Papers by Abhilasha Ravichander

20 papers
Inference-Time Policy Adapters (IPA): Tailoring Extreme-Scale LMs without Fine-tuning (2023.emnlp-main)

Copied to clipboard

Challenge: Extreme-scale language models have shown exceptional performance on a variety of language tasks, but the degree of control offered by these models through pure prompting is limited.
Approach: They propose an inference-time policy adapter which tailors a large base model without fine-tuning it.
Outcome: The proposed model outperforms baseline methods on five challenging text generation tasks and even over GPT-4.
Measuring and Improving Consistency in Pretrained Language Models (2021.tacl-1)

Copied to clipboard

Challenge: In this paper, we examine whether pretrained language models are consistent with factual knowledge.
Approach: They propose a method to improve consistency of pretrained language models . consistency is a desirable property of a good language understanding model, they argue .
Outcome: The proposed model improves consistency and shows that it is effective.
Agent Lumos: Unified and Modular Training for Open-Source Language Agents (2024.acl-long)

Copied to clipboard

Challenge: Lumos is a framework for training open-source agents on complex interactive tasks.
Approach: They propose a framework for training open-source LLM-based agents called Lumos . Lumos features a learnable, unified and modular architecture with a planning module that learns high-level subgoal generation and a grounding module trained to translate these into the actions using various tools in the execution module.
Outcome: The framework outperforms open-source agents on QA and web tasks.
Reverse Question Answering: Can an LLM Write a Question so Hard (or Bad) that it Can’t Answer? (2025.naacl-short)

Copied to clipboard

Challenge: Question answering (QA) is a popular task, but we test both separately . a recent study found that LLMs are less accurate in numerical RQA than RQA .
Approach: We run 16 LLMs on QA and RQA with trivia questions/answers . they find question and answer types that lead to RQA errors and suggest improvements .
Outcome: The results show that LLMs are less accurate in RQA for numerical answers than RQA . RQA errors correlate with question difficulty and inversely correlate with answer frequencies .
Exploring Numeracy in Word Embeddings (P19-1)

Copied to clipboard

Challenge: Existing word embeddings are inadequate at capturing numerical properties of numbers.
Approach: They propose to use word embeddings to capture numerical properties of numbers . they hope to develop methods which better capture numeric properties .
Outcome: The proposed models lack the ability to capture numeric properties of numbers, the authors show . their findings provide a starting point for the development of better models .
Question Answering for Privacy Policies: Combining Computational and Legal Perspectives (D19-1)

Copied to clipboard

Challenge: Privacy policies are long and complex documents that are difficult for users to read and understand.
Approach: They present a corpus of 1750 questions about privacy policies of mobile applications and over 3500 expert annotations of relevant answers.
Outcome: The proposed corpus of 1750 questions on privacy policies shows that a strong neural baseline underperforms human performance by almost 0.3 F1 on PrivacyQA.
Erratum: Measuring and Improving Consistency in Pretrained Language Models (2021.tacl-1)

Copied to clipboard

Challenge: During production of this paper, an error was introduced to the formula on the bottom of the right column of page 1020.
Approach: the formula was changed in the last two terms of the paper .
Outcome: the correct formula is now available on the web.
On the Systematicity of Probing Contextualized Word Representations: The Case of Hypernymy in BERT (2020.starsem-1)

Copied to clipboard

Challenge: Existing studies have found that BERT can correctly retrieve noun hypernyms in cloze tasks, but this does not correspond to systematic knowledge in BERT.
Approach: They propose to use BERT to probe for hypernymy knowledge encoded in representations for cloze tasks to find out whether it is systematic or not .
Outcome: The proposed model can retrieve hypernyms in cloze tasks, but not systematic knowledge in BERT.
When and Why Does Bias Mitigation Work? (2023.findings-emnlp)

Copied to clipboard

Challenge: Neural models exploit shallow surface features to perform language understanding tasks, rather than learning the deeper language understanding and reasoning skills that practitioners desire.
Approach: They propose to use model debiasing techniques to pressure models away from spurious features and to use them to learn useful representations instead.
Outcome: The proposed methods increase models' reliance on hidden biases instead of learning robust features that help them solve a task.
A Tale of Two Regulatory Regimes: Creation and Analysis of a Bilingual Privacy Policy Corpus (2022.lrec-1)

Copied to clipboard

Challenge: With the introduction of new privacy regulations, disclosures made by the same organization are not always the same in different languages.
Approach: They propose a language annotation scheme to capture nuances of two new privacy regulations, namely the EU’s GDPR and California’s CCPA/CPRA.
Outcome: The proposed method captures the nuances of two new privacy regulations and compares them to a corpus of 64 privacy policies in English and 91 in German with manual annotations for 8K and 19K fine-grained data practices.
CONDAQA: A Contrastive Reading Comprehension Dataset for Reasoning about Negation (2022.emnlp-main)

