This repository is for creating a finite-state transducer (FST) in the Ojibwe language. The FST can be used to generate morphological analyses for an inflected form, or vice versa.
Morphological information about Ojibwe words is housed here. Combined with the FST-generating code in FSTmorph and the Ojibwe lexical information stored in OjibweLexicon, the FST can be generated as specified below.
These results reflect the performance of an FST built from the morphology stored in OjibweMorph and the lemmas stored in OjibweLexicon/OPD and OjibweLexicon/HammerlyFieldwork.
Each test form is inputted to the FST, then the corresponding analysis outputted by the FST is checked for correctness.
These inflected test forms come from the OPD, and are stored in OjibweLexicon/OPD/for_yaml. This is a large test set covering a variety of word forms.
For these and the paradigm tests, the "# of Forms Without Results" counts the test forms for which the FST provides no analysis whatsoever. The "Precision" captures how many outputted analyses were correct, while the "Recall" captures how many of the correct analyses were outputted. Note that some forms have multiple correct analyses.
| Date Last Updated | # of Forms Tested | # of Forms Without Results | Precision | Recall |
|---|---|---|---|---|
| 2025-07-21 | 66801 | 135 | 77.25% | 97.01% |
| Date Last Updated | # of Forms Tested | # of Forms Without Results | Precision | Recall |
|---|---|---|---|---|
| 2025-07-21 | 8565 | 15 | 83.4% | 96.92% |
The inflected forms used in these tests come from the NounSpreadsheets/ and VerbSpreadsheets/ folders here in OjibweMorph. This smaller test set is used largely as a sanity check.
| Date Last Updated | # of Forms Tested | # of Forms Without Results | Precision | Recall |
|---|---|---|---|---|
| 2025-07-21 | 8089 | 0 | 93.93% | 100.0% |
| Date Last Updated | # of Forms Tested | # of Forms Without Results | Precision | Recall |
|---|---|---|---|---|
| 2025-07-21 | 14330 | 0 | 99.98% | 100.0% |
The inflected forms used in these tests come from example sentences in the OPD, stored in OjibweLexicon/OPD/example_sentences.
The overall results are given at the bottom of the table, but a breakdown by the speaker of the example sentence is also provided first. Because the speakers come from a variety of communities, this can give an impression of how well the FST is covering different varieties of Ojibwe. You can learn more about the speakers here.
In the table below, we are simply counting 'failures' -- forms that receive no analysis whatsover from the FST. This is because unlike with the OPD and paradigm tests, we do not have a "gold standard" analysis to check. The "by-token" failure covers every token (word) in the example sentences, whereas the "by-type" failures consider every unique token (i.e., so that each token only counts once towards the score regardless of its frequency).
| Speaker | Region | Community | By-Token Failure | By-Type Failure |
|---|---|---|---|---|
| NJ | Border Lakes | Nigigoonsiminikaaning | 5.03% (335/6651) | 7.32% (314/4285) |
| GJ | Border Lakes | Lac La Croix | 12.32% (9/73) | 12.5% (9/72) |
| ES | Red Lake | Obaashiing | 5.65% (539/9531) | 10.51% (518/4925) |
| RG | Red Lake | Odaawaa-Zaaga'iganiing | 2.54% (56/2197) | 4.44% (55/1237) |
| GH | Leech Lake | Jaachaabaaning | 2.71% (7/258) | 3.39% (7/206) |
| LW | Leech Lake | Jaachaabaaning | 2.63% (5/190) | 3.24% (5/154) |
| LS | Mille Lacs | Aazhomog | 8.19% (5/61) | 9.61% (5/52) |
| LSA | Mille Lacs | Lake Lena | 3.22% (1/31) | 3.44% (1/29) |
| Unknown | N/A | N/A | 0.0% (0/5) | 0.0% (0/5) |
| Overall | 5.03% (957/18997) | 9.24% (906/9803) |
Date Last Updated: 2025-07-21
To make use of the FST, you will need to install the foma compiler (and the flookup program, included as part of the foma toolkit). On Mac or Linux, the easiest way to install is via homebrew. Just use the command brew install foma. Alternatively, there are other installation instructions here (including for Windows users).
Note for Windows users: In addition to the page given above, we found these instructions useful for installing. Also, ensure that the directory you add to your PATH immediately contains
foma.exeandflookup.exe. For example, if the path tofoma.exeisC:\Program Files (x86)\Foma\win32\foma.exe, then addC:\Program Files (x86)\Foma\win32(notC:\Program Files (x86)\Foma\) to your PATH.
-
Clone OjibweLexicon
In addition to this repository, you'll also need to get OjibweLexicon installed locally. -
Install FSTmorph
The FST is created using code in FSTmorph, which makes use of language-specific information stored in both OjibweMorph and OjibweLexicon.
FSTmorph can be installed via pip:
pip install fstmorph -
Make edits to the
Makefileas needed
The Makefile in this repo contains variables for various file locations. For the most part the pre-set values should work fine, but you should ensure that the location of OjibweLexicon (i.e., theOJIBWE_LEXICONvar) is correct for your local installation. -
Use the
Makefile
make allto simply build the FSTmake checkto run tests on the FSTmake cleanto remove generated files
Note: When running these commands, we have sometimes encountered an error message related to
malloc. It seems to happen randomly, and you can just run the command again (perhaps running thecleancommand above in between) until the error does not occur.
