditional Historical Linguistics Characteristics Research Object Internal Language History (ontogenesis) etymology historical grammar historical phonology External Language History (phylogenesis) linguistic reconstruction proof of language relationship genetic classiﬁcation General Tendencies in Language History processes and mechanisms of sound change grammaticalization lexical change 6 / 50

Slide 15

Slide 15 text

Traditional Historical Linguistics Characteristics Origins Uniformitarianism “universality of change” – change is independent of time and space “graduality of change” – change is neither abrupt nor chaotic “uniformity of change” – change is not heterogeneous, but uniform Founding Fathers Franz Bopp (1791–1867): language comparison (Bopp 1816) Rasmus Rask (1787-1832) and Jacob Grimm (1785-1863): sound law (Rask 1818, Grimm 1822) August Schleicher (1821–1868): family tree and linguistic reconstruction (Schleicher 1853 & 1861) 7 / 50

Slide 16

Slide 16 text

Traditional Historical Linguistics Achievements Achievements 8 / 50

Slide 17

Slide 17 text

Traditional Historical Linguistics Achievements Methods and Theories Comparative Method (Meillet 1925) Basic procedure for proving language relationship and reconstructing unattested ancestral language states, etymologies, and genetic classiﬁcations. Family Tree Model and Wave Theory (Schleicher 1853, Schmidt 1872) Two partially incompatible models to describe historical language relations. Regularity Hypothesis (Osthoﬀ & Brugmann 1878) Fundamental working hypothesis that states that certain sound change processes proceed regularly (universally, gradually, and in a uniform manner). 9 / 50

Slide 18

Slide 18 text

Traditional Historical Linguistics Achievements Comparative Method proof of relationship identification of cognates identification of sound correspondences reconstruction of proto-forms internal classification 10 / 50

Slide 19

Slide 19 text

Slide 20

Slide 20 text

Traditional Historical Linguistics Achievements Comparative Method Cognate List Alignment Correspondence List German dünn d ʏ n GER ENG Frequ. d θ 3 x d d 1 x n n 1 x m m 1 x ŋ ŋ 1 x English thin θ ɪ n German Ding d ɪ ŋ English thing θ ɪ ŋ German dumm d ʊ m English dumb d ʌ m German Dorn d ɔɐ n English thorn d ɔː n 10 / 50

Slide 21

Slide 21 text

Slide 22

Slide 22 text

Traditional Historical Linguistics Achievements Comparative Method Cognate List Alignment Correspondence List German dünn d ʏ n GER ENG Frequ. d θ 2 x d d 1 x n n 1 x m m 1 x ŋ ŋ 1 x English thin θ ɪ n German Ding d ɪ ŋ English thing θ ɪ ŋ German dumm d ʊ m English dumb d ʌ m German Dorn d ɔɐ n English thorn d ɔː n 10 / 50

Slide 23

Slide 23 text

Slide 24

Slide 24 text

Traditional Historical Linguistics Achievements Comparative Method Cognate List Alignment Correspondence List German dünn d ʏ n GER ENG Frequ. d θ 3 x d d 1 x ? n n 2 x m m 1 x ŋ ŋ 1 x English thin θ ɪ n German Ding d ɪ ŋ English thing θ ɪ ŋ German dumm d ʊ m English dumb d ʌ m German Dorn d ɔɐ n English thorn θ ɔː n 10 / 50

Slide 25

Slide 25 text

Traditional Historical Linguistics Achievements Comparative Method Cognate List Alignment Correspondence List German dünn d ʏ n GER ENG Frequ. d θ 3 x d d 1 x n n 2 x m m 1 x ŋ ŋ 1 x English thin θ ɪ n German Ding d ɪ ŋ English thing θ ɪ ŋ German dumm d ʊ m English dumb d ʌ m German Dorn d ɔɐ n English thorn θ ɔː n 10 / 50

Slide 26

Slide 26 text

Traditional Historical Linguistics Achievements Insights Internal Language History Thanks to historical linguistics, the history of a considerable (but still small) amount of languages has been thoroughly investigated. External Language History Thanks to historical linguistics, a considerable amount of the languages in the world has been genetically classiﬁed (although there remain many unsolved and controversially discussed questions). General Language History Some work on general processes of language history has been done, yet many questions still remain unsolved or are controversially debated. 11 / 50

