DENOASR: Debiasing ASRs through Selective Denoising

Automatic Speech Recognition (ASR) systems have been examined and shown to exhibit biases toward particular groups of individuals, influenced by factors such as demographic traits, accents, and speech styles. Noise can disproportionately impact speakers with certain accents, dialects, or speaking styles, leading to biased error rates. In this work, we introduce a novel framework DENOASR, which is a selective denoising technique to reduce the disparity in the word error rates between the two gender groups, male and female. We find that a combination of two popular speech denoising techniques, viz. DEMUCS and LE, can be effectively used to mitigate ASR disparity without compromising their overall performance. Experiments using two state-of-the-art open-source ASRs - OpenAI WHISPER and NVIDIA NEMO - on multiple benchmark datasets, including TIE, VOX-POPULI, TEDLIUM, and FLEURS, show that there is a promising reduction in the average word error rate gap across the two gender groups. For a given dataset, the denoising is selectively applied on speech samples having speech intelligibility below a certain threshold, estimated using a small validation sample, thus ameliorating the need for large-scale human-written ground-truth transcripts. Our findings suggest that selective denoising can be an elegant approach to mitigate biases in present-day ASR systems.

Navigating the Cultural Kaleidoscope: A Hitchhiker's Guide to Sensitivity in Large Language Models

As LLMs are increasingly deployed in global applications, the importance of cultural sensitivity becomes paramount, ensuring that users from diverse backgrounds feel respected and understood. Cultural harm can arise when these models fail to align with specific cultural norms, resulting in misrepresentations or violations of cultural values. This work addresses the challenges of ensuring cultural sensitivity in LLMs, especially in small-parameter models that often lack the extensive training data needed to capture global cultural nuances. We present two key contributions: (1) A cultural harm test dataset, created to assess model outputs across different cultural contexts through scenarios that expose potential cultural insensitivities, and (2) A culturally aligned preference dataset, aimed at restoring cultural sensitivity through fine-tuning based on feedback from diverse annotators. These datasets facilitate the evaluation and enhancement of LLMs, ensuring their ethical and safe deployment across different cultural landscapes. Our results show that integrating culturally aligned feedback leads to a marked improvement in model behavior, significantly reducing the likelihood of generating culturally insensitive or harmful content. Ultimately, this work paves the way for more inclusive and respectful AI systems, fostering a future where LLMs can safely and ethically navigate the complexities of diverse cultural landscapes.

CrowdCounter: A benchmark type-specific multi-target counterspeech dataset

Counterspeech presents a viable alternative to banning or suspending users for hate speech while upholding freedom of expression. However, writing effective counterspeech is challenging for moderators/users. Hence, developing suggestion tools for writing counterspeech is the need of the hour. One critical challenge in developing such a tool is the lack of quality and diversity of the responses in the existing datasets. Hence, we introduce a new dataset - CrowdCounter containing 3,425 hate speech-counterspeech pairs spanning six different counterspeech types (empathy, humor, questioning, warning, shaming, contradiction), which is the first of its kind. The design of our annotation platform itself encourages annotators to write type-specific, non-redundant and high-quality counterspeech. We evaluate two frameworks for generating counterspeech responses - vanilla and type-controlled prompts - across four large language models. In terms of metrics, we evaluate the responses using relevance, diversity and quality. We observe that Flan-T5 is the best model in the vanilla framework across different models. Type-specific prompts enhance the relevance of the responses, although they might reduce the language quality. DialoGPT proves to be the best at following the instructions and generating the type-specific counterspeech accurately.

Breaking the Global North Stereotype: A Global South-centric Benchmark Dataset for Auditing and Mitigating Biases in Facial Recognition Systems

Facial Recognition Systems (FRSs) are being developed and deployed globally at unprecedented rates. Most platforms are designed in a limited set of countries but deployed in worldwide, without adequate checkpoints. This is especially problematic for Global South countries which lack strong legislation to safeguard persons facing disparate performance of these systems. A combination of unavailability of datasets, lack of understanding of FRS functionality and low-resource bias mitigation measures accentuate the problem. In this work, we propose a new face dataset composed of 6,579 unique male and female sportspersons from eight countries around the world. More than 50% of the dataset comprises individuals from the Global South countries and is demographically diverse. To aid adversarial audits and robust model training, each image has four adversarial variants, totaling over 40,000 images. We also benchmark five popular FRSs, both commercial and open-source, for the task of gender prediction (and country prediction for one of the open-source models as an example of red-teaming). Experiments on industrial FRSs reveal accuracies ranging from 98.2%--38.1%, with a large disparity between males and females in the Global South (max difference of 38.5%). Biases are also observed in all FRSs between females of the Global North and South (max difference of ~50%). Grad-CAM analysis identifies the nose, forehead and mouth as the regions of interest on one of the open-source FRSs. Utilizing this insight, we design simple, low-resource bias mitigation solutions using few-shot and novel contrastive learning techniques significantly improving the accuracy with disparity between males and females reducing from 50% to 1.5% in one of the settings. In the red-teaming experiment with the open-source Deepface model, contrastive learning proves more effective than simple fine-tuning.

Sponsored is the New Organic: Implications of Sponsored Results on Quality of Search Results in the Amazon Marketplace

Interleaving sponsored results (advertisements) amongst organic results on search engine result pages (SERP) has become a common practice across multiple digital platforms. Advertisements have catered to consumer satisfaction and fostered competition in digital public spaces; making them an appealing gateway for businesses to reach their consumers. However, especially in the context of digital marketplaces, due to the competitive nature of the sponsored results with the organic ones, multiple unwanted repercussions have surfaced affecting different stakeholders. From the consumers' perspective the sponsored ads/results may cause degradation of search quality and nudge consumers to potentially irrelevant and costlier products. The sponsored ads may also affect the level playing field of the competition in the marketplaces among sellers. To understand and unravel these potential concerns, we analyse the Amazon digital marketplace in four different countries by simulating 4,800 search operations. Our analyses over SERPs consisting 2M organic and 638K sponsored results show items with poor organic ranks (beyond 100th position) appear as sponsored results even before the top organic results on the first page of Amazon SERP. Moreover, we also observe that in majority of the cases, these top sponsored results are costlier and are of poorer quality than the top organic results. We believe these observations can motivate researchers for further deliberation to bring in more transparency and guard rails in the advertising practices followed in digital marketplaces.

Auditing the Grid-Based Placement of Private Label Products on E-commerce Search Result Pages

E-commerce platforms support the needs and livelihoods of their two most important stakeholders -- customers and producers/sellers. Multiple algorithmic systems, like ``search'' systems mediate the interactions between these stakeholders by connecting customers to producers with relevant items. Search results include (i) private label (PL) products that are manufactured/sold by the platform itself, as well as (ii) third-party products on advertised / sponsored and organic positions. In this paper, we systematically quantify the extent of PL product promotion on e-commerce search results for the two largest e-commerce platforms operating in India -- Amazon.in and Flipkart. By analyzing snapshots of search results across the two platforms, we discover high PL promotion on the initial result pages (~ 15% PLs are advertised on the first SERP of Amazon). Both platforms use different strategies to promote their PL products, such as placing more PLs on the advertised positions -- while Amazon places them on the first, middle, and last rows of the search results, Flipkart places them on the first two positions and the (entire) last column of the search results. We discover that these product placement strategies of both platforms conform with existing user attention strategies proposed in the literature. Finally, to supplement the findings from the collected data, we conduct a survey among 68 participants on Amazon Mechanical Turk. The click pattern from our survey shows that users strongly prefer to click on products placed at positions that correspond to the PL products on the search results of Amazon, but not so strongly on Flipkart. The click-through rate follows previously proposed theoretically grounded user attention distribution patterns in a two-dimensional layout.

Investigating Nudges toward Related Sellers on E-commerce Marketplaces: A Case Study on Amazon

E-commerce marketplaces provide business opportunities to millions of sellers worldwide. Some of these sellers have special relationships with the marketplace by virtue of using their subsidiary services (e.g., fulfillment and/or shipping services provided by the marketplace) -- we refer to such sellers collectively as Related Sellers. When multiple sellers offer to sell the same product, the marketplace helps a customer in selecting an offer (by a seller) through (a) a default offer selection algorithm, (b) showing features about each of the offers and the corresponding sellers (price, seller performance metrics, seller's number of ratings etc.), and (c) finally evaluating the sellers along these features. In this paper, we perform an end-to-end investigation into how the above apparatus can nudge customers toward the Related Sellers on Amazon's four different marketplaces in India, USA, Germany and France. We find that given explicit choices, customers' preferred offers and algorithmically selected offers can be significantly different. We highlight that Amazon is adopting different performance metric evaluation policies for different sellers, potentially benefiting Related Sellers. For instance, such policies result in notable discrepancy between the actual performance metric and the presented performance metric of Related Sellers. We further observe that among the seller-centric features visible to customers, sellers' number of ratings influences their decisions the most, yet it may not reflect the true quality of service by the seller, rather reflecting the scale at which the seller operates, thereby implicitly steering customers toward larger Related Sellers. Moreover, when customers are shown the rectified metrics for the different sellers, their preference toward Related Sellers is almost halved.

Demarked: A Strategy for Enhanced Abusive Speech Moderation through Counterspeech, Detoxification, and Message Management

Despite regulations imposed by nations and social media platforms, such as recent EU regulations targeting digital violence, abusive content persists as a significant challenge. Existing approaches primarily rely on binary solutions, such as outright blocking or banning, yet fail to address the complex nature of abusive speech. In this work, we propose a more comprehensive approach called Demarcation scoring abusive speech based on four aspect -- (i) severity scale; (ii) presence of a target; (iii) context scale; (iv) legal scale -- and suggesting more options of actions like detoxification, counter speech generation, blocking, or, as a final measure, human intervention. Through a thorough analysis of abusive speech regulations across diverse jurisdictions, platforms, and research papers we highlight the gap in preventing measures and advocate for tailored proactive steps to combat its multifaceted manifestations. Our work aims to inform future strategies for effectively addressing abusive speech online.

SafeInfer: Context Adaptive Decoding Time Safety Alignment for Large Language Models

Safety-aligned language models often exhibit fragile and imbalanced safety mechanisms, increasing the likelihood of generating unsafe content. In addition, incorporating new knowledge through editing techniques to language models can further compromise safety. To address these issues, we propose SafeInfer, a context-adaptive, decoding-time safety alignment strategy for generating safe responses to user queries. SafeInfer comprises two phases: the safety amplification phase, which employs safe demonstration examples to adjust the model's hidden states and increase the likelihood of safer outputs, and the safety-guided decoding phase, which influences token selection based on safety-optimized distributions, ensuring the generated content complies with ethical guidelines. Further, we present HarmEval, a novel benchmark for extensive safety evaluations, designed to address potential misuse scenarios in accordance with the policies of leading AI tech giants.

Breaking Boundaries: Investigating the Effects of Model Editing on Cross-linguistic Performance

The integration of pretrained language models (PLMs) like BERT and GPT has revolutionized NLP, particularly for English, but it has also created linguistic imbalances. This paper strategically identifies the need for linguistic equity by examining several knowledge editing techniques in multilingual contexts. We evaluate the performance of models such as Mistral, TowerInstruct, OpenHathi, Tamil-Llama, and Kan-Llama across languages including English, German, French, Italian, Spanish, Hindi, Tamil, and Kannada. Our research identifies significant discrepancies in normal and merged models concerning cross-lingual consistency. We employ strategies like 'each language for itself' (ELFI) and 'each language for others' (ELFO) to stress-test these models. Our findings demonstrate the potential for LLMs to overcome linguistic barriers, laying the groundwork for future research in achieving linguistic inclusivity in AI technologies.