Large language models have shown their ability to become effective few-shot learners with prompting, revoluting the paradigm of learning with data scarcity. However, this approach largely depends on the quality of prompt initialization and always exhibits large variability among different runs. Such property makes prompt tuning highly unreliable and vulnerable to poorly constructed prompts, which limits its extension to more real-world applications. To tackle this issue, we propose to treat the hard prompt and soft prompt as separate inputs to mitigate noise brought by the prompt initialization. Furthermore, we optimize soft prompts with contrastive learning for utilizing class-aware information in the training process to maintain model performance. Experimental results demonstrate that StablePT outperforms state-of-the-art methods by 6.97% in accuracy and reduces the standard deviation by 1.92 on average. Furthermore, extensive experiments underscore its robustness and stability across 8 datasets covering various tasks.

Standard fine-tuning of language models typically performs well on in-distribution data, but suffers with generalization to distribution shifts. In this work, we aim to improve the generalization of adapter-based cross-lingual task transfer where such cross-language distribution shifts are imminent. We investigate scheduled unfreezing algorithms –originally proposed to mitigate catastrophic forgetting in transfer learning – for fine-tuning task adapters. Our experiments show that scheduled unfreezing methods close the gap to full fine-tuning and achieve stronger cross-lingual transfer performance, suggesting that these methods can go beyond just mitigating catastrophic forgetting. Next, aiming to understand these empirical findings, we investigate the learning dynamics of scheduled unfreezing using Fisher Information. Our experiments reveal that scheduled unfreezing induces different learning dynamics compared to standard fine-tuning, and provide evidence that the dynamics of Fisher Information during training correlate with cross-lingual generalization performance. We additionally propose a general scheduled unfreezing algorithm that achieves an average of 2 points improvement over four datasets compared to standard fine-tuning and provides empirical evidence for a theory-based justification of the heuristic unfreezing schedule for task adapter training.

Large language models (LLMs) are highly adept at question answering and reasoning tasks, but when reasoning in a situational context, human expectations vary depending on the relevant cultural common ground. As languages are associated with diverse cultures, LLMs should also be culturally-diverse reasoners. In this paper, we study the ability of a wide range of state-of-the-art multilingual LLMs (mLLMs) to reason with proverbs and sayings in a conversational context. Our experiments reveal that: (1) mLLMs “know” limited proverbs and memorizing proverbs does not mean understanding them within a conversational context; (2) mLLMs struggle to reason with figurative proverbs and sayings, and when asked to select the wrong answer (instead of asking it to select the correct answer); and (3) there is a “culture gap” in mLLMs when reasoning about proverbs and sayings translated from other languages. We construct and release our evaluation dataset MAPS (MulticulturAl Proverbs and Sayings) for proverb understanding with conversational context for six different languages.

Visual question answering (VQA) is one of the crucial vision-and-language tasks. Yet, existing VQA research has mostly focused on the English language, due to a lack of suitable evaluation resources. Previous work on cross-lingual VQA has reported poor zero-shot transfer performance of current multilingual multimodal Transformers with large gaps to monolingual performance, without any deeper analysis. In this work, we delve deeper into the different aspects of cross-lingual VQA, aiming to understand the impact of 1) modeling methods and choices, including architecture, inductive bias, fine-tuning; 2) learning biases: including question types and modality biases in cross-lingual setups. The key results of our analysis are: 1. We show that simple modifications to the standard training setup can substantially reduce the transfer gap to monolingual English performance, yielding +10 accuracy points over existing methods. 2. We analyze cross-lingual VQA across different question types of varying complexity for different multilingual multimodal Transformers, and identify question types that are the most difficult to improve on. 3. We provide an analysis of modality biases present in training data and models, revealing why zero-shot performance gaps remain for certain question types and languages.

Current multimodal models, aimed at solving Vision and Language (V+L) tasks, predominantly repurpose Vision Encoders (VE) as feature extractors. While many VEs—of different architectures, trained on different data and objectives—are publicly available, they are not designed for the downstream V+L tasks. Nonetheless, most current work assumes that a single pre-trained VE can serve as a general-purpose encoder. In this work, we focus on analysis and aim to understand whether the information stored within different VEs is complementary, i.e. if providing the model with features from multiple VEs can improve the performance on a target task, and how they are combined. We exhaustively experiment with three popular VEs on six downstream V+L tasks and analyze the attention and VE-dropout patterns. Our analyses suggest that diverse VEs complement each other, resulting in improved downstream V+L task performance, where the improvements are not due to simple ensemble effects (i.e. the performance does not always improve when increasing the number of encoders). We demonstrate that future VEs, which are not repurposed, but explicitly designed for V+L tasks, have the potential of improving performance on the target V+L tasks.

Real-world politically-opinionated memes often rely on figurative language to cloak propaganda and radical ideas to help them spread. It is not only a scientific challenge to develop machine learning models to recognize them in memes, but also sociologically beneficial to understand hidden meanings at scale and raise awareness. These memes are fast-evolving (in both topics and visuals) and it remains unclear whether current multimodal machine learning models are robust to such distribution shifts. To enable future research into this area, we first present FigMemes, a dataset for figurative language classification in politically-opinionated memes. We evaluate the performance of state-of-the-art unimodal and multimodal models and provide comprehensive benchmark results. The key contributions of this proposed dataset include annotations of six commonly used types of figurative language in politically-opinionated memes, and a wide range of topics and visual styles.We also provide analyses on the ability of multimodal models to generalize across distribution shifts in memes. Our dataset poses unique machine learning challenges and our results show that current models have significant room for improvement in both performance and robustness to distribution shifts.

Hierarchical Text Classification (HTC), which aims to predict text labels organized in hierarchical space, is a significant task lacking in investigation in natural language processing. Existing methods usually encode the entire hierarchical structure and fail to construct a robust label-dependent model, making it hard to make accurate predictions on sparse lower-level labels and achieving low Macro-F1. In this paper, we explore the level dependency and path dependency of the label hierarchy in a generative way for building the knowledge of upper-level labels of current path into lower-level ones, and thus propose a novel PAAM-HiA-T5 model for HTC: a hierarchy-aware T5 model with path-adaptive attention mechanism. Specifically, we generate a multi-level sequential label structure to exploit hierarchical dependency across different levels with Breadth-First Search (BFS) and T5 model. To further improve label dependency prediction within each path, we then propose an original path-adaptive attention mechanism (PAAM) to lead the model to adaptively focus on the path where the currently generated label is located, shielding the noise from other paths. Comprehensive experiments on three benchmark datasets show that PAAM-HiA-T5 greatly outperforms all state-of-the-art HTC approaches especially in Macro-F1.

This work demonstrates the development process of a machine learning architecture for inference that can scale to a large volume of requests. We used a BERT model that was fine-tuned for emotion analysis, returning a probability distribution of emotions given a paragraph. The model was deployed as a gRPC service on Kubernetes. Apache Spark was used to perform inference in batches by calling the service. We encountered some performance and concurrency challenges and created solutions to achieve faster running time. Starting with 200 successful inference requests per minute, we were able to achieve as high as 18 thousand successful requests per minute with the same batch job resource allocation. As a result, we successfully stored emotion probabilities for 95 million paragraphs within 96 hours.

In natural language processing (NLP), state-of-the-art (SOTA) semi-supervised learning (SSL) frameworks have shown great performance on deep pre-trained language models such as BERT, and are expected to significantly reduce the demand for manual labeling. However, our empirical studies indicate that these frameworks are not suitable for lightweight models such as TextCNN, LSTM and etc. In this work, we develop a new SSL framework called FLiText, which stands for Faster and Lighter semi-supervised Text classification. FLiText introduces an inspirer network together with the consistency regularization framework, which leverages a generalized regular constraint on the lightweight models for efficient SSL. As a result, FLiText obtains new SOTA performance for lightweight models across multiple SSL benchmarks on text classification. Compared with existing SOTA SSL methods on TextCNN, FLiText improves the accuracy of lightweight model TextCNN from 51.00% to 90.49% on IMDb, 39.8% to 58.06% on Yelp-5, and from 55.3% to 65.08% on Yahoo! Answer. In addition, compared with the fully supervised method on the full dataset, FLiText just uses less than 1% of labeled data to improve the accuracy by 6.59%, 3.94%, and 3.22% on the datasets of IMDb, Yelp-5, and Yahoo! Answer respectively.

One daunting problem for semantic parsing is the scarcity of annotation. Aiming to reduce nontrivial human labor, we propose a two-stage semantic parsing framework, where the first stage utilizes an unsupervised paraphrase model to convert an unlabeled natural language utterance into the canonical utterance. The downstream naive semantic parser accepts the intermediate output and returns the target logical form. Furthermore, the entire training process is split into two phases: pre-training and cycle learning. Three tailored self-supervised tasks are introduced throughout training to activate the unsupervised paraphrase model. Experimental results on benchmarks Overnight and GeoGranno demonstrate that our framework is effective and compatible with supervised training.

Semantic parsing converts natural language queries into structured logical forms. The lack of training data is still one of the most serious problems in this area. In this work, we develop a semantic parsing framework with the dual learning algorithm, which enables a semantic parser to make full use of data (labeled and even unlabeled) through a dual-learning game. This game between a primal model (semantic parsing) and a dual model (logical form to query) forces them to regularize each other, and can achieve feedback signals from some prior-knowledge. By utilizing the prior-knowledge of logical form structures, we propose a novel reward signal at the surface and semantic levels which tends to generate complete and reasonable logical forms. Experimental results show that our approach achieves new state-of-the-art performance on ATIS dataset and gets competitive performance on OVERNIGHT dataset.

We introduce a new dataset for multi-class emotion analysis from long-form narratives in English. The Dataset for Emotions of Narrative Sequences (DENS) was collected from both classic literature available on Project Gutenberg and modern online narratives avail- able on Wattpad, annotated using Amazon Mechanical Turk. A number of statistics and baseline benchmarks are provided for the dataset. Of the tested techniques, we find that the fine-tuning of a pre-trained BERT model achieves the best results, with an average micro-F1 score of 60.4%. Our results show that the dataset provides a novel opportunity in emotion analysis that requires moving beyond existing sentence-level techniques.

We study methods for learning sentence embeddings with syntactic structure. We focus on methods of learning syntactic sentence-embeddings by using a multilingual parallel-corpus augmented by Universal Parts-of-Speech tags. We evaluate the quality of the learned embeddings by examining sentence-level nearest neighbours and functional dissimilarity in the embedding space. We also evaluate the ability of the method to learn syntactic sentence-embeddings for low-resource languages and demonstrate strong evidence for transfer learning. Our results show that syntactic sentence-embeddings can be learned while using less training data, fewer model parameters, and resulting in better evaluation metrics than state-of-the-art language models.

In this paper we describe an approach that both creates crosslingual acoustic monophone model sets for speech recognition tasks and objectively predicts their performance without target-language speech data or acoustic measurement techniques. This strategy is based on a series of linguistic metrics characterizing the articulatory phonetic and phonological distances of target-language phonemes from source-language phonemes. We term these algorithms the Combined Phonetic and Phonological Crosslingual Distance (CPP-CD) metric and the Combined Phonetic and Phonological Crosslingual Prediction (CPP-CP) metric. The particular motivations for this project are the current unavailability and often prohibitively high production cost of speech databases for many strategically important low- and middle-density languages. First, we describe the CPP-CD approach and compare the performance of CPP-CD-specified models to both native language models and crosslingual models selected by the Bhattacharyya acoustic-model distance metric in automatic speech recognition (ASR) experiments. Results confirm that the CPP-CD approach nearly matches those achieved by the acoustic distance metric. We then test the CPP-CP algorithm on the CPP-CD models by comparing the CPP-CP scores to the recognition phoneme error rates. Based on this comparison, we conclude that the CPP-CP algorithm is a reliable indicator of crosslingual model performance in speech recognition tasks.