Slide 27

Slide 27 text

Traditional Historical Linguistics Problems Problems 12 / 50

Slide 28

Slide 28 text

Traditional Historical Linguistics Problems Transparency Part of the process of “becoming” a competent Indo-Europeanist has always been recognized as coming to grasp “intuitively” concepts and types of changes in language so as to be able to pick and choose between alternative explanations for the history and development of speciﬁc features of the reconstructed language and its oﬀspring. Schwink (1994) 13 / 50

Slide 29

Slide 29 text

Traditional Historical Linguistics Problems Applicability 14 / 50

Slide 30

Slide 30 text

Traditional Historical Linguistics Problems Applicability – 7,106 languages (Lewis & Fennig 2013) – 147 language families (ibid.) – 25244065 languages which could be compared 14 / 50

Slide 31

Slide 31 text

Traditional Historical Linguistics Problems Applicability The amount of digitally available data for the languages of the world is growing from day to day, while there are only a few historical linguists who are trained to carry out the comparison of these languages. It seems impossible to handle this task when relying only on the traditional, time- consuming manual procedures developed in traditional historical linguistics. 14 / 50

Slide 32

Slide 32 text

Traditional Historical Linguistics Problems Adequacy One time is never, two times is ever! (a mathematician friend on the treatment of probability in Indo-European linguistics) 15 / 50

Slide 33

Slide 33 text

Traditional Historical Linguistics Problems Summary Despite its achievements, traditional historical linguistics has some clear shortcomings, such as a lack of transparency in methodology, the “philological” form of knowledge representation, and the questionable validity of certain results. 16 / 50

Slide 34

Slide 34 text

Traditional Historical Linguistics Problems Example on “Philological Knowledge Representation” Frucht. Sf std. (9. Jh.), mhd. vruht, ahd. fruht, as. fruht. Ent- lehnt aus l. frūctus m. gleicher Bedeutung (zu l. fruī “ge- nieße”). Das deutsche Wort ist Femininum geworden im Anschluß an die ti- Abstrakta wie Flucht² usw. Adjekti- ve: fruchtig, fruchtbar; Verb: (be-)fruchten. Ebenso nndl. vrucht, ne. fruit, nfrz. fruit, nschw. frukt, nnorw. frukt; frugal. (Kluge und Seebold 2002) 17 / 50

Slide 35

Slide 35 text

Quantitative Historical Linguistics 18 / 50

Slide 36

Slide 36 text

Quantitative Historical Linguistics Characteristics Characteristics P(A|B)=(P(B|A)P(A))/(P(B) 19 / 50

Slide 37

Slide 37 text

Quantitative Historical Linguistics Characteristics Characteristics “Indo-European and computational cladistics” (Ringe, Warnow and Taylor 2002) “Language-tree divergence times support the Anatolian theory of Indo-European origin” (Gray und Atkinson 2003) “Language classiﬁcation by numbers” (McMahon und McMahon 2005) “Curious Parallels and Curious Connections: Phylogenetic Thinking in Biology and Historical Linguistics” (Atkinson und Gray 2005) “Automated classiﬁcation of the world’s languages” (Brown et al. 2008) “Computational Feature-Sensitive Reconstruction of Language Relationships: Developing the ALINE Distance for Comparative Historical Linguistic Reconstruction” (Downey et al. 2008) “Networks uncover hidden lexical borrowing in Indo-European language evolution” (Nelson-Sathi et al. 2011) “A pipeline for computational historical linguistics” (Steiner, Stadler, und Cysouw 2011) 20 / 50

Slide 38

Slide 38 text

Quantitative Historical Linguistics Characteristics Points of Interest and Goals phylogenetic reconstruction sequence comparison general questions of language development 21 / 50

Slide 39

Slide 39 text

Quantitative Historical Linguistics Characteristics Points of Interest and Goals phylogenetic reconstruction sequence comparison general questions of language development Goals If we cannot guarantee getting the same results from the same data considered by diﬀerent linguists, we jeopardize the essential scientiﬁc criterion of repeatability. (McMahon & McMahon 2005) 21 / 50

Slide 40

Slide 40 text

Quantitative Historical Linguistics Characteristics Methods and Theories phylogenetic reconstruction (cf., among others, Gray & Atkinson 2003 Ringe et al. 2002, Brown et al. 2008) phonetic alignment (cf., among others, Kondrak 2000, Prokić et al. 2009, List 2012a) cognate detection (cf. Steiner et al. 2011, List 2012b) borrowing detection (cf. Nelson-Sathi et al. 2011, List et al. 2014a) 22 / 50

