Simulating the Emergence of Dependency Length Minimization Preferences

Introduction

Portions of this ongoing work have been published as Endowing Neural Language Learners with Human-like Biases: A Case Study on Dependency Length Minimization at LREC-COLING 2024 and presented as Neural-agent Language Learning and Communication: Emergence of Dependency Length Minimization at CogSci 2024 Posters.

The implementation is based on the NeLLCom framework and EGG toolkit.

Running experiments

1. Installing EGG toolkit;
2. Moving to the EGG game design folder:

   cd EGG/egg/zoo
   

3. Cloning the current repo into the EGG game design folder:

   git clone https://github.com/yuqing0304/DLM_exp.git
   cd DLM_exp
   

4. Then, we can run a game, for example, communicating with an impatient listener (Impa) with the RNN architecture (rnn) using the verb-final subject-modified language (finalSM) of half meaning space (half):

   cd DLM_exp_main2/DLM_halffinalSMrnnImpa
   sbatch run.sh

Explanations for parameters

speaker_hidden_size/listener_hidden_size: Size of the hidden layers in the speaker/listener networks.
meaning_embedding_dim/listener_embedding_size: Embedding size in the speaker/listener networks. 
word_dropout_p: Word dropout rate for the input, interpreted as noise. 

Citation

If you find this study useful in your research, please cite this paper:

@inproceedings{zhang-etal-2024-endowing,
    title = "Endowing Neural Language Learners with Human-like Biases: A Case Study on Dependency Length Minimization",
    author = "Zhang, Yuqing  and
      Verhoef, Tessa  and
      van Noord, Gertjan  and
      Bisazza, Arianna",
    editor = "Calzolari, Nicoletta  and
      Kan, Min-Yen  and
      Hoste, Veronique  and
      Lenci, Alessandro  and
      Sakti, Sakriani  and
      Xue, Nianwen",
    booktitle = "Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024)",
    month = may,
    year = "2024",
    address = "Torino, Italia",
    publisher = "ELRA and ICCL",
    url = "https://aclanthology.org/2024.lrec-main.516/",
    pages = "5819--5832"
}