Chinese Poetry Generation with Recurrent Neural Networks
This project includes the code/model for the paper
Chinese Poetry Generation with Recurrent Neural Networks
@InProceedings{zhang-lapata:2014:EMNLP2014,
author = {Zhang, Xingxing and Lapata, Mirella},
title = {Chinese Poetry Generation with Recurrent Neural Networks},
booktitle = {Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP)},
month = {October},
year = {2014},
address = {Doha, Qatar},
publisher = {Association for Computational Linguistics},
pages = {670--680},
url = {http://www.aclweb.org/anthology/D14-1074}
}
Acknowledgement
Our implementation is greatly inspired by Tomas Mikolov's rnnlm toolkit. We would like to thank Tomas Mikolov for making his code public available.
Dataset
Download the complete dataset from here
Dependencies
- KenLM
- G++ (4.4.7)
- Java (1.8.0_51, 1.6 or 1.7 should also be fine)
- Python (2.7)
Installation
- Install KenLM. Also remember to add kenlm to your LD_LIBRARY_PATH
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/afs/inf.ed.ac.uk/user/s12/s1270921/usr/kenlm/lib
-
Go to rnnpg folder and modify the Makefile (see below). Direct
INCLUDESandLDFLAGSto your KenLM library. Also modify the Makefiles in rnnpg-decoder and rnnpg-generator. -
Make everything by ./INSTALL.sh
OUT_EXEC = rnnpg
OBJS = $(patsubst %.cpp, %.o, $(wildcard *.cpp))
CC = g++
CPPFLAGS = -O3 -funroll-loops -ffast-math -finline-functions -Wall -Winline -pipe -DKENLM_MAX_ORDER=6
INCLUDES = -I/afs/inf.ed.ac.uk/user/s12/s1270921/usr/kenlm
LDFLAGS = -L/afs/inf.ed.ac.uk/user/s12/s1270921/usr/kenlm/lib -lkenlm
all : $(OUT_EXEC)
rm *.o
$(OUT_EXEC) : $(OBJS)
$(CC) -o $@ $^ $(INCLUDES) $(LDFLAGS)
%.o : %.cpp
$(CC) $(CPPFLAGS) -c $< -o $@ $(INCLUDES)
clean:
rm -f *.o
rm -f $(OUT_EXEC)
Run Experiments
Download data/model from here
# move MISC.tar.bz2 to the root folder of this project, then
tar jxvf MISC.tar.bz2
1. Perplexity
cd experiments
./ppl.sh
2. Generation
cd experiments
./generation.sh
Enjoy the generated poems!
3. BLEU
Download from here
tar jxvf BLEU2-final.tar.bz2
cd BLEU2-final
cd MERT_channel-1_RNN-CB-POS-LM-Eval-BLEU2
python showBLEU.py .