Kevin Tang

Harrison Santiago, Joshua Martin, Sarah Moeller et al.

Recent research has highlighted that natural language processing (NLP) systems exhibit a bias against African American speakers. The bias errors are often caused by poor representation of linguistic features unique to African American English (AAE), due to the relatively low probability of occurrence of many such features in training data. We present a workflow to overcome such bias in the case of habitual "be". Habitual "be" is isomorphic, and therefore ambiguous, with other forms of "be" found in both AAE and other varieties of English. This creates a clear challenge for bias in NLP technologies. To overcome the scarcity, we employ a combination of rule-based filters and data augmentation that generate a corpus balanced between habitual and non-habitual instances. With this balanced corpus, we train unbiased machine learning classifiers, as demonstrated on a corpus of AAE transcribed texts, achieving .65 F$_1$ score disambiguating habitual "be".

Yixin Nie, Lin Guan, Zhongyao Ma et al.

This report presents CharacterFlywheel, an iterative flywheel process for improving large language models (LLMs) in production social chat applications across Instagram, WhatsApp, and Messenger. Starting from LLaMA 3.1, we refined models across 15 generations using data from both internal and external real-user traffic. Through continuous deployments from July 2024 to April 2025, we conducted controlled 7-day A/B tests showing consistent engagement improvements: 7 of 8 newly deployed models demonstrated positive lift over the baseline, with the strongest performers achieving up to 8.8% improvement in engagement breadth and 19.4% in engagement depth. We also observed substantial gains in steerability, with instruction following increasing from 59.2% to 84.8% and instruction violations decreasing from 26.6% to 5.8%. We detail the CharacterFlywheel process which integrates data curation, reward modeling to estimate and interpolate the landscape of engagement metrics, supervised fine-tuning (SFT), reinforcement learning (RL), and both offline and online evaluation to ensure reliable progress at each optimization step. We also discuss our methods for overfitting prevention and navigating production dynamics at scale. These contributions advance the scientific rigor and understanding of LLMs in social applications serving millions of users.

Dana Serditova, Kevin Tang

Automatic Speech Recognition (ASR) systems are widely used in everyday communication, education, healthcare, and industry, yet their performance remains uneven across speakers, particularly when dialectal variation diverges from the mainstream accents represented in training data. This study investigates ASR bias through a sociolinguistic analysis of Newcastle English, a regional variety of North-East England that has been shown to challenge current speech recognition technologies. Using spontaneous speech from the Diachronic Electronic Corpus of Tyneside English (DECTE), we evaluate the output of a state-of-the-art commercial ASR system and conduct a fine-grained analysis of more than 3,000 transcription errors. Errors are classified by linguistic domain and examined in relation to social variables including gender, age, and socioeconomic status. In addition, an acoustic case study of selected vowel features demonstrates how gradient phonetic variation contributes directly to misrecognition. The results show that phonological variation accounts for the majority of errors, with recurrent failures linked to dialect-specific features like vowel quality and glottalisation, as well as local vocabulary and non-standard grammatical forms. Error rates also vary across social groups, with higher error frequencies observed for men and for speakers at the extremes of the age spectrum. These findings indicate that ASR errors are not random but socially patterned and can be explained from a sociolinguistic perspective. Thus, the study demonstrates the importance of incorporating sociolinguistic expertise into the evaluation and development of speech technologies and argues that more equitable ASR systems require explicit attention to dialectal variation and community-based speech data.

Yuchao Gu, Yipin Zhou, Bichen Wu et al.

Current diffusion-based video editing primarily focuses on structure-preserved editing by utilizing various dense correspondences to ensure temporal consistency and motion alignment. However, these approaches are often ineffective when the target edit involves a shape change. To embark on video editing with shape change, we explore customized video subject swapping in this work, where we aim to replace the main subject in a source video with a target subject having a distinct identity and potentially different shape. In contrast to previous methods that rely on dense correspondences, we introduce the VideoSwap framework that exploits semantic point correspondences, inspired by our observation that only a small number of semantic points are necessary to align the subject's motion trajectory and modify its shape. We also introduce various user-point interactions (\eg, removing points and dragging points) to address various semantic point correspondence. Extensive experiments demonstrate state-of-the-art video subject swapping results across a variety of real-world videos.

Jeffy Yu, Maximilian Huber, Kevin Tang

This paper investigates the ethical implications of aligning Large Language Models (LLMs) with financial optimization, through the case study of GreedLlama, a model fine-tuned to prioritize economically beneficial outcomes. By comparing GreedLlama's performance in moral reasoning tasks to a base Llama2 model, our results highlight a concerning trend: GreedLlama demonstrates a marked preference for profit over ethical considerations, making morally appropriate decisions at significantly lower rates than the base model in scenarios of both low and high moral ambiguity. In low ambiguity situations, GreedLlama's ethical decisions decreased to 54.4%, compared to the base model's 86.9%, while in high ambiguity contexts, the rate was 47.4% against the base model's 65.1%. These findings emphasize the risks of single-dimensional value alignment in LLMs, underscoring the need for integrating broader ethical values into AI development to ensure decisions are not solely driven by financial incentives. The study calls for a balanced approach to LLM deployment, advocating for the incorporation of ethical considerations in models intended for business applications, particularly in light of the absence of regulatory oversight.

Dana Serditova, Kevin Tang, Jochen Steffens

Automatic Speech Recognition (ASR) systems struggle with regional dialects due to biased training which favours mainstream varieties. While previous research has identified racial, age, and gender biases in ASR, regional bias remains underexamined. This study investigates ASR performance on Newcastle English, a well-documented regional dialect known to be challenging for ASR. A two-stage analysis was conducted: first, a manual error analysis on a subsample identified key phonological, lexical, and morphosyntactic errors behind ASR misrecognitions; second, a case study focused on the systematic analysis of ASR recognition of the regional pronouns ``yous'' and ``wor''. Results show that ASR errors directly correlate with regional dialectal features, while social factors play a lesser role in ASR mismatches. We advocate for greater dialectal diversity in ASR training data and highlight the value of sociolinguistic analysis in diagnosing and addressing regional biases.

Rayyan Merchant, Akhilesh Kakolu Ramarao, Kevin Tang

Despite speaking mutually intelligible varieties of the same language, speakers of Tajik Persian, written in a modified Cyrillic alphabet, cannot read Iranian and Afghan texts written in the Perso-Arabic script. As the vast majority of Persian text on the Internet is written in Perso-Arabic, monolingual Tajik speakers are unable to interface with the Internet in any meaningful way. Due to overwhelming similarity between the formal registers of these dialects and the scarcity of Tajik-Farsi parallel data, machine transliteration has been proposed as more a practical and appropriate solution than machine translation. This paper presents a transformer-based G2P approach to Tajik-Farsi transliteration, achieving chrF++ scores of 58.70 (Farsi to Tajik) and 74.20 (Tajik to Farsi) on novel digraphic datasets, setting a comparable baseline metric for future work. Our results also demonstrate the non-trivial difficulty of this task in both directions. We also provide an overview of the differences between the two scripts and the challenges they present, so as to aid future efforts in Tajik-Farsi transliteration.

Rahul Porwal, Alice Rozet, Pryce Houck et al.

African American English (AAE) presents unique challenges in natural language processing (NLP). This research systematically compares the performance of available NLP models--rule-based, transformer-based, and large language models (LLMs)--capable of identifying key grammatical features of AAE, namely Habitual Be and Multiple Negation. These features were selected for their distinct grammatical complexity and frequency of occurrence. The evaluation involved sentence-level binary classification tasks, using both zero-shot and few-shot strategies. The analysis reveals that while LLMs show promise compared to the baseline, they are influenced by biases such as recency and unrelated features in the text such as formality. This study highlights the necessity for improved model training and architectural adjustments to better accommodate AAE's unique linguistic characteristics. Data and code are available.

Akhilesh Kakolu Ramarao, Kevin Tang, Dinah Baer-Henney

Over the past decade, various studies have addressed how speakers solve the so-called `The Paradigm Cell Filling Problem' (PCFP) \citep{ackerman2009parts} across different languages. The PCFP addresses a fundamental question in morphological processing: how do speakers accurately generate inflected forms of words when presented with incomplete paradigms? This problem is particularly salient when modeling complex inflectional systems. We focus on Spanish verbal paradigms, where certain verbs follow an irregular L-shaped pattern, where the first-person singular present indicative stem matches the stem used throughout the present subjunctive mood. We formulate the problem as a morphological reinflection task. Specifically, we investigate the role of input frequency in the acquisition of regular versus irregular L-shaped patterns in transformer models. By systematically manipulating the input distributions and analyzing model behavior, we reveal four key findings: 1) Models perform better on L-shaped verbs compared to regular verbs, especially in uneven frequency conditions; 2) Robust primacy effects are observed, but no consistent recency effects; 3) Memorization becomes more prominent as the proportion of L-shaped verbs increases; 4) There is a tendency to regularize L-shaped verbs when their consonant alternation pairs are rare or absent in the training data.

Akhilesh Kakolu Ramarao, Kevin Tang, Dinah Baer-Henney

Whether neural networks can serve as cognitive models of morphological learning remains an open question. Recent work has shown that encoder-decoder models can acquire irregular patterns, but evidence that they generalize these patterns like humans is mixed. We investigate this using the Spanish \emph{L-shaped morphome}, where only the first-person singular indicative (e.g., \textit{pongo} `I put') shares its stem with all subjunctive forms (e.g., \textit{ponga, pongas}) despite lacking apparent phonological, semantic, or syntactic motivation. We compare five encoder-decoder transformers varying along two dimensions: sequential vs. position-invariant positional encoding, and atomic vs. decomposed tag representations. Positional encoding proves decisive: position-invariant models recover the correct L-shaped paradigm clustering even when L-shaped verbs are scarce in training, whereas sequential positional encoding models only partially capture the pattern. Yet none of the models productively generalize this pattern to novel forms. Position-invariant models generalize the L-shaped stem across subjunctive cells but fail to extend it to the first-person singular indicative, producing a mood-based generalization rather than the L-shaped morphomic pattern. Humans do the opposite, generalizing preferentially to the first-person singular indicative over subjunctive forms. None of the models reproduce the human pattern, highlighting the gap between statistical pattern reproduction and morphological abstraction.

Rayyan Merchant, Kevin Tang

As a digraphic language, the Persian language utilizes two written standards: Perso-Arabic in Afghanistan and Iran, and Tajik-Cyrillic in Tajikistan. Despite the significant similarity between the dialects of each country, script differences prevent simple one-to-one mapping, hindering written communication and interaction between Tajikistan and its Persian-speaking ``siblings''. To overcome this, previously-published efforts have investigated machine transliteration models to convert between the two scripts. Unfortunately, most efforts did not use datasets other than those they created, limiting these models to certain domains of text such as archaic poetry or word lists. A truly usable transliteration system must be capable of handling varied domains, meaning that suck models lack the versatility required for real-world usage. The contrast in domain between data also obscures the task's true difficulty. We present a new state-of-the-art sequence-to-sequence model for Tajik-Farsi transliteration trained across all available datasets, and present two datasets of our own. Our results across domains provide clearer understanding of the task, and set comprehensive comparable leading benchmarks. Overall, our model achieves chrF++ and Normalized CER scores of 87.91 and 0.05 from Farsi to Tajik and 92.28 and 0.04 from Tajik to Farsi. Our model, data, and code are available at https://anonymous.4open.science/r/ParsTranslit-FB30/.

Ruixuan Liu, Philip Huang, Ava Pun et al. · cmu

Creating assembly products demands significant manual effort and expert knowledge in 1) designing the assembly and 2) constructing the product. This paper introduces Prompt-to-Product, an automated pipeline that generates real-world assembly products from natural language prompts. Specifically, we leverage LEGO bricks as the assembly platform and automate the process of creating brick assembly structures. Given the user design requirements, Prompt-to-Product generates physically buildable brick designs, and then leverages a bimanual robotic system to construct the real assembly products, bringing user imaginations into the real world. We conduct a comprehensive user study, and the results demonstrate that Prompt-to-Product significantly lowers the barrier and reduces manual effort in creating assembly products from imaginative ideas.

Akhilesh Kakolu Ramarao, Kevin Tang, Dinah Baer-Henney

This study investigates the cognitive plausibility of the Spanish irregular morphomic pattern by directly comparing transformer-based neural networks to human behavioral data from \citet{Nevins2015TheRA}. Using the same analytical framework as the original human study, we evaluate whether transformer models can replicate human-like sensitivity to a complex linguistic phenomena, the morphome, under controlled input conditions. Our experiments focus on three frequency conditions: natural, low-frequency, and high-frequency distributions of verbs exhibiting irregular morphomic patterns. While the models outperformed humans in stem and suffix accuracy, a clear divergence emerged in response preferences. Unlike humans, who consistently favored natural responses across all test items, models' preferred irregular responses and were influenced by the proportion of irregular verbs in their training data. Additionally, models trained on the natural and low-frequency distributions, but not the high-frequency distribution, were sensitive to the phonological similarity between test items and real Spanish L-shaped verbs.

Nitin Venkateswaran, Kevin Tang, Ratree Wayland

Traditional models of accent perception underestimate the role of gradient variations in phonological features which listeners rely upon for their accent judgments. We investigate how pretrained representations from current self-supervised learning (SSL) models of speech encode phonological feature-level variations that influence the perception of segmental accent. We focus on three segments: the labiodental approximant, the rhotic tap, and the retroflex stop, which are uniformly produced in the English of native speakers of Hindi as well as other languages in the Indian sub-continent. We use the CSLU Foreign Accented English corpus (Lander, 2007) to extract, for these segments, phonological feature probabilities using Phonet (Vásquez-Correa et al., 2019) and pretrained representations from Wav2Vec2-BERT (Barrault et al., 2023) and WavLM (Chen et al., 2022) along with accent judgements by native speakers of American English. Probing analyses show that accent strength is best predicted by a subset of the segment's pretrained representation features, in which perceptually salient phonological features that contrast the expected American English and realized non-native English segments are given prominent weighting. A multinomial logistic regression of pretrained representation-based segment distances from American and Indian English baselines on accent ratings reveals strong associations between the odds of accent strength and distances from the baselines, in the expected directions. These results highlight the value of self-supervised speech representations for modeling accent perception using interpretable phonological features.

Anna Stein, Kevin Tang

This study compares probabilistic predictors based on information theory with Naive Discriminative Learning (NDL) predictors in modeling acoustic word duration, focusing on probabilistic reduction. We examine three models using the Buckeye corpus: one with NDL-derived predictors using information-theoretic formulas, one with traditional NDL predictors, and one with N-gram probabilistic predictors. Results show that the N-gram model outperforms both NDL models, challenging the assumption that NDL is more effective due to its cognitive motivation. However, incorporating information-theoretic formulas into NDL improves model performance over the traditional model. This research highlights a) the need to incorporate not only frequency and contextual predictability but also average contextual predictability, and b) the importance of combining information-theoretic metrics of predictability and information derived from discriminative learning in modeling acoustic reduction.

Lian Remme, Kevin Tang

This paper provides a proof of concept that audio of tabletop role-playing games (TTRPG) could serve as a challenge for diarization systems. TTRPGs are carried out mostly by conversation. Participants often alter their voices to indicate that they are talking as a fictional character. Audio processing systems are susceptible to voice conversion with or without technological assistance. TTRPG present a conversational phenomenon in which voice conversion is an inherent characteristic for an immersive gaming experience. This could make it more challenging for diarizers to pick the real speaker and determine that impersonating is just that. We present the creation of a small TTRPG audio dataset and compare it against the AMI and the ICSI corpus. The performance of two diarizers, pyannote.audio and wespeaker, were evaluated. We observed that TTRPGs' properties result in a higher confusion rate for both diarizers. Additionally, wespeaker strongly underestimates the number of speakers in the TTRPG audio files. We propose TTRPG audio as a promising challenge for diarization systems.

Kevin Tang, Jason A. Shaw

The average predictability (aka informativity) of a word in context has been shown to condition word duration (Seyfarth, 2014). All else being equal, words that tend to occur in more predictable environments are shorter than words that tend to occur in less predictable environments. One account of the informativity effect on duration is that the acoustic details of probabilistic reduction are stored as part of a word's mental representation. Other research has argued that predictability effects are tied to prosodic structure in integral ways. With the aim of assessing a potential prosodic basis for informativity effects in speech production, this study extends past work in two directions; it investigated informativity effects in another large language, Mandarin Chinese, and broadened the study beyond word duration to additional acoustic dimensions, pitch and intensity, known to index prosodic prominence. The acoustic information of content words was extracted from a large telephone conversation speech corpus with over 400,000 tokens and 6,000 word types spoken by 1,655 individuals and analyzed for the effect of informativity using frequency statistics estimated from a 431 million word subtitle corpus. Results indicated that words with low informativity have shorter durations, replicating the effect found in English. In addition, informativity had significant effects on maximum pitch and intensity, two phonetic dimensions related to prosodic prominence. Extending this interpretation, these results suggest that predictability is closely linked to prosodic prominence, and that the lexical representation of a word includes phonetic details associated with its average prosodic prominence in discourse. In other words, the lexicon absorbs prosodic influences on speech production.

Linjie Yang, Kevin Tang, Jianchao Yang et al.

Dense captioning is a newly emerging computer vision topic for understanding images with dense language descriptions. The goal is to densely detect visual concepts (e.g., objects, object parts, and interactions between them) from images, labeling each with a short descriptive phrase. We identify two key challenges of dense captioning that need to be properly addressed when tackling the problem. First, dense visual concept annotations in each image are associated with highly overlapping target regions, making accurate localization of each visual concept challenging. Second, the large amount of visual concepts makes it hard to recognize each of them by appearance alone. We propose a new model pipeline based on two novel ideas, joint inference and context fusion, to alleviate these two challenges. We design our model architecture in a methodical manner and thoroughly evaluate the variations in architecture. Our final model, compact and efficient, achieves state-of-the-art accuracy on Visual Genome for dense captioning with a relative gain of 73\% compared to the previous best algorithm. Qualitative experiments also reveal the semantic capabilities of our model in dense captioning.

Kevin Tang, Manohar Paluri, Li Fei-Fei et al.

With the widespread availability of cellphones and cameras that have GPS capabilities, it is common for images being uploaded to the Internet today to have GPS coordinates associated with them. In addition to research that tries to predict GPS coordinates from visual features, this also opens up the door to problems that are conditioned on the availability of GPS coordinates. In this work, we tackle the problem of performing image classification with location context, in which we are given the GPS coordinates for images in both the train and test phases. We explore different ways of encoding and extracting features from the GPS coordinates, and show how to naturally incorporate these features into a Convolutional Neural Network (CNN), the current state-of-the-art for most image classification and recognition problems. We also show how it is possible to simultaneously learn the optimal pooling radii for a subset of our features within the CNN framework. To evaluate our model and to help promote research in this area, we identify a set of location-sensitive concepts and annotate a subset of the Yahoo Flickr Creative Commons 100M dataset that has GPS coordinates with these concepts, which we make publicly available. By leveraging location context, we are able to achieve almost a 7% gain in mean average precision.

Vignesh Ramanathan, Kevin Tang, Greg Mori et al.

In this paper, we propose to learn temporal embeddings of video frames for complex video analysis. Large quantities of unlabeled video data can be easily obtained from the Internet. These videos possess the implicit weak label that they are sequences of temporally and semantically coherent images. We leverage this information to learn temporal embeddings for video frames by associating frames with the temporal context that they appear in. To do this, we propose a scheme for incorporating temporal context based on past and future frames in videos, and compare this to other contextual representations. In addition, we show how data augmentation using multi-resolution samples and hard negatives helps to significantly improve the quality of the learned embeddings. We evaluate various design decisions for learning temporal embeddings, and show that our embeddings can improve performance for multiple video tasks such as retrieval, classification, and temporal order recovery in unconstrained Internet video.