Slide 41

Slide 41 text

Quantitative Historical Linguistics Achievements Achievements 23 / 50

Slide 42

Slide 42 text

Quantitative Historical Linguistics Achievements New Perspectives external language history receives more attention than before “Indo-Euro-Centrism” is replaced by a more cross-linguistic paradigm new questions regarding general language history new proposals to model language history 24 / 50

Slide 43

Slide 43 text

Quantitative Historical Linguistics Achievements New Approaches empirical data becomes the center of interest probabilistic approaches replace “historical” approaches databases replace philological knowledge representation “informal” methods are formalized and automatized 25 / 50

Slide 44

Slide 44 text

Quantitative Historical Linguistics Achievements Examples 26 / 50

Slide 45

Slide 45 text

Quantitative Historical Linguistics Achievements Examples: Phonetic Alignment Alignment Analyses Alignment analyses display sequence similarities by representing multiple sequences as rows of a matrix in which common segments are placed in the same column. Alignments are a formal way to deal with general tasks of sequence comparison. Although never explicitly labeled or displayed, alignments are virtually present in all analyses in historical linguistics dealing with the comparison of sound sequences (words, morphemes). 26 / 50

Slide 46

Slide 46 text

Slide 47

Slide 47 text

Slide 48

Slide 48 text

Slide 49

Slide 49 text

Quantitative Historical Linguistics Achievements Examples: Phonetic Alignment Sound Classes Sounds which frequently occur in correspondence relations in genetically related languages can be clustered into classes. It is thereby assumed that “phonetic correspondences inside a ‘type’ are more regular than those between diﬀerent ‘types’” (Dolgopolsky 1986[1964]: 35). 27 / 50

Slide 50

Slide 50 text

Slide 51

Slide 51 text

Slide 52

Slide 52 text

Slide 53

Slide 53 text

Slide 54

Slide 54 text

Slide 55

Slide 55 text

Quantitative Historical Linguistics Achievements Examples: Phonetic Alignment Sound-Class-Based Phonetic Alignment (SCA, List 2012 a) Sound classes and alignment analyses can be combined. Sound sequences are internally represented as sound classes. Alignments are carried out using standard algorithms developed in evolutionary biology. INPUT tɔxtər dɔːtər TOKENIZATION t, ɔ, x, t, ə, r d, ɔː, t, ə, r CONVERSION t ɔ x … → T O G … d ɔː t … → T O T … ALIGNMENT T O G T E R T O - T E R CONVERSION T O G … → t ɔ x … T O - … → d oː - … OUTPUT t ɔ x t ə r d ɔː - t ə r 1 28 / 50

Slide 56

Slide 56 text

Quantitative Historical Linguistics Achievements Examples: Phonetic Alignment SCA reaches an accuracy of more than 90 % for multiple alignment analyses, using the conservative column scores as evaluation scores. SCA can be applied to almost all languages, including tone languages (clicks are not yet supported), provided the data is given in regular phonetic transcription. SCA models prosodic properties of sound sequences and scores sound segments diﬀerently, depending on their position in the sequence, thereby accounting for general theories of prosodic strength . SCA is integrated in LingPy (http://lingpy.org, List & Moran 2013, an open source Python toolkit for quantitative tasks in historical linguistics and has been successfully tested on all major platforms (Mac, Linux, Microsoft). 29 / 50

Slide 57

Slide 57 text

Quantitative Historical Linguistics Achievements Examples: Automatic Cognate Detection LexStat (List 2012, List 2014) LexStat is a method for automatic cognate detection in multilingual wordlists. It uses on sound-class-based sequence alignment (SCA) analyses as a proxy to infer language-speciﬁc sound similarities (similar to the notion of sound correspondences in historical linguistics). Using the automatically inferred sound similarities, LexStat partitions words into cognate sets. 30 / 50

Slide 58

Slide 58 text

Quantitative Historical Linguistics Achievements Examples: Automatic Cognate Detection Basic Procedure for Multilingual Cognate Detection WORDLIST DATA 31 / 50

Slide 59

Slide 59 text

Quantitative Historical Linguistics Achievements Examples: Automatic Cognate Detection Basic Procedure for Multilingual Cognate Detection WORDLIST DATA PAIRWISE DISTANCES BETWEEN WORDS PAIRWISE COMPARISON 31 / 50

Slide 60

Slide 60 text

Slide 61

Slide 61 text

Quantitative Historical Linguistics Achievements Examples: Automatic Cognate Detection Cognate Clustering Analysis ID Taxa Word Gloss GlossID IPA ... ... ... ... ... ... 21 German Frau woman 20 frau 22 Dutch vrouw woman 20 vrɑu 23 English woman woman 20 wʊmən 24 Danish kvinde woman 20 kvenə 25 Swedish kvinna woman 20 kviːna 26 Norwegian kvine woman 20 kʋinə ... ... ... ... ... ... 31 / 50

