Complex Segment Learner

Complex Segment Learner

Yindjibarndi (Pama-Nyungan)

Yindjibarndi, a language not discussed in our paper, is described by Wordick (1982). Wordick does not posit an inventory containing any complex segments, although fn. 3 on p. 40 suggests that the language may have a series of prenasalized stops: a speaker, when asked to divide multi into syllables, provided mu.nti.
Our corpus is a hand-entered version of each headword provided in Wordick's lexicon; the short version contains all of the 1-2 syllable forms from that lexicon. In both cases, the learner's reaction to the Yindjibarndi data is similar to its reaction to the Quechua word corpus: it unifies all clusters in the data whose frequency is significantly different from 0.
We suspect that two factors lead to this result. First, Yindjibarndi consonantal phonotactics are quite restrictive: clusters appear only between vowels, and are limited in their composition (see Wordick). Second, Wordick's lexicon is small and includes a large number of morphologically complex forms. Unfortunately it is not easy to see whether or not the results would change given a different corpus, because Wordick's lexicon is the only one that we are aware of. Yindjibarndi thus stands as a difficult case for our learner.

Simulation data at a glance

Click on simulation name to view additional simulation details.

Simulation nameInitial state Learning DataInitial state features
All LearningData.txt Features.txt
Short LearningData.txt Features.txt

Simulation details for Yindjibarndi all

Input:

This corpus contains all headwords from Wordick's (1982) lexicon.

LearningData.txt | Features.txt

Summary of iterations:

IterationLearning Data producedFeatures producedInseparabilityNew Segments addedSegments removed
1 LearningData.txt Features.txt [download] [view] ngk, nhth, nyty, nt, rnrt None
2 LearningData.txt Features.txt [download] [view] mp, nk, rnk, rp, rk None
3 LearningData.txt Features.txt [download] [view] tp, tty, nyk, np, nty, rrk, rrw None
4 LearningData.txt Features.txt [download] [view] typ, nym, nm, rnp, rnm, ly None
5 LearningData.txt Features.txt [download] [view] rrm None
6 No new learning data No new features [download] [view] None None

Summary of inventory changes

StageConsonant set
Inputp k th ty t rt m ng nh ny n rn w yh y rr r l rl
Outputp k th ty t rt m ng nh ny n rn w yh y rr r l rl ngk nhth nyty nt rnrt mp nk rnk rp rk tp tty nyk np nty rrk rrw typ nym nm rnp rnm ly rrm

Simulation Plots

/media/yindjibarndi/all/simulation/insep_plots.png


Simulation details for Yindjibarndi short

Input:

This corpus contains all monosyllabic and disyllabic headwords words from Wordick's (1982) lexicon.

LearningData.txt | Features.txt

Summary of iterations:

IterationLearning Data producedFeatures producedInseparabilityNew Segments addedSegments removed
1 LearningData.txt Features.txt [download] [view] ngk, nhth, nyty, nt, rnrt None
2 LearningData.txt Features.txt [download] [view] mp, nk, rnk, rrw, rp, rk None
3 LearningData.txt Features.txt [download] [view] tp, np, rrk None
4 No new learning data No new features [download] [view] None None

Summary of inventory changes

StageConsonant set
Inputp k th ty t rt m ng nh ny n rn w yh y rr r l rl
Outputp k th ty t rt m ng nh ny n rn w yh y rr r l rl ngk nhth nyty nt rnrt mp nk rnk rrw rp rk tp np rrk

Simulation Plots

/media/yindjibarndi/short/simulation/insep_plots.png