Copied to clipboard

Challenge: Negation is fundamental to human communication.
Approach: They propose a dataset which requires reasoning about implications of negated statements in paragraphs . they collect paragraphs with diverse negation cues and crowdworkers ask questions about implications .
Outcome: The first dataset in english requires reasoning about implications of negated statements in paragraphs . it features 14,182 question-answer pairs with over 200 unique negation cues based on crowd-workers . the best performing model achieves only 42% on consistency metric, well below human performance of 81%.
Breaking Down Walls of Text: How Can NLP Benefit Consumer Privacy? (2021.acl-long)

Copied to clipboard

Challenge: Privacy policies are long and complex documents that are difficult for users to read and comprehend.
Approach: They propose language technologies to help users reclaim control over their privacy . they highlight many remaining opportunities to develop more precise or nuanced language technologies .
Outcome: The proposed language technologies can address the privacy information gap . they can be more precise or nuanced in the way they use the text of privacy policies.
MacGyver: Are Large Language Models Creative Problem Solvers? (2024.naacl-long)

Copied to clipboard

Challenge: a new study examines the creative problem-solving capabilities of modern LLMs . it provides insight into the constrained problem- solving capabilities of both humans and AI .
Approach: They use an automatically generated dataset to compare and contrast LLMs and humans to find out their creative problem-solving abilities.
Outcome: The proposed dataset compares LLMs and humans in a constrained setting . it shows that humans excel in tasks they are familiar with but struggle with domain-specific knowledge .
HALoGEN: Fantastic LLM Hallucinations and Where to Find Them (2025.acl-long)

Copied to clipboard

Challenge: generative large language models produce hallucinations that are not aligned with world knowledge or input context.
Approach: They propose a hallucination benchmark framework that measures hallucinism in large language models . they evaluate 150,000 generations from 14 language models and find they are riddled with hallucinos .
Outcome: The proposed framework evaluates 150,000 generations from 14 language models.
NoiseQA: Challenge Set Evaluation for User-Centric Question Answering (2021.eacl-main)

Copied to clipboard

Challenge: Question-Answering (QA) systems are deployed in the real world . a lack of research attention has been devoted to studying the issues that arise when people use QA systems.
Approach: They show that component components that precede an answering engine can introduce varied and considerable sources of error.
Outcome: The proposed evaluations highlight the need for QA evaluation to expand to consider real-world use.
Stress Test Evaluation for Natural Language Inference (C18-1)

Copied to clipboard

Challenge: Existing models perform well at standard datasets for NLI, achieving impressive results across different genres of text.
Approach: They propose to use automatic stress tests to evaluate models' ability to make inferential decisions.
Outcome: The proposed model performs well across genres of text, but lacks the ability to make inferential decisions.
What Has Been Lost with Synthetic Evaluation? (2025.findings-emnlp)

Copied to clipboard

Challenge: a recent study evaluated the validity and difficulty of large language models for evaluation benchmarks . large language model evaluation benchmarking is challenging and requires specific phenomena to be addressed .
Approach: They compare LLM-generated reasoning-over-text benchmarks to those generated through crowdsourcing . they find they are *less challenging for LLMs* than their human-authored counterparts .
Outcome: The results show that LLMs can generate variants that are valid according to annotation guidelines, but less challenging than human-authored counterparts.
Information-Guided Identification of Training Data Imprint in (Proprietary) Large Language Models (2025.naacl-long)

Copied to clipboard

Challenge: Lack of transparency in training data is limiting external oversight and inspection of LLMs for issues such as copyright infringement and data contamination.
Approach: They propose a method to identify training data known to proprietary LLMs without requiring access to model weights or token probabilities by using information-guided probes.
Outcome: The proposed method can identify training data known to proprietary LLMs without access to model weights or token probabilities.
Probing the Probing Paradigm: Does Probing Accuracy Entail Task Relevance? (2021.eacl-main)

Copied to clipboard

Challenge: Neural models have established state-of-the-art performance on several NLP benchmarks, but little is understood about the mechanisms by which they operate.
Approach: They examine the probing paradigm through a set of controlled synthetic tasks and show that pretrained word embeddings play a considerable role in encoding these properties rather than the training task itself.
Outcome: The proposed model can encode linguistic properties above chance-level even when distributed in the data as random noise, reversing the interpretation of absolute claims on probing tasks.
Artifacts or Abduction: How Do LLMs Answer Multiple-Choice Questions Without the Question? (2024.acl-long)

Copied to clipboard

Challenge: Multiple-choice question answering (MCQA) is often used to evaluate large language models . a recent study found that LLMs perform MCQA with choices-only prompts .
Approach: They investigate whether LLMs can perform multiple-choice question answering (MCQA) with choices-only prompts . they find no evidence that the choices- only accuracy stems from memorization alone .
Outcome: The results show that LLMs perform MCQA with choices-only prompts with 0.33 accuracy gain.

What is GenGO?

GenGO is an NLP powered publication search system. It currenctly indexes 30k+ papers from ACL Anthology, and implements multi-aspect summarization, semantic search, and more!

Information

About
Limitations