Reading: Slangvolution: A Causal Analysis of Semantic Change and Frequency Dynamics in Slang

[a1da4]

0. Paper

• authors: Daphna Keidar, Andreas Opedal, Zhijing Jin, Mrinmaya Sachan
• paper: arxiv

1. What is it?

They analyze the relationship between semantic shift and frequency shift of slang words.

2. What is amazing compared to previous works?

In the analysis, they use the Causal Directed Acyclic Graphs (DAGs)

3. Where is the key to technologies and techniques?

To construct DAGs, they used following features:

• word type: slang or not

• word frequency

• frequency shift:

  

• polysemy

• part of speech

• semantic change: obtain word embeddings from RoBERTa and calculate Average Pairwise Distances:

  • cosine distance: d_cos [0, 2]
  • Euclidean distance: d_2 [0, inf]
  • combine (proposal): [0, 1]

4. How did evaluate it?

From the above DAG,

• word type (slang or not) has a direct effect on the semantic shift and the frequency shift
• PoS does not have any effect on semantic shift
• word polysemy has a direct effect on word frequency

5. Is there a discussion?

6. Which paper should read next?