Milestones
List view
We have been getting requests to document the glosser under FieldDB, which is actually a morpheme segmenter from one of our student papers. To get this show on the road, we decided to prepare the segmenter for a conference presentation... In this milestone * Discuss the context of low resource morpheme segmentation * Discuss previous work and approaches to this problem * Discuss the implementation of our algorithm * Discuss the evaluation of our algorithm using known data from the morpho challenge as benchmarks Future * Evaluate our algorithm in the context of Glossing in FieldDB * Evaluate our algorithm in combination with active learning in the context of Glossing in FieldDB
No due date•1/4 issues closedCreate a WordPress plugin to test MorphoParser in the context of generating a word cloud, per post and per tag.
No due dateCreate a media wiki plugin to test MorphoParser in the task of indexing in fieldlinguistics' app's wiki.
No due dateModify Glossy Chrome to provide a user interface to the MorphoParser to test MorphoParser assistance in the context of the fieldlinguistics web widgets.
No due dateSet up the restful api using Node.js for * stemming * training * setting parameters (range) manually * returning potential parses
No due dateIterate between Dense Corpus and Precedence Relations until "learned." This milestone will use compression gain to stop the iteration, and consider the precedence relations learned.
No due dateUsing the dense corpus created previously, extract new precedence relations and new morphemes.
No due dateUsing the finite state machine generated in the previous milestone, write a function which generates a dense corpus.
No due dateUsing the ranked list of possible morphemes, design a simple function to turn high confidence morphemes into seeds. Create a UI in HTML which lets a human see the possible morphs, go online or work with an informant to get the morphs in context to confirm that they are indeed morphemes, then create a seed list. Output: * Finite State Machine using seed morphemes
No due dateUsing the wordlist generated in the previous iteration, extract a list of possible morphemes. To be used directly, or to be evaluated by a human (informant or fieldlinguist) to arrive at the seed list. This step provides a hook for human contribution.
No due dateUsing the wordlist created in the previous component, extract parameters such as * degree of agglutination * potential morpheme length distribution (gamma, and mean) * degree of prefixing * degree of suffixing * degree of vowel harmony * degree of allophonic writing system * phonotactic ontology These paramaters will be used/learned to iterate the model towards termination.
No due dateThis component takes in a corpus, and returns a list of words, sorted by rhyming frequency.
No due date•2/4 issues closed