Slide 62

Slide 62 text

Quantitative Historical Linguistics Achievements Examples: Automatic Cognate Detection Cognate Clustering Swedish English Danish Norwegian Dutch German kvinna woman kvinde kvine vrouw Frau Swedish kvina 0.00 0.69 0.07 0.12 0.71 0.78 English wumin 0.69 0.00 0.66 0.57 0.68 0.87 Danish kveni 0.07 0.66 0.00 0.08 0.67 0.71 Norwegian kwini 0.12 0.57 0.08 0.00 0.75 0.74 Dutch frou 0.71 0.68 0.67 0.75 0.00 0.17 German frau 0.78 0.87 0.71 0.74 0.17 0.00 Analysis ID Taxa Word Gloss GlossID IPA ... ... ... ... ... ... 21 German Frau woman 20 frau 22 Dutch vrouw woman 20 vrɑu 23 English woman woman 20 wʊmən 24 Danish kvinde woman 20 kvenə 25 Swedish kvinna woman 20 kviːna 26 Norwegian kvine woman 20 kʋinə ... ... ... ... ... ... 31 / 50

Slide 63

Slide 63 text

Quantitative Historical Linguistics Achievements Examples: Automatic Cognate Detection Cognate Clustering Swedish English Danish Norwegian Dutch German kvinna woman kvinde kvine vrouw Frau Swedish kvina 0.00 0.69 0.07 0.12 0.71 0.78 English wumin 0.69 0.00 0.66 0.57 0.68 0.87 Danish kveni 0.07 0.66 0.00 0.08 0.67 0.71 Norwegian kwini 0.12 0.57 0.08 0.00 0.75 0.74 Dutch frou 0.71 0.68 0.67 0.75 0.00 0.17 German frau 0.78 0.87 0.71 0.74 0.17 0.00 German Frau frau Dutch vrouw vrou English woman wumin Danish kvinde kveni Swedish kvinna kvina Norwegian kvine kwini 31 / 50

Slide 64

Slide 64 text

Quantitative Historical Linguistics Achievements Examples: Automatic Cognate Detection Cognate Clustering Swedish English Danish Norwegian Dutch German kvinna woman kvinde kvine vrouw Frau Swedish kvina 0.00 0.69 0.07 0.12 0.71 0.78 English wumin 0.69 0.00 0.66 0.57 0.68 0.87 Danish kveni 0.07 0.66 0.00 0.08 0.67 0.71 Norwegian kwini 0.12 0.57 0.08 0.00 0.75 0.74 Dutch frou 0.71 0.68 0.67 0.75 0.00 0.17 German frau 0.78 0.87 0.71 0.74 0.17 0.00 German Frau frau Dutch vrouw vrou English woman wumin Danish kvinde kveni Swedish kvinna kvina Norwegian kvine kwini 31 / 50

Slide 65

Slide 65 text

Quantitative Historical Linguistics Achievements Examples: Automatic Cognate Detection Cognate Clustering German Frau frau Dutch vrouw vrou English woman wumin Danish kvinde kveni Swedish kvinna kvina Norwegian kvine kwini Analysis ID Taxa Word Gloss GlossID IPA CogID ... ... ... ... ... ... ... 21 German Frau woman 20 frau 1 22 Dutch vrouw woman 20 vrɑu 1 23 English woman woman 20 wʊmən 2 24 Danish kvinde woman 20 kvenə 3 25 Swedish kvinna woman 20 kviːna 3 26 Norwegian kvine woman 20 kʋinə 3 ... ... ... ... ... ... ... 31 / 50

Slide 66

Slide 66 text

Quantitative Historical Linguistics Achievements Examples: Automatic Cognate Detection INPUT TOKENIZATION PREPROCESSING LOG-ODDS D ISTANCE COGNATE OUTPUT CORRESPONDENCE DETECTION USING PHONETIC ALIGNMENT LOOP DISTRIBUTION LexStat Algorithm (List 2014) EXPECTED ATTESTED DISTRIBUTION CALCULATION CLUSTERING 31 / 50

Slide 67

Slide 67 text

Quantitative Historical Linguistics Achievements Examples: Automatic Cognate Detection B-Cubed F-Scores on BDCD Benchmark (List 2014) Bai (Tibeto-Burman) Indo-European Japanese and Ryukyu Ob-Ugrian Austronesian Sinitic (Chinese Dialects) 60 65 70 75 80 85 90 95 Turchin NED SCA LexStat 32 / 50

Slide 68

Slide 68 text

Slide 69

Slide 69 text

Slide 70

Slide 70 text

Quantitative Historical Linguistics Achievements Examples: Phylogenetic Networks Hugo Schuchardt (1842-1927) 33 / 50

Slide 71

Slide 71 text

Quantitative Historical Linguistics Achievements Examples: Phylogenetic Networks Hugo Schuchardt (1842-1927) “We connect the branches and twigs of the tree with countless horizon- tal lines and it ceases to be a tree.” (Schuchardt 1870 [1900]: 11) 33 / 50

Slide 72

Slide 72 text

Quantitative Historical Linguistics Achievements Examples: Phylogenetic Networks Hugo Schuchardt (1842-1927) 33 / 50

Slide 73

Slide 73 text

Quantitative Historical Linguistics Achievements Examples: Phylogenetic Networks Hugo Schuchardt (1842-1927) 33 / 50

Slide 74

Slide 74 text

Quantitative Historical Linguistics Achievements Examples: Phylogenetic Networks Biological Workflow (Dagan & Martin 2007, Dagan et al. 2008) 1 collect phyletic pattern data (shared gene families) of the taxa that shall be investigated 2 use gain-loss mapping techniques with different weighting models, allowing for different amounts of gain events to analyze how the gene families evolved along a given reference tree 3 use ancestral genome sizes as an external criterion to determine the best weighting model 4 assume that all patterns for which the best model yields more than one gain event result from lateral gene transfer 5 reconstruct a minimal lateral network (MLN) by connecting multiple gains for the same gene family by lateral edges 34 / 50

Slide 75

Slide 75 text

Quantitative Historical Linguistics Achievements Examples: Phylogenetic Networks Linguistic Workflow (Nelson-Sathi et al. 2011, List et al. 2014a) 1 collect phyletic pattern data (shared cognates) of the languages that shall be investigated 2 use gain-loss mapping techniques with different weighting models, allowing for different amounts of to analyze how the cognates evolved along a given reference tree 3 use ancestral vocabulary size distributions as an external criterion to determine the best weighting model 4 allow for a substantial amount (5%) of parallel evolution 5 assume that all patterns for which the best model yields more than one gain event result from lateral gene transfer 6 reconstruct a minimal lateral network by connecting multiple gains of the same cognate by lateral edges 35 / 50

Slide 76

Slide 76 text

Quantitative Historical Linguistics Achievements Examples: Phylogenetic Networks 36 / 50

Slide 77

Slide 77 text

Quantitative Historical Linguistics Achievements Examples: Phylogenetic Networks -- Spanish -- French -- Italian Danish -- English -- German -- 36 / 50

Slide 78

Slide 78 text

Quantitative Historical Linguistics Achievements Examples: Phylogenetic Networks -- Spanish -- French -- Italian Danish -- English -- German -- 36 / 50

Slide 79

Slide 79 text

Quantitative Historical Linguistics Achievements Examples: Phylogenetic Networks -- Spanish -- French -- Italian Danish -- English -- German -- 36 / 50

Slide 80

Slide 80 text

Quantitative Historical Linguistics Achievements Examples: Phylogenetic Networks -- Spanish -- French -- Italian Danish -- English -- German -- 36 / 50

Slide 81

Slide 81 text

Quantitative Historical Linguistics Achievements Examples: Phylogenetic Networks List et al. (2014b) . . ---Lánzhōu . Fùzhōu -- . Xiāngtàn -- . M ěixiàn -- . H ongkong -- . ---Wǔhàn . ---Běijīng . ---Kùnmíng . Hángzhōu -- . Xiàmén -- . ---Chéngdū . Sùzhōu -- . Shànghǎi -- . Táiběi -- . ---Zhèngzhōu . Shèxiàn -- . ---Nánjīng . ---Guìyáng . W énzhōu -- . N ánníng -- . Tūnxī -- . ---Tiānjìn . Shāntóu -- . ---Xīníng . ---Q īngdǎo . ---Ürüm qi . ---Píngyáo . Nánchàng -- . ---Tàiyuán . Chángshā -- . Hǎikǒu -- . ---Héfèi . Jiàn'ǒu -- . ---Yīnchuàn . ---Hohhot . Táoyuán -- . ---Xī'ān . G uǎngzhōu -- . ---Harbin . ---Jìnán . 0 . 0 . 0 . Inferred Links 36 / 50

