Towards Understanding the Robustness of LLM-based Evaluations under Perturbations

Traditional evaluation metrics like BLEU and ROUGE fall short when capturing the nuanced qualities of generated text, particularly when there is no single ground truth. In this paper, we explore the potential of Large Language Models (LLMs), specifically Google Gemini 1, to serve as automatic evaluators for non-standardized metrics in summarization and dialog-based tasks. We conduct experiments across multiple prompting strategies to examine how LLMs fare as quality evaluators when compared with human judgments on the SummEval and USR datasets, asking the model to generate both a score as well as a justification for the score. Furthermore, we explore the robustness of the LLM evaluator by using perturbed inputs. Our findings suggest that while LLMs show promise, their alignment with human evaluators is limited, they are not robust against perturbations and significant improvements are required for their standalone use as reliable evaluators for subjective metrics.

iREL at SemEval-2024 Task 9: Improving Conventional Prompting Methods for Brain Teasers

This paper describes our approach for SemEval-2024 Task 9: BRAINTEASER: A Novel Task Defying Common Sense. The BRAINTEASER task comprises multiple-choice Question Answering designed to evaluate the models' lateral thinking capabilities. It consists of Sentence Puzzle and Word Puzzle subtasks that require models to defy default common-sense associations and exhibit unconventional thinking. We propose a unique strategy to improve the performance of pre-trained language models, notably the Gemini 1.0 Pro Model, in both subtasks. We employ static and dynamic few-shot prompting techniques and introduce a model-generated reasoning strategy that utilizes the LLM's reasoning capabilities to improve performance. Our approach demonstrated significant improvements, showing that it performed better than the baseline models by a considerable margin but fell short of performing as well as the human annotators, thus highlighting the efficacy of the proposed strategies.

BrainStorm @ iREL at SMM4H 2024: Leveraging Translation and Topical Embeddings for Annotation Detection in Tweets

The proliferation of LLMs in various NLP tasks has sparked debates regarding their reliability, particularly in annotation tasks where biases and hallucinations may arise. In this shared task, we address the challenge of distinguishing annotations made by LLMs from those made by human domain experts in the context of COVID-19 symptom detection from tweets in Latin American Spanish. This paper presents BrainStorm @ iREL's approach to the SMM4H 2024 Shared Task, leveraging the inherent topical information in tweets, we propose a novel approach to identify and classify annotations, aiming to enhance the trustworthiness of annotated data.

MetaCheckGPT -- A Multi-task Hallucination Detector Using LLM Uncertainty and Meta-models

Hallucinations in large language models (LLMs) have recently become a significant problem. A recent effort in this direction is a shared task at Semeval 2024 Task 6, SHROOM, a Shared-task on Hallucinations and Related Observable Overgeneration Mistakes. This paper describes our winning solution ranked 1st and 2nd in the 2 sub-tasks of model agnostic and model aware tracks respectively. We propose a meta-regressor framework of LLMs for model evaluation and integration that achieves the highest scores on the leaderboard. We also experiment with various transformer-based models and black box methods like ChatGPT, Vectara, and others. In addition, we perform an error analysis comparing GPT4 against our best model which shows the limitations of the former.

Can LLMs Generate Architectural Design Decisions? -An Exploratory Empirical study

Architectural Knowledge Management (AKM) involves the organized handling of information related to architectural decisions and design within a project or organization. An essential artifact of AKM is the Architecture Decision Records (ADR), which documents key design decisions. ADRs are documents that capture decision context, decision made and various aspects related to a design decision, thereby promoting transparency, collaboration, and understanding. Despite their benefits, ADR adoption in software development has been slow due to challenges like time constraints and inconsistent uptake. Recent advancements in Large Language Models (LLMs) may help bridge this adoption gap by facilitating ADR generation. However, the effectiveness of LLM for ADR generation or understanding is something that has not been explored. To this end, in this work, we perform an exploratory study that aims to investigate the feasibility of using LLM for the generation of ADRs given the decision context. In our exploratory study, we utilize GPT and T5-based models with 0-shot, few-shot, and fine-tuning approaches to generate the Decision of an ADR given its Context. Our results indicate that in a 0-shot setting, state-of-the-art models such as GPT-4 generate relevant and accurate Design Decisions, although they fall short of human-level performance. Additionally, we observe that more cost-effective models like GPT-3.5 can achieve similar outcomes in a few-shot setting, and smaller models such as Flan-T5 can yield comparable results after fine-tuning. To conclude, this exploratory study suggests that LLM can generate Design Decisions, but further research is required to attain human-level generation and establish standardized widespread adoption.