By default, the output will go in a local directory called FST/. In there, the directory generated contains the FST, lexc files and XFST rules. The FST itself is FST/generated/ojibwe.fomabin.
By default, the lemma list will be taken from OjibweLexicon/OPD and OjibweLexicon/HammerlyFieldwork. You can change this in the Makefile to look elsewhere. You can use multiple lists by giving a comma-separated list of directories.
This FST is run using foma. In addition to the example give below, some documentation from their team is available here.
-
Start the Foma program with the following command:
foma
Some information about foma should appear, and your prompt should now sayfoma[0]:. -
Load the Ojibwe FST you created in the previous section with:
load FST/generated/ojibwe.fomabin
The output should look something like this:
5.0 MB. 160042 states, 328686 arcs, Cyclic. -
Now you can use the FST.
a. To get an analysis for an inflected form (e.g., wiigwaas), use this command:
up wiigwaas
The output should be:
wiigwaas+NI+Sg
wiigwaas+NA+ProxSg
b. To get an inflected form from an analysis, use this command:
down wiigwaas+NI+Sg
The output should be:
wiigwaas
You can input as many of these commands as you like. -
Once you're done using the FST, you can exit Foma with:
quitorexit
A call to make check will test the generated FST. As outlined in the Test Results section of this README, three sets of tests are run: OPD tests, paradigm tests, and corpus tests.
The OPD and paradigm tests are conducted using testing infrastructure in FSTmorph, based on giella-core. Detailed information about the results of the (most recent run of) tests can be found in the generated log files: FST/opd-test.log and FST/paradigm-test.log. A CSV summarizing the results of the tests over time is also generated: FST/opd_noun_test_summary.csv etc. In the CSV, each row corresponds to a new run of the tests (though it will not add duplicate rows, i.e., no new rows will get printed if the test results and date have not changed).
The corpus tests are specific to this FST and are thus not run via FSTmorph -- the relevant code is all in scripts/ here in OjibweMorph. Similarly to the other tests, you can view details about the most recent test results in FST/corpus_test.txt and a summary of results over time in FST/corpus_test_summary.csv.
The XSpreadsheets/ directories contain CSVs with example forms for the various paradigm and class categories within each part-of-speech (POS) category. The NounSpreadsheets/, PVSpreadsheets/, and VerbSpreadsheets/ directories all contain detailed READMEs discussing their respective CSV contents, which also provides quite a lit of linguistic information similar to what one might find in a grammar. There is also detailed documentation of the phonological rules. Here are links to each one:
The directories config/, xsft/, and templates contain additional Ojibwe-specific files used to create the FST.
Unless otherwise indicated, the work and content within this repository is copyrighted by The Experimental Linguistics and Fieldwork Lab (ELF-Lab; https://github.com/ELF-Lab) at The University of British Columbia (UBC) in collaboration with the Alberta Language Technology Lab (ALT-Lab; https://altlab.ualberta.ca/) at the University of Alberta (UofA) and the Ojibwe People's Dictionary (OPD; http://ojibwe.lib.umn.edu) at the University of Minnesota (UofM), as well as various other organizations, unless otherwise attributed.
Unless otherwise indicated, this repository and its contents are copyrighted under the Creative Commons Attribution NonCommercial-ShareAlike 4.0 International License (https://creativecommons.org/licenses/by-nc-sa/4.0/). This means you are free to share the materials (copy and redistribute the material in any medium or format) and adapt the materials (remix, transform, and build upon the material) within this repository under the following conditions:
Attribution: You must give appropriate credit when using this material, indicate changes that were made to the original material, and include a statement such as:
"Copyrighted under the Creative Commons Attribution NonCommercial-ShareAlike 4.0 International License (https://creativecommons.org/licenses/by-nc-sa/4.0/) by The Experimental Linguistics and Fieldwork Lab (ELF-Lab; https://github.com/ELF-Lab) at The University of British Columbia (UBC) in collaboration with the Alberta Language Technology Lab (ALT-Lab; https://altlab.ualberta.ca/) at the University of Alberta (UofA) and the Ojibwe People's Dictionary (OPD; http://ojibwe.lib.umn.edu) at the University of Minnesota (UofM)."
Your attribution should include the above links, and should not in any way that suggest that ELF-Lab, UBC, ALT-Lab, UofA, the OPD, or the UofM endorses you or your use of the work unless otherwise indicated with a written endorsement.
Noncommercial: You may not use any material for commercial purposes unless otherwise authorized by ELF-Lab.
Share Alike: If you remix, transform, or build upon our materials, you may distribute the resulting work only under the same the same Creative Commons BY-NC-SA license, and include a link to the license.
Several people and organizations have (directly or indirectly) contributed code, advice, tools and/or materials to this project. We extend our sincerest gratitude for the help!
- Antti Arppe
- Chris Hammerly
- Ogimaawigwanebiik Nancy Jones
- Nora Livesay
- Minh Nguyen
- John Nichols
- Scott Parkhill
- Sandra Radic
- Miikka Silfverberg
- Anna Stacey
- Reed Steiner
This work was supported by a SSHRC Insight Grant (435-2023-0474) awarded to Hammerly, Arppe, and Silfverberg and a SSHRC Partnership Grant (895-2019-1012) to Arppe, Silfverberg and Hammerly (among others).
To cite this work or the contents of the repository (including, but not limited to, datasets, tables, explanations, methods, analysis, structure, etc) in an academic work, please use the following: