Pre-trained language models trained on large-scale data have learned serious levels of social biases. Consequently, various methods have been proposed to debias pre-trained models. Debiasing methods need to mitigate only discriminatory bias information from the pre-trained models, while retaining information that is useful for the downstream tasks. In previous research, whether useful information is retained has been confirmed by the performance of downstream tasks in debiased pre-trained models. On the other hand, it is not clear whether these benchmarks consist of data pertaining to social biases and are appropriate for investigating the impact of debiasing. For example in gender-related social biases, data containing female words (e.g. ``she, female, woman''), male words (e.g. ``he, male, man''), and stereotypical words (e.g. ``nurse, doctor, professor'') are considered to be the most affected by debiasing. If there is not much data containing these words in a benchmark dataset for a target task, there is the possibility of erroneously evaluating the effects of debiasing. In this study, we compare the impact of debiasing on performance across multiple downstream tasks using a wide-range of benchmark datasets that containing female, male, and stereotypical words. Experiments show that the effects of debiasing are consistently \emph{underestimated} across all tasks. Moreover, the effects of debiasing could be reliably evaluated by separately considering instances containing female, male, and stereotypical words than all of the instances in a benchmark dataset.
Despite their impressive performance in a wide range of NLP tasks, Large Language Models (LLMs) have been reported to encode worrying-levels of gender bias. Prior work has proposed debiasing methods that require human labelled examples, data augmentation and fine-tuning of the LLMs, which are computationally costly. Moreover, one might not even have access to the internal parameters for performing debiasing such as in the case of commercially available LLMs such as GPT-4. To address this challenge we propose bias suppression, a novel alternative to debiasing that does not require access to model parameters. We show that text-based preambles, generated from manually designed templates covering counterfactual statements, can accurately suppress gender biases in LLMs. Moreover, we find that descriptive sentences for occupations can further suppress gender biases. Interestingly, we find that bias suppression has a minimal adverse effect on downstream task performance, while effectively mitigating the gender biases.
Prior work has shown that the ordering in which concepts are shown to a commonsense generator plays an important role, affecting the quality of the generated sentence. However, it remains a challenge to determine the optimal ordering of a given set of concepts such that a natural sentence covering all the concepts could be generated from a pretrained generator. To understand the relationship between the ordering of the input concepts and the quality of the generated sentences, we conduct a systematic study considering multiple language models (LMs) and concept ordering strategies. We find that BART-large model consistently outperforms all other LMs considered in this study when fine-tuned using the ordering of concepts as they appear in CommonGen training data as measured using multiple evaluation metrics. Moreover, the larger GPT3-based large language models (LLMs) variants do not necessarily outperform much smaller LMs on this task, even when fine-tuned on task-specific training data. Interestingly, human annotators significantly reorder input concept sets when manually writing sentences covering those concepts, and this ordering provides the best sentence generations independently of the LM used for the generation, outperforming a probabilistic concept ordering baseline
The missing signal caused by the objects being occluded or an unstable sensor is a common challenge during data collection. Such missing signals will adversely affect the results obtained from the data, and this issue is observed more frequently in robotic tactile perception. In tactile perception, due to the limited working space and the dynamic environment, the contact between the tactile sensor and the object is frequently insufficient and unstable, which causes the partial loss of signals, thus leading to incomplete tactile data. The tactile data will therefore contain fewer tactile cues with low information density. In this paper, we propose a tactile representation learning method, named TacMAE, based on Masked Autoencoder to address the problem of incomplete tactile data in tactile perception. In our framework, a portion of the tactile image is masked out to simulate the missing contact region. By reconstructing the missing signals in the tactile image, the trained model can achieve a high-level understanding of surface geometry and tactile properties from limited tactile cues. The experimental results of tactile texture recognition show that our proposed TacMAE can achieve a high recognition accuracy of 71.4% in the zero-shot transfer and 85.8% after fine-tuning, which are 15.2% and 8.2% higher than the results without using masked modeling. The extensive experiments on YCB objects demonstrate the knowledge transferability of our proposed method and the potential to improve efficiency in tactile exploration.
Tactile sensing plays an irreplaceable role in robotic material recognition. It enables robots to distinguish material properties such as their local geometry and textures, especially for materials like textiles. However, most tactile recognition methods can only classify known materials that have been touched and trained with tactile data, yet cannot classify unknown materials that are not trained with tactile data. To solve this problem, we propose a tactile zero-shot learning framework to recognise unknown materials when they are touched for the first time without requiring training tactile samples. The visual modality, providing tactile cues from sight, and semantic attributes, giving high-level characteristics, are combined together to bridge the gap between touched classes and untouched classes. A generative model is learnt to synthesise tactile features according to corresponding visual images and semantic embeddings, and then a classifier can be trained using the synthesised tactile features of untouched materials for zero-shot recognition. Extensive experiments demonstrate that our proposed multimodal generative model can achieve a high recognition accuracy of 83.06% in classifying materials that were not touched before. The robotic experiment demo and the dataset are available at https://sites.google.com/view/multimodalzsl.
