Skip to content
KAIST 18S CS570 Term Project: Automatic Classification of Poetry by Neural Scansion
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
classification
data
parsers
tagger
.gitignore
README.md
Team13_Final Paper_Automatic Classification of Poetry by Neural Scansion.pdf

README.md

Automatic_Classification_of_Poetry_by_Neural_Scansion

CS570 Term Project: Automatic Classification of Poetry by Neural Scansion

Detailed Report

Base Paper

"Automatic Classification of Poetry by Meter and Rhyme"

Proceedings of AAAI 2016

University of Ottawa

Chris Tanasescu, Bryan Paget, and Diana Inkpen

Our Model

End-to-End model of

poem crawler

from https://www.poetryfoundation.org/

Syllable Extraction

Model: Character-level Bi-LSTM-CRF model

Dataset: WebCelex (160,595 words with syllable structures)

For poetryfoundation poem data (to extract syllables for ultimate goal)

34,324 words are included in WebCelex <- Known words

25,729 words are not included (need to extract by our model) <- Unknown words

Trainset 154,595 words(including all known words), Devset 3,000, Testset 3,000 words

Meter(강세) Generation

Model: Character-level Bi-LSTM-CRF model

Dataset: For Better For Verse (4B4V) (87 poems, 1,187 lines with syllable, meter features)

Trainset 951 lines, Devset 118, Testset 118 lines

Include special characters (?, !, ‘, .) those might have information for meters

By using syllables, we lose the word structures

add Word-Boundaries (*) to syllables (ex. And | sings | a | me | lan | cho | ly | strain )

Also the baseline model by "Automatic Classification of Poetry by Meter and Rhyme"

Results and Evaluations

in Detailed Report

You can’t perform that action at this time.