Multilingual Bias Detection and Mitigation for Indian Languages

Lack of diverse perspectives causes neutrality bias in Wikipedia content leading to millions of worldwide readers getting exposed by potentially inaccurate information. Hence, neutrality bias detection and mitigation is a critical problem. Although previous studies have proposed effective solutions for English, no work exists for Indian languages. First, we contribute two large datasets, mWikiBias and mWNC, covering 8 languages, for the bias detection and mitigation tasks respectively. Next, we investigate the effectiveness of popular multilingual Transformer-based models for the two tasks by modeling detection as a binary classification problem and mitigation as a style transfer problem. We make the code and data publicly available.

Neural models for Factual Inconsistency Classification with Explanations

Factual consistency is one of the most important requirements when editing high quality documents. It is extremely important for automatic text generation systems like summarization, question answering, dialog modeling, and language modeling. Still, automated factual inconsistency detection is rather under-studied. Existing work has focused on (a) finding fake news keeping a knowledge base in context, or (b) detecting broad contradiction (as part of natural language inference literature). However, there has been no work on detecting and explaining types of factual inconsistencies in text, without any knowledge base in context. In this paper, we leverage existing work in linguistics to formally define five types of factual inconsistencies. Based on this categorization, we contribute a novel dataset, FICLE (Factual Inconsistency CLassification with Explanation), with ~8K samples where each sample consists of two sentences (claim and context) annotated with type and span of inconsistency. When the inconsistency relates to an entity type, it is labeled as well at two levels (coarse and fine-grained). Further, we leverage this dataset to train a pipeline of four neural models to predict inconsistency type with explanations, given a (claim, context) sentence pair. Explanations include inconsistent claim fact triple, inconsistent context span, inconsistent claim component, coarse and fine-grained inconsistent entity types. The proposed system first predicts inconsistent spans from claim and context; and then uses them to predict inconsistency types and inconsistent entity types (when inconsistency is due to entities). We experiment with multiple Transformer-based natural language classification as well as generative models, and find that DeBERTa performs the best. Our proposed methods provide a weighted F1 of ~87% for inconsistency type classification across the five classes.

LLM-RM at SemEval-2023 Task 2: Multilingual Complex NER using XLM-RoBERTa

Named Entity Recognition(NER) is a task of recognizing entities at a token level in a sentence. This paper focuses on solving NER tasks in a multilingual setting for complex named entities. Our team, LLM-RM participated in the recently organized SemEval 2023 task, Task 2: MultiCoNER II,Multilingual Complex Named Entity Recognition. We approach the problem by leveraging cross-lingual representation provided by fine-tuning XLM-Roberta base model on datasets of all of the 12 languages provided -- Bangla, Chinese, English, Farsi, French, German, Hindi, Italian, Portuguese, Spanish, Swedish and Ukrainian

Summarizing Indian Languages using Multilingual Transformers based Models

With the advent of multilingual models like mBART, mT5, IndicBART etc., summarization in low resource Indian languages is getting a lot of attention now a days. But still the number of datasets is low in number. In this work, we (Team HakunaMatata) study how these multilingual models perform on the datasets which have Indian languages as source and target text while performing summarization. We experimented with IndicBART and mT5 models to perform the experiments and report the ROUGE-1, ROUGE-2, ROUGE-3 and ROUGE-4 scores as a performance metric.

XWikiGen: Cross-lingual Summarization for Encyclopedic Text Generation in Low Resource Languages

Lack of encyclopedic text contributors, especially on Wikipedia, makes automated text generation for \emph{low resource (LR) languages} a critical problem. Existing work on Wikipedia text generation has focused on \emph{English only} where English reference articles are summarized to generate English Wikipedia pages. But, for low-resource languages, the scarcity of reference articles makes monolingual summarization ineffective in solving this problem. Hence, in this work, we propose \task{}, which is the task of cross-lingual multi-document summarization of text from multiple reference articles, written in various languages, to generate Wikipedia-style text. Accordingly, we contribute a benchmark dataset, \data{}, spanning $\sim$69K Wikipedia articles covering five domains and eight languages. We harness this dataset to train a two-stage system where the input is a set of citations and a section title and the output is a section-specific LR summary. The proposed system is based on a novel idea of neural unsupervised extractive summarization to coarsely identify salient information followed by a neural abstractive model to generate the section-specific text. Extensive experiments show that multi-domain training is better than the multi-lingual setup on average.