Pretrained language models (PLMs) are key components in NLP, but they contain strong social biases. Quantifying these biases is challenging because current methods focusing on fill-the-mask objectives are sensitive to slight changes in input. To address this, we propose LABDet, a robust language-agnostic method for evaluating bias in PLMs. For nationality as a case study, we show that LABDet "surfaces" nationality bias by training a classifier on top of a frozen PLM on non-nationality sentiment detection. Collaborating with political scientists, we find consistent patterns of nationality bias across monolingual PLMs in six languages that align with historical and political context. We also show for English BERT that bias surfaced by LABDet correlates well with bias in the pretraining data; thus, our work is one of the few studies that directly links pretraining data to PLM behavior. Finally, we verify LABDet's reliability and applicability to different templates and languages through an extensive set of robustness checks.
Instruction tuning enables language models to generalize more effectively and better follow user intent. However, obtaining instruction data can be costly and challenging. Prior works employ methods such as expensive human annotation, crowd-sourced datasets with alignment issues, or generating noisy examples via LLMs. We introduce the LongForm dataset, which is created by leveraging English corpus examples with augmented instructions. We select a diverse set of human-written documents from existing corpora such as C4 and Wikipedia and generate instructions for the given documents via LLMs. This approach provides a cheaper and cleaner instruction-tuning dataset and one suitable for long text generation. We finetune T5, OPT, and LLaMA models on our dataset and show that even smaller LongForm models have good generalization capabilities for text generation. Our models outperform 10x larger language models without instruction tuning on various tasks such as story/recipe generation and long-form question answering. Moreover, LongForm models outperform prior instruction-tuned models such as FLAN-T5 and Alpaca by a large margin. Finally, our models can effectively follow and answer multilingual instructions; we demonstrate this for news generation. We publicly release our data and models: https://github.com/akoksal/LongForm.
Large pretrained language models have been widely used in downstream NLP tasks via task-specific fine-tuning. Recently, an array of Parameter-Efficient Fine-Tuning (PEFT) methods have also achieved strong task performance while updating a much smaller number of parameters compared to full model tuning. However, it is non-trivial to make informed per-task design choices (i.e., to create PEFT configurations) concerning the selection of PEFT architectures and modules, the number of tunable parameters, and even the layers in which the PEFT modules are inserted. Consequently, it is highly likely that the current, manually set PEFT configurations might be suboptimal for many tasks from the perspective of the performance-to-efficiency trade-off. To address the core question of the PEFT configuration selection that aims to control and maximise the balance between performance and parameter efficiency, we first define a rich configuration search space spanning multiple representative PEFT modules along with finer-grained configuration decisions over the modules (e.g., parameter budget, insertion layer). We then propose AutoPEFT, a novel framework to traverse this configuration space: it automatically configures multiple PEFT modules via high-dimensional Bayesian optimisation. We show the resource scalability and task transferability of AutoPEFT-found configurations, outperforming existing PEFT methods on average on the standard GLUE benchmark while conducting the configuration search on a single task. The per-task AutoPEFT-based configuration search even outperforms full-model fine-tuning.
Task-oriented dialogue (TOD) systems have been applied in a range of domains to support human users to achieve specific goals. Systems are typically constructed for a single domain or language and do not generalise well beyond this. Their extension to other languages in particular is restricted by the lack of available training data for many of the world's languages. To support work on Natural Language Understanding (NLU) in TOD across multiple languages and domains simultaneously, we constructed MULTI3NLU++, a multilingual, multi-intent, multi-domain