Heidelberg-NLP's repositories
discourse-aware-semantic-self-attention
Repository for code and data from the EMNLP-IJCNLP 2019 paper "Discourse-aware Semantic Self-Attention for Narrative Reading Comprehension"
LMs4Implicit-Knowledge-Generation
Code for equipping pretrained language models (BART, GPT-2, XLNet) with commonsense knowledge for generating implicit knowledge statements between two sentences, by (i) finetuning the models on corpora enriched with implicit information; and by (ii) constraining models with key concepts and commonsense knowledge paths connecting them.
xsrl_mbert_aligner
X-SRL Dataset. Including the code for the SRL annotation projection tool and an out-of-the-box word alignment tool based on Multilingual BERT embeddings.
amr-metric-suite
This project collects methods that enhance the comparison between AMR graphs.
CO-NNECT
This repository contains our path generation framework Co-NNECT, in which we combine two models for establishing knowledge relations and paths between concepts from sentences, as a form of explicitation of implicit knowledge: COREC-LM (COmmonsense knowledge RElation Classification using Language Models), a relation classification system that we developed for classifying commonsense knowledge relations; and COMET, a target prediction system developed by Bosselut et al., 2019.
AMRParseEval
Code and data for the paper *Better Smatch = Better Parser? AMR evaluation is not so simple anymore*
CC-SHAP-VLM
Official code implementation for the paper "Do Vision & Language Decoders use Images and Text equally? How Self-consistent are their Explanations?"
counting-probe
Counting dataset for Vision & Language models. Introduced in the paper "Seeing Past Words: Testing the Cross-Modal Capabilities of Pretrained V&L Models". https://arxiv.org/abs/2012.12352
GraphLanguageModels
Code for our paper "Graph Language Models"
simple-xamr
Strong X-AMR baseline with NMT and parser pipeline.
abstract-anaphora-data
Scripts for processing data for resolution of abstract anaphora
Multi-Hop-Knowledge-Paths-Human-Needs
Ranking and Selecting Multi-Hop Knowledge Paths to Better Predict Human Needs
renji_abarai
This repository shows the code and data for our submission to the "Argument Retrieval for Controversial Questions" task at Touché 2023.