Slide 82

Slide 82 text

Slide 83

Slide 83 text

Slide 84

Slide 84 text

Slide 85

Slide 85 text

Quantitative Historical Linguistics Problems Problems 37 / 50

Slide 86

Slide 86 text

Quantitative Historical Linguistics Problems Transparency 38 / 50

Slide 87

Slide 87 text

Quantitative Historical Linguistics Problems Transparency Evaluation criteria for applied automatic methods are not very intuitive and vary greatly. 38 / 50

Slide 88

Slide 88 text

Slide 89

Slide 89 text

Quantitative Historical Linguistics Problems Transparency Evaluation criteria for applied automatic methods are not very intuitive and vary greatly. Benchmark databases are rarely used, especially in phylogenetic approaches eyeballing of phylogenetic trees is sold as proof for “valid approaches”. It is diﬃcult to communicate the results to traditional linguists. 38 / 50

Slide 90

Slide 90 text

Slide 91

Slide 91 text

Slide 92

Slide 92 text

Slide 93

Slide 93 text

Slide 94

Slide 94 text

Quantitative Historical Linguistics Problems Applicability 39 / 50

Slide 95

Slide 95 text

Quantitative Historical Linguistics Problems Applicability Method Multilingual? No additional requirements? Freely Available? Mackay & Kondrak 2005 ✗ ✓ ✗ Bergsma & Kondrak 2007 ✓ ✓ ✗ Turchin et al. 2010 ✓ ✓ ✓ Berg-Kirkpatrick & Klein 2011 ✗ ✓ ✗ Hauer & Kondrak 2011 ✓ ✓ ✗ Steiner et al. 2011 ✓ ✓ ✗ List 2012 & List 2014 ✓ ✓ ✓ Beinborn et al. 2013 ✗ ? ✗ Bouchard-Côté et al. 2013 ✓ ✗ ✗ Rama 2013 ✗ ✓ ✗ Ciobanu & Dinu 2014 ✗ ✓ ✗ … … … … 39 / 50

Slide 96

Slide 96 text

Quantitative Historical Linguistics Problems Applicability Method Multilingual? No additional requirements? Freely Available? Mackay & Kondrak 2005 ✗ ✓ ✗ Bergsma & Kondrak 2007 ✓ ✓ ✗ Turchin et al. 2010 ✓ ✓ ✓ Berg-Kirkpatrick & Klein 2011 ✗ ✓ ✗ Hauer & Kondrak 2011 ✓ ✓ ✗ Steiner et al. 2011 ✓ ✓ ✗ List 2012 & 2014 ✓ ✓ ✓ Beinborn et al. 2013 ✗ ? ✗ Bouchard-Côté et al. 2013 ✓ ✗ ✗ Rama 2013 ✗ ✓ ✗ Ciobanu & Dinu 2014 ✗ ✓ ✗ … … … … 39 / 50

Slide 97

Slide 97 text

Quantitative Historical Linguistics Problems Applicability Method Multilingual? No additional requirements? Freely Available? Mackay & Kondrak 2005 ✗ ✓ ✗ Bergsma & Kondrak 2007 ✓ ✓ ✗ Turchin et al. 2010 ✓ ✓ ✓ Berg-Kirkpatrick & Klein 2011 ✗ ✓ ✗ Hauer & Kondrak 2011 ✓ ✓ ✗ Steiner et al. 2011 ✓ ✓ ✗ List 2012 & 2014 ✓ ✓ ✓ Beinborn et al. 2013 ✗ ? ✗ Bouchard-Côté et al. 2013 ✓ ✗ ✗ Rama 2013 ✗ ✓ ✗ Ciobanu & Dinu 2014 ✗ ✓ ✗ … … … … 39 / 50

Slide 98

Slide 98 text

Quantitative Historical Linguistics Problems Applicability Method Multilingual? No additional requirements? Freely Available? Mackay & Kondrak 2005 ✗ ✓ ✗ Bergsma & Kondrak 2007 ✓ ✓ ✗ Turchin et al. 2010 ✓ ✓ ✓ Berg-Kirkpatrick & Klein 2011 ✗ ✓ ✗ Hauer & Kondrak 2011 ✓ ✓ ✗ Steiner et al. 2011 ✓ ✓ ✗ List 2012 & 2014 ✓ ✓ ✓ Beinborn et al. 2013 ✗ ? ✗ Bouchard-Côté et al. 2013 ✓ ✗ ✗ Rama 2013 ✗ ✓ ✗ Ciobanu & Dinu 2014 ✗ ✓ ✗ … … … … 39 / 50

