Overview of my work on semantic graphs

This is an overview over the code/data associated with my dissertation.

MR tools

MR metrics & MR-to-MR alignment

Structural matching

• Smatch++, 📜: optimal and standardized Smatch
• Smaragd, 📜: fast approximated Smatch with neural network

Advanced matching beyond structure

• S2match, 📜: Smatch matching with node labels matched via word embeddings
• WWLK, 📜: Weisfeiler Leman MR Kernel with embeddings for contextualied matching, many-to-many node alignment

Graph processing, reading, writing

• Smatch++, 📜: can read and modify graphs, standardization, sub-graph extraction

MR-to-text alignment

• WWLK-aligner, 📜: Easy-to-use Wasserstein amr-text alignment

MR metric evaluation

• BAMBOO, 📜: Live BAMBOO benchmark for MR metric comparison

MR applications

• Explainable sentence embeddings (S3BERT), 📜: Infusing fine-grained similarity into a state-of-the-art sentence embedding model for explainable and decomposable embeddings
• Explainable NLG evaluation, 📜: Evaluation of NLG systems in meaning space
• AMR4NLI, 📜: Asymmetric and unsupervised application of MR metrics for transparent entailment rating
• Textual argument similarity, 📜

Neural graph encoder

• Penman-informed CNN, 📜: Simple and efficient neural graph encoding of graphs such as AMRs, dependency trees, etc.

Corpora of (A)MR annotations

• NLI AMR, 📜: >1 mio Silver AMR pairs of five NLI data sets
• ParsEval, 📜: 800 Parsed AMRs with human quality annoations (domain: little Prince, AMR3)
• Textual Similarity, 📜: Silver AMRs of text similarity benchmarks (e.g., STS) that are annoated with human textual similarity ratings
• Textual argument similarity, 📜: Silver AMRs of textual argument similarity that are annoated with human textual similarity ratings