Sense embedding learning methods learn multiple vectors for a given ambiguous word, corresponding to its different word senses. For this purpose, different methods have been proposed in prior work on sense embedding learning that use different sense inventories, sense-tagged corpora and learning methods. However, not all existing sense embeddings cover all senses of ambiguous words equally well due to the discrepancies in their training resources. To address this problem, we propose the first-ever meta-sense embedding method -- Neighbour Preserving Meta-Sense Embeddings, which learns meta-sense embeddings by combining multiple independently trained source sense embeddings such that the sense neighbourhoods computed from the source embeddings are preserved in the meta-embedding space. Our proposed method can combine source sense embeddings that cover different sets of word senses. Experimental results on Word Sense Disambiguation (WSD) and Word-in-Context (WiC) tasks show that the proposed meta-sense embedding method consistently outperforms several competitive baselines.
Clustering is a common task in machine learning, but clusters of unlabelled data can be hard to quantify. The application of clustering algorithms in chemistry is often dependant on material representation. Ascertaining the effects of different representations, clustering algorithms, or data transformations on the resulting clusters is difficult due to the dimensionality of these data. We present a thorough analysis of measures for isotropy of a cluster, including a novel implantation based on an existing derivation. Using fractional anisotropy, a common method used in medical imaging for comparison, we then expand these measures to examine the average isotropy of a set of clusters. A use case for such measures is demonstrated by quantifying the effects of kernel approximation functions on different representations of the Inorganic Crystal Structure Database. Broader applicability of these methods is demonstrated in analysing learnt embedding of the MNIST dataset. Random clusters are explored to examine the differences between isotropy measures presented, and to see how each method scales with the dimensionality. Python implementations of these measures are provided for use by the community.
Cosine similarity between two words, computed using their contextualised token embeddings obtained from masked language models (MLMs) such as BERT has shown to underestimate the actual similarity between those words (Zhou et al., 2022). This similarity underestimation problem is particularly severe for highly frequent words. Although this problem has been noted in prior work, no solution has been proposed thus far. We observe that the L2 norm of contextualised embeddings of a word correlates with its log-frequency in the pretraining corpus. Consequently, the larger L2 norms associated with the highly frequent words reduce the cosine similarity values measured between them, thus underestimating the similarity scores. To solve this issue, we propose a method to discount the L2 norm of a contextualised word embedding by the frequency of that word in a corpus when measuring the cosine similarities between words. We show that the so called stop words behave differently from the rest of the words, which require special consideration during their discounting process. Experimental results on a contextualised word similarity dataset show that our proposed discounting method accurately solves the similarity underestimation problem.
Languages are dynamic entities, where the meanings associated with words constantly change with time. Detecting the semantic variation of words is an important task for various NLP applications that must make time-sensitive predictions. Existing work on semantic variation prediction have predominantly focused on comparing some form of an averaged contextualised representation of a target word computed from a given corpus. However, some of the previously associated meanings of a target word can become obsolete over time (e.g. meaning of gay as happy), while novel usages of existing words are observed (e.g. meaning of cell as a mobile phone). We argue that mean representations alone cannot accurately capture such semantic variations and propose a method that uses the entire cohort of the contextualised embeddings of the target word, which we refer to as the sibling distribution. Experimental results on SemEval-2020 Task 1 benchmark dataset for semantic variation prediction show that our method outperforms prior work that consider only the mean embeddings, and is comparable to the current state-of-the-art. Moreover, a qualitative analysis shows that our method detects important semantic changes in words that are not captured by the existing methods. Source code is available at https://github.com/a1da4/svp-gauss .
Discrete prompts have been used for fine-tuning Pre-trained Language Models for diverse NLP tasks. In particular, automatic methods that generate discrete prompts from a small set of training instances have reported superior performance. However, a closer look at the learnt prompts reveals that they contain noisy and counter-intuitive lexical constructs that would not be encountered in manually-written prompts. This raises an important yet understudied question regarding the robustness of automatically learnt discrete prompts when used in downstream tasks. To address this question, we conduct a systematic study of the robustness of discrete prompts by applying carefully designed perturbations into an application using AutoPrompt and then measure their performance in two Natural Language Inference (NLI) datasets. Our experimental results show that although the discrete prompt-based method remains relatively robust against perturbations to NLI inputs, they are highly sensitive to other types of perturbations such as shuffling and deletion of prompt tokens. Moreover, they generalize poorly across different NLI datasets. We hope our findings will inspire future work on robust discrete prompt learning.