Slide 99

Slide 99 text

Slide 100

Slide 100 text

Quantitative Historical Linguistics Problems Accuracy Data Problems (Geisler & List forthcoming) Comparing two independently produced lexicostatistical datasets: database # languages # concepts Dyen et al. 1997 95 200 Tower of Babel 98 110 intersection 46 103 Results up to 10 % diﬀerence in concept translations many undetected borrowings in both datasets up to 30 % diﬀerences in tree topologies for Bayesian analyses 40 / 50

Slide 101

Slide 101 text

Quantitative Historical Linguistics Problems Accuracy 40 / 50

Slide 102

Slide 102 text

Quantitative Historical Linguistics Problems Summary Many quantitative methods which are based on manually compiled datasets cannot cope with errors resulting from inconsistent data compilation. They are only as objective as the data being fed to them! Many quantitative approaches are insuﬃciently tested, and scholars are often content with results traditional linguists would never accept. Additionally, quantitative approaches are often presented in a way that makes it hard (not only for traditional linguists) to understand what they are based upon. Results are reported in an intransparent way, supplementary data is often lacking, concrete examples are seldom provided and source code (essential to check and replicate analyses) is missing in almost all recent publications. 41 / 50

Slide 103

Slide 103 text

Computer-Assisted Language Comparison 42 / 50

Slide 104

Slide 104 text

Computer-Assisted Language Comparison Bridging the Gap Bridging the Gap So far, the majority of computational approaches in historical linguistics largely disregards the actual needs of historical linguistics. Despite the frequent claims that the algorithms are intended to supplement traditional research, many of them are mere attempts to prove the power of modern machine learning approaches and completely disregard the achievements of traditional research in historical linguistics. 43 / 50

Slide 105

Slide 105 text

Computer-Assisted Language Comparison Bridging the Gap Bridging the Gap If we really want to make a diﬀerence with computational approaches and not simply seek to replace every expert who likes books with a computer or abacus, we need to work much, much harder, on a real integration of computational and traditional approaches. 43 / 50

Slide 106

Slide 106 text

Computer-Assisted Language Comparison Bridging the Gap Bridging the Gap P(A|B)=(P(B|A)P(A))/(P(B) FRANZ BOPP VERY, VERY LONG TITLE 43 / 50

Slide 107

Slide 107 text

Computer-Assisted Language Comparison Bridging the Gap Bridging the Gap P(A|B)=(P(B|A)P(A))/(P(B) FRANZ BOPP VERY, VERY LONG TITLE 43 / 50

Slide 108

Slide 108 text

Computer-Assisted Language Comparison Bridging the Gap Bridging the Gap P(A|B)=(P(B|A)P(A))/(P(B) FRANZ BOPP VERY, VERY LONG TITLE Apart from “computational historical linguistics”, we need to establish a new discipline of “computer-aided historical linguistics”. Such a framework needs bench- marks and new standards to cope with general problems of quantitative approaches. However, such a framework will also need additional resources that help traditional approaches to leave the “realm of intuition”. 43 / 50

Slide 109

Slide 109 text

Computer-Assisted Language Comparison Examples Examples 44 / 50

Slide 110

Slide 110 text

Computer-Assisted Language Comparison Examples Benchmark Databases for Historical Linguistics 45 / 50

Slide 111

Slide 111 text

Computer-Assisted Language Comparison Examples Benchmark Databases for Historical Linguistics First benchmark databases have been compiled and published: Benchmark Database of Phonetic Alignments (BDPA, List & Prokić 2014, http://alignments.lingpy.org) Benchmark Database for Cognate Detection (BDCD, presented in List 2014, http://sequencecomparison.github.io). Benchmark Database for Linguistic Reconstruction (BDLR, in preparation). 45 / 50

Slide 112

Slide 112 text

Computer-Assisted Language Comparison Examples Benchmark Databases for Historical Linguistics All data is given in phonetic transcriptions (IPA), tokenized into phonemic units, freely available for download, and can be directly used in LingPy. 45 / 50

