5. Inflection classes¶

Qumin’s basic unit when looking at inflection classes is the microclass.

Microclass:: A set of lexemes that share the exact same set of alternation patterns across all pairs of cells.

The three algorithms presented below manipulate microclasses in various ways to provide meaningful insights into their structure.

Microclass heatmaps¶

The microclass heatmap script uses a clustering algorithm to identify simmilarities between microclasses:

qumin action=heatmap data=parakar/parakar.package.json \
patterns=results/patterns/metadata.json \
cells="[abl.sg, gen.sg, part.sg]" hydra.run.dir=results/microclasses

In the example below, two large groups emerge, which are identifiable both through the colors on the heatmap and through the tree on the side of the picture:

Heatmap of Livvi Karelian microclass similarities.

Tip

Have a look at the microclass heatmap reference to see all available options for microclass computations.

Inflection class lattices¶

Inflection classe lattices (Beniamine 2018, Beniamine 2021) show which microclasses share which patterns:

qumin action=lattice data=parakar/parakar.package.json \
patterns=results/patterns/metadata.json \
cells="[abl.sg, gen.sg, part.sg]" hydra.run.dir=results/lattice \
lattice.png=True lattice.html=True

Each node at the bottom of the hierarchy represents a microclass. Each node at higher levels represents a set of microclasses that share one or several alternation patterns:

Lattice of Livvi Karelian inflection system.

Tip

Have a look at the lattice reference to see all available options for inflection class lattices.

Macroclasses¶

Warning

This algorithm is much slower than the others. It is implemented to provide reproductibility for the results of Beniamine & al (2018) but we recommend using other algorithms for a more in depth understanding of inflection classes.

qumin action=macroclasses data=parakar/parakar.package.json \
patterns=results/patterns/metadata.json \
cells="[abl.sg, gen.sg, part.sg]" hydra.run.dir=results/macroclasses

The macroclass algorithm merges microclasses into larger classes, trying to minimize the length of the description of the system. It outputs a list of macroclasses and a merge tree:

Tree of macroclasses in a sample of Livvi Karelian cells.

Tip

Have a look at the macroclass reference to see all available options for macroclass computations.