Kʼicheʼ (ISO-639-3: quc) is a Mayan language spoken mainly in Guatemala.
Kʼicheʼ is an agglutinative language, meaning that words are built up of multiple morphs which express inflectional and derivational categories. For example,
| Kʼicheʼ | Segmentation | Gloss | Translation |
|---|---|---|---|
| Xatinwilo | X-at-inw-il-o | PERF-O2SG-S1SG-see-SS |
I saw you. |
| qatinamit | qa-tinamit | P1PL-village |
Our village. |
| nutijoxelabʼ | nu-tijoxel-abʼ | P1SG-student-PL |
My students. |
| Xkunataj | X-kuna-taj | PERF-S3SG-cure-PASS |
Was cured. |
Being able to segment these words into their corresponding morphs can be useful for other natural language processing tasks, such as predictive text and machine translation. The objective of this task is to produce a system which given a sentence in Kʼicheʼ produces the sentence segmented into morphs.
In the data/ subdirectory you will find a training set. The development set is very small, so
for setting hyperparameters of your model you should probably use cross-validation.
Here we use > as a delimiter for the morphs.
Retaʼm uwach jachin xloqʼow ri pas. R>etaʼm u>wach jachin x>loqʼow ri pas.
Na xwaj taj xinbʼij Na x>w>a>j taj x>in>bʼi>j
Xinpaj ri nutriko. X>in>paj ri nu>triko.
Xinta chi katulik X>in>ta chi k>at>ul>ik
Na xwaj taj xinchʼawik Na x>w>aj taj x>in>chʼaw>ik
Kinwaʼik rech na kinyowaj taj K>in>waʼ>ik r>ech na k>in>yowaj taj
The baseline system is a simple weighted finite-state transducer which tries to maximise the frequency of sets of morphemes per token.
Evaluation is F-score, that is,
Where P is Precision and R is Recall.
The script evaluate.py in this repository calculates the precision, recall and F-score for
a test file. You can use it as follows:
$ python3 evaluate.py <ref_file> <tst_file>The script expects the input to both files to be a two-column tab-separated file where the first column is the input and the second column is the output of the segmenter. For example,
$ python3 evaluate.py data/test.tsv output.tsv
10 sentences read, 82 tokens
P: 0.9573170731707317
R: 0.9542682926829268
F: 0.9552845528455286Where data/test.tsv is the test file and output.tsv is the output from your system.