Slide 113

Slide 113 text

Computer-Assisted Language Comparison Examples Visualizations and Interactive Applications Often, automatic approaches hide essential aspects of their analyses. These aspects are not only valid to test the power of methods, but also to get the best out of the results. Aggregation of results is useful for publications, but we know, that “every word has its own history”, and traditional research has always been concerned with this. Visualizing and reporting all detailed decisions and findings of automatic methods will not only increase their transparency, it may also help convincing traditional scholars that computational approaches may provide valuable insights. Apart from static visualizations, JavaScript and HTML5 offer unique ways for interactive data visualization and make it easy to produce, share, and explore what automatic methods have produced. So far, we have develop JavaScript prototype tools that – visualize phonetic alignments of cognate sets (JavaScript Cognate Viewer, JCOV, http://github.com/dighl/jcov/), – allow to edit and refine alignments and cognate sets online using online tools (Etymological Dictionary Editor, EDICTOR, http://tsv.lingpy.org), and – tools that visualize phylogenetic trees in geographic space (together with T. Mayer, Tree Explorer, TREX, http://github.com/dighl/TREX). 46 / 50

Slide 114

Slide 114 text

Computer-Assisted Language Comparison Challenges Challenges 47 / 50

Slide 115

Slide 115 text

Computer-Assisted Language Comparison Challenges Challenges German m oː n t - English m uː n - - Danish m ɔː n - ə Swedish m oː n - e 47 / 50

Slide 116

Slide 116 text

Computer-Assisted Language Comparison Challenges Challenges German m oː n t - English m uː n - - Danish m ɔː n - ə Swedish m oː n - e Fúzhōu ŋ u o ʔ ⁵ - - - - - - - - - - Měixiàn ŋ i a t ⁵ - - - - - k u o ŋ ⁴⁴ Guǎngzhōu j - y t ² l - œ ŋ ²² - - - - - Běijīng - y ɛ - ⁵¹ l i ɑ ŋ - - - - - - 47 / 50

Slide 117

Slide 117 text

Slide 118

Slide 118 text

Computer-Assisted Language Comparison Challenges Challenges Fúzhōu Měixiàn Guǎngzhōu Běijīng 47 / 50

Slide 119

Slide 119 text

Computer-Assisted Language Comparison Challenges Challenges Fúzhōu Měixiàn Guǎngzhōu Běijīng INNO VATIO N INNO VATIO N INNO VATIO N BO RRO W ING LO SS INNO VATIO N INNO VATIO N 47 / 50

Slide 120

Slide 120 text

Computer-Assisted Language Comparison Challenges Challenges SEMANTIC CHANGE MORPHOLOGICAL CHANGE S T R A T IC C H A N G E Three Dimensions of Lexical Change (Gévaudan 2007) 47 / 50

Slide 121

Slide 121 text

Computer-Assisted Language Comparison Challenges Challenges In order to cope with the multiple dimensions of lexical change, we need new methods and models in historical linguistics, which explicitly deal with borrowing, partial cognacy, and semantic change. Following the lead of evolutionary biology, these methods could be combined under a uniﬁed framework of tree reconciliation (Page & Cotton 2002) in historical linguistics. 48 / 50

Slide 122

Slide 122 text

Computer-Assisted Language Comparison Challenges Challenges Fúzhōu Měixiàn Guǎngzhōu Běijīng Fúzhōu Měixiàn Guǎngzhōu Běijīng 48 / 50

Slide 123

Slide 123 text

Computer-Assisted Language Comparison Challenges Challenges Fúzhōu Měixiàn Guǎngzhōu Běijīng Fúzhōu Měixiàn Guǎngzhōu Běijīng 48 / 50

Slide 124

Slide 124 text

Computer-Assisted Language Comparison Challenges Challenges Fúzhōu Měixiàn Guǎngzhōu Běijīng 48 / 50

Slide 125

Slide 125 text

Computer-Assisted Language Comparison Challenges Challenges Fúzhōu Měixiàn Guǎngzhōu Běijīng 48 / 50

Slide 126

Slide 126 text

Computer-Assisted Language Comparison Challenges Challenges LOSS INNO VATIO N INNO VATIO N BORROWING 48 / 50

Slide 127

Slide 127 text

Conclusion Conclusion 49 / 50

Slide 128

Slide 128 text

Conclusion Conclusion Automatic approaches are constantly gaining ground in historical linguistics. 49 / 50

Slide 129

Slide 129 text

Slide 130

Slide 130 text

Slide 131

Slide 131 text

Slide 132

Slide 132 text

Conclusion Conclusion Automatic approaches are constantly gaining ground in historical linguistics. Nevertheless, the majority of the new approaches shows a great lack in transparency and applicability. One reason for this is the gap between traditional and computational approaches which are mostly applied independently from each other. In order to increase the interaction between traditional and computational historical linguists, we need a paradigm shift in historical linguistic. Computational linguists need to increase the transparency of their results, focusing on their detailed and interactive presentation instead of hiding behind numbers. 49 / 50