Australasian Language Technology Association

Professor Steven Bird of Charles Darwin University gave a challenging, incisive, resonant and deeply heartfelt keynote at #ALTA2024 at The Australian National University last week - where he contested the prevailing epistemic value of #LLMs, and showed how they contribute to the #metacrisis - the intertwined crises of truth, reality, justice and environment. A must-watch for anyone in #NLP.

In Canberra this week, to present our paper "Generating bilingual example sentences with LLMs". It won best paper award at the Australasian Language Technology Association (ALTA) conference! Supervised by Katerina Vylomova and Kemal Kurniawan. TLDR: (1) We work with French / Indonesian / Tetun <> English, and generate examples with GPT-4o and Llama 3.1. (2) Humans disagree with each other on what makes a "good" dictionary example but LLMs can be trained to align with a specific human (3) LLMs, including open ones (Llama 3.1 405b), can generate pretty good example sentences, but (as expected) their perf degrades with language resource level. Also summer in Canberra is very nice! Very Mediterranean. Paper at https://lnkd.in/gSNRncnk

During closing remarks, #ALTA2024 Conference General Chair, Dr Gabriela Ferraro from the ANU School of Cybernetics gives thanks to her conference team: Program Chairs: Tim Baldwin of MBZUAI (Mohamed bin Zayed University of Artificial Intelligence), Nicholas I-Hsien Kuo of UNSW, Sergio José (羅士豪) Rodríguez Méndez of ANU College of Engineering, Computing & Cybernetics Publication Chair: Anton Malko, ANU College of Arts & Social Sciences Technology Chair: Dawei Chen, ANU College of Engineering, Computing & Cybernetics Finance Chair: Ishihara Shunichi, ANU College of Arts & Social Sciences Sponsorship Chair: Charbel El-Khaissi, ANU College of Arts & Social Sciences Publicity Chair: Kathy Reid MBA, ANU College of Engineering, Computing & Cybernetics Local Chair: Ned Cooper, ANU College of Engineering, Computing & Cybernetics She passes the baton to the #ALTA2025 team at University of Technology Sydney and we look forward to meeting again next year 🎉

During Closing Remarks at #ALTA2024, Conference Chair, Dr Gabriela Ferraro of the ANU School of Cybernetics extends gratitude on behalf of the Australasian Language Technology Association to the student volunteers, from many disciplines at The Australian National University who have helped to deliver the conference. Please join us in thanking them for their excellent efforts! 👏 Jiacheng Zhang from ANU College of Asia & the Pacific 🌏 Taylor Liu and Daniel M. from ANU College of Arts & Social Sciences 🗣️ Jeevika S., Ke Tan, Vito Rihaldijiran, Zitao Chi, Xuke Du, Praju Ghatpande, Shifan Zhang, Iko-Ojo Simon from ANU College of Engineering, Computing & Cybernetics 💻

Minghan Wang from Monash University is presenting his long paper today at #ALTA2024 titled "Simultaneous Machine Translation with Large Language Models" online. ➡️ This paper investigates the possibility of applying Large Language Models (#LLM) to #SimulMT tasks by using existing incremental-decoding methods with a newly proposed #RALCP algorithm for latency reduction. ➡️ They conducted experiments using the #Llama2-7b-chat model on nine different languages from the #MUST-C dataset. 📷 Taylor Liu

Milindi Kodikara from RMIT University is presenting the long paper today at #ALTA2024 titled "Lesser the Shots, Higher the #Hallucinations: Exploration of Genetic Information Extraction using Generative Large Language Models". This paper systematically evaluates the performance of generative large language models (#LLMs) on the extraction of specialised genetic information, focusing on end-to-end information extraction encompassing both named entity recognition #NER and relation extraction. 📷 Taylor Liu #NLP

Kritesh Rauniyar from Delhi Technological University (Formerly DCE) is presenting their paper today at #ALTA2024 on (joint work, primary author Surendrabikram Thapa of Virginia Tech) "Which Side Are You On? Investigating Politico-Economic #Bias in #Nepali Language Models" online. This paper addresses this gap by investigating the political and economic biases present in five fill-mask models and eleven generative models trained for the Nepali language. 📷 Taylor Liu

In our session on ethical considerations and social media analysis at #ALTA2024 today, Daniel Cabrera Lozoya from University of Melbourne is presenting his paper titled "Truth in the Noise: Unveiling Authentic #Dementia Self-Disclosure Statements in Social Media with #LLMs" This study implemented a genetic algorithm that evolves prompts using various state-of-the-art prompt engineering techniques, including chain of thought, self-critique, generated knowledge, and expert prompting. 📷 Taylor Liu

Minkang Liu from Universiti Sains Malaysia presenting his paper today at #ALTA2024, titled "#GenABSA-Vec: Generative Aspect-Based #Sentiment Feature #Vectorization for Document-Level Sentiment Classification". This paper proposes a method to construct a #GenABSA feature vector containing five aspect-sentiment scores to represent each review document. #NLP #classification #NLProcessing 📷 Taylor Liu

In the #ALTA2024 session on advances in #NLP models and techniques today, Emmanuele Chersoni of The Hong Kong Polytechnic University is presenting joint work (first author Carina Kauf from Massachusetts Institute of Technology) titled "Comparing Plausibility Estimates in Base and Instruction-Tuned Large Language Models " This research aims to explore the effect of the instruction #tuning process on a model’s knowledge of semantic plausibility. And as prompting does not suffer from the confounders of log-likelihoods (e.g. frequency, word length etc.), could it turn out to be a better method for extracting #plausibility knowledge? 📷 Taylor Liu