Guillaume Segerer CNRS - LLACAN - France [e-mail]

1
Guillaume SegererCNRS - LLACAN -
Francesegerer_at_vjf.cnrs.fr

• Niger-Congo Languages as a playground for lexical
  comparison

LYON, May 12-14, 2008 New Directions in
Historical Linguistics Paper presented May
14 Document revised May 16
2
Languages of Africa
3
Niger-Congo Languages
4
Niger-Congo clusters
5
The experiment

• The present experiment consists in
• - testing the validity of the Niger-Congo phylum
  by measuring its homogeneity
• - doing real mass comparison 506 languages
  examined ( 1/3 of all NC languages)
• - with only a few lexical roots, chosen
  intuitively from empirical experience

It can be further refined by - considering more
languages (more data is actually available) -
chosing different lexical roots (but how ?) -
taking into account adjacent phyla (Nilo-Saharan,
Afro-Asiatic, Khoisan)
6
Language Sample (506 lgs)
7
10 supposedly common NC lexical roots

• 1 TU(P) to spit
• 2 MED MOD to swallow
• 3 NYU to drink
• 4 DUM to bite
• 5 TE tree
• 6 NYI(N) tooth
• 7 TU ear
• 8 DEM tongue
• 9 DI to eat
• 10 TAT three

8
Distribution of root 1
9
Distribution of root 2
10
Distribution of root 3
11
Distribution of root 4
12
Distribution of root 5
13
Distribution of root 6
14
Distribution of root 7
15
Distribution of root 8
16
Distribution of root 9
17
Distribution of root 10
18
Weighted sample
19
Probabilities 1
consonants labial symbol P dental/coronal
symbol T palatal symbol C velar/uvular symbol
K
vowels front symbol I central symbol A back
symbol U
example DEM tongue coded as TIP probability 1/4
x 1/3 x 1/4 1/48 tolerance IA gt new prob.
1/24 Out of the 506 sample languages, 1/24 21
languages may by chance have a word for tongue
of the shape TIP TAP
20
Probabilities 2
Probabilities for each of the 10 roots TAT -
three TAT TIT gt 1/24 DUM - bite TUP TUT
gt 1/24 DEM - tongue TIP TAP gt 1/24 MED
MOD - swallow PIT PUT gt 1/24 TU(P) - spit
TU CU gt 1/6 TU - ear TU CU gt 1/6 NYI(N) -
tooth CI TI gt 1/6 TE - tree TI TA CI gt
1/4 DI - eat TI CA CI gt 1/4 NYU - drink
CU TU KUgt 1/3
probability to have all 10 items 1/ 143 327
232 4 languages in the sample have all 10 items
Akpafu (Kwa), Sukuma (Bantu F21, Runyankore
(Bantu E13), Andoni (Benue-Congo)
21
Probabilities 3
probability to have at least 1 item 18 of 1565
lgs gt 286 lgsprobability to have at least 2
items 19 27 of 286 lgs gt 55 79
lgsprobability to have at least 3 items 29
37 of 55 79 lgs gt 16 29 lgs
22
Some questions...

• Can this method be used to classify a language ?
• What is the minimal number of items needed to
  identify a language cluster ?
• Is there a method (other than intuitive) to
  identify these items ?
• Can this technique be applied to any language
  family / cluster ?
• What are the implications of these phenomena ?
• ...

23
A restricted distribution
GOP possible Atlantic lexical innovation