Selected Papers

FDSHyper

Distributional Inclusion Hypothesis and Quantifications: Probing for Hypernymy in Functinoal Distributional Semantics

Chun Hei Lo, Wai Lam, Hong Cheng, and Guy Emerson

In ACL 2024; Area Chair Award

TL;DR: This work reveals that Functional Distributional Semantics (FDS) captures hypernymy if trained on a corpus that follows the Distributional Inclusion Hypothesis (DIH). We further propose an alternative FDS training objective that handles simple universal quantifications, which enables hypernymy learning under the reverse of the DIH and is shown to improve hypernymy detection from corpora.

Paper Slides Poster

TCSFromDMRS

Functional Distributional Semantics at Scale

Chun Hei Lo, Hong Cheng, Wai Lam, and Guy Emerson

In *SEM 2023

TL;DR: Functional Distributional Semantics (FDS) attempts to learn truth-conditional meanings of words from distributional information in a corpus. This work extends the applicability of FDS to sentences with more complex structures.

Paper Slides Poster Code

pshrgOnDMRS

Semantic Composition with PSHRG for Derivation Tree Reconstruction from Graph-Based Meaning Representations

Chun Hei Lo, Wai Lam, and Hong Cheng

In ACL 2022

TL;DR: This work explores the use of a formal graph grammar in approximating the composition of meaning representation graphs and recovering their synatctic derivations. Surface realization becomes more explainable with the syntax trees.

Paper Slides Video

DSTC9

Task-Oriented Dialogue System with Unstructured Knowledge Access

In arXiv

TL;DR: This is a joint work with the CUHK team in participating the Ninth Dialog System Technology Challenge (DSTC9). We developed a dialog system using retrieval-augmented generation. I am responsbile for developing the retrieval algorithm of knowledge snippets that best answer the user's query given the dialogue history. We ranked 12 out of the 24 participating teams. Visit the links below to see how we appproached it!

Task Paper Code