Gender bias in language models has gained increasing attention in the field of natural language processing. Encoder-based transformer models, which have achieved state-of-the-art performance in various language tasks, have been shown to exhibit strong gender biases inherited from their training data. This paper investigates gender bias in contextualized word embeddings, a crucial component of transformer-based models. We focus on prominent architectures such as BERT, ALBERT, RoBERTa, and DistilBERT to examine their vulnerability to gender bias. To quantify the degree of bias, we introduce a novel metric, MALoR, which assesses bias based on model probabilities for filling masked tokens. We further propose a mitigation approach involving continued pre-training on a gender-balanced dataset generated via Counterfactual Data Augmentation. Our experiments reveal significant reductions in gender bias scores across different pronoun pairs. For instance, in BERT-base, bias scores for "he-she" dropped from 1.27 to 0.08, and "his-her" from 2.51 to 0.36 following our mitigation approach. We also observed similar improvements across other models, with "male-female" bias decreasing from 1.82 to 0.10 in BERT-large. Our approach effectively reduces gender bias without compromising model performance on downstream tasks.

Topic modeling is traditionally applied to word counts without accounting for the context in which words appear. Recent advancements in large language models (LLMs) offer contextualized word embeddings, which capture deeper meaning and relationships between words. We aim to leverage such embeddings to improve topic modeling. We use a pre-trained LLM to convert each document into a sequence of word embeddings. This sequence is then modeled as a Poisson point process, with its intensity measure expressed as a convex combination of $K$ base measures, each corresponding to a topic. To estimate these topics, we propose a flexible algorithm that integrates traditional topic modeling methods, enhanced by net-rounding applied before and kernel smoothing applied after. One advantage of this framework is that it treats the LLM as a black box, requiring no fine-tuning of its parameters. Another advantage is its ability to seamlessly integrate any traditional topic modeling approach as a plug-in module, without the need for modifications Assuming each topic is a $\beta$-H\"{o}lder smooth intensity measure on the embedded space, we establish the rate of convergence of our method. We also provide a minimax lower bound and show that the rate of our method matches with the lower bound when $\beta\leq 1$. Additionally, we apply our method to several datasets, providing evidence that it offers an advantage over traditional topic modeling approaches.

Virtual particles are peculiar objects. They figure prominently in much of theoretical and experimental research in elementary particle physics. But exactly what they are is far from obvious. In particular, to what extent they should be considered "real" remains a matter of controversy in philosophy of science. Also their origin and development has only recently come into focus of scholarship in the history of science. In this study, we propose using the intriguing case of virtual particles to discuss the efficacy of Semantic Change Detection (SCD) based on contextualized word embeddings from a domain-adapted BERT model in studying specific scientific concepts. We find that the SCD metrics align well with qualitative research insights in the history and philosophy of science, as well as with the results obtained from Dependency Parsing to determine the frequency and connotations of the term "virtual". Still, the metrics of SCD provide additional insights over and above the qualitative research and the Dependency Parsing. Among other things, the metrics suggest that the concept of the virtual particle became more stable after 1950 but at the same time also more polysemous.

In this paper, we study different approaches for classifying emotions from speech using acoustic and text-based features. We propose to obtain contextualized word embeddings with BERT to represent the information contained in speech transcriptions and show that this results in better performance than using Glove embeddings. We also propose and compare different strategies to combine the audio and text modalities, evaluating them on IEMOCAP and MSP-PODCAST datasets. We find that fusing acoustic and text-based systems is beneficial on both datasets, though only subtle differences are observed across the evaluated fusion approaches. Finally, for IEMOCAP, we show the large effect that the criteria used to define the cross-validation folds have on results. In particular, the standard way of creating folds for this dataset results in a highly optimistic estimation of performance for the text-based system, suggesting that some previous works may overestimate the advantage of incorporating transcriptions.

There are two primary approaches to addressing cross-lingual transfer: multilingual pre-training, which implicitly aligns the hidden representations of various languages, and translate-test, which explicitly translates different languages into an intermediate language, such as English. Translate-test offers better interpretability compared to multilingual pre-training. However, it has lower performance than multilingual pre-training(Conneau and Lample, 2019; Conneau et al, 2020) and struggles with word-level tasks due to translation altering word order. As a result, we propose a new Machine-created Universal Language (MUL) as an alternative intermediate language. MUL comprises a set of discrete symbols forming a universal vocabulary and a natural language to MUL translator for converting multiple natural languages to MUL. MUL unifies shared concepts from various languages into a single universal word, enhancing cross-language transfer. Additionally, MUL retains language-specific words and word order, allowing the model to be easily applied to word-level tasks. Our experiments demonstrate that translating into MUL yields improved performance compared to multilingual pre-training, and our analysis indicates that MUL possesses strong interpretability. The code is at: https://github.com/microsoft/Unicoder/tree/master/MCUL.

Word Sense Disambiguation (WSD) is one of the hardest tasks in natural language understanding and knowledge engineering. The glass ceiling of 80% F1 score is recently achieved through supervised deep-learning, enriched by a variety of knowledge graphs. Here, we propose a novel neurosymbolic methodology that is able to push the F1 score above 90%. The core of our methodology is a neurosymbolic sense embedding, in terms of a configuration of nested balls in n-dimensional space. The centre point of a ball well-preserves word embedding, which partially fix the locations of balls. Inclusion relations among balls precisely encode symbolic hypernym relations among senses, and enable simple logic deduction among sense embeddings, which cannot be realised before. We trained a Transformer to learn the mapping from a contextualized word embedding to its sense ball embedding, just like playing the game of darts (a game of shooting darts into a dartboard). A series of experiments are conducted by utilizing pre-training n-ball embeddings, which have the coverage of around 70% training data and 75% testing data in the benchmark WSD corpus. The F1 scores in experiments range from 90.1% to 100.0% in all six groups of test data-sets (each group has 4 testing data with different sizes of n-ball embeddings). Our novel neurosymbolic methodology has the potential to break the ceiling of deep-learning approaches for WSD. Limitations and extensions of our current works are listed.

Language models are trained on large-scale corpora that embed implicit biases documented in psychology. Valence associations (pleasantness/unpleasantness) of social groups determine the biased attitudes towards groups and concepts in social cognition. Building on this established literature, we quantify how social groups are valenced in English language models using a sentence template that provides an intersectional context. We study biases related to age, education, gender, height, intelligence, literacy, race, religion, sex, sexual orientation, social class, and weight. We present a concept projection approach to capture the valence subspace through contextualized word embeddings of language models. Adapting the projection-based approach to embedding association tests that quantify bias, we find that language models exhibit the most biased attitudes against gender identity, social class, and sexual orientation signals in language. We find that the largest and better-performing model that we study is also more biased as it effectively captures bias embedded in sociocultural data. We validate the bias evaluation method by overperforming on an intrinsic valence evaluation task. The approach enables us to measure complex intersectional biases as they are known to manifest in the outputs and applications of language models that perpetuate historical biases. Moreover, our approach contributes to design justice as it studies the associations of groups underrepresented in language such as transgender and homosexual individuals.

In this study, we aimed to address the growing concern of trolling behavior on social media by developing and evaluating a set of model architectures for the automatic detection of troll tweets. Utilizing deep learning techniques and pre-trained word embedding methods such as BERT, ELMo, and GloVe, we evaluated the performance of each architecture using metrics such as classification accuracy, F1 score, AUC, and precision. Our results indicate that BERT and ELMo embedding methods performed better than the GloVe method, likely due to their ability to provide contextualized word embeddings that better capture the nuances and subtleties of language use in online social media. Additionally, we found that CNN and GRU encoders performed similarly in terms of F1 score and AUC, suggesting their effectiveness in extracting relevant information from input text. The best-performing method was found to be an ELMo-based architecture that employed a GRU classifier, with an AUC score of 0.929. This research highlights the importance of utilizing contextualized word embeddings and appropriate encoder methods in the task of troll tweet detection, which can assist social-based systems in improving their performance in identifying and addressing trolling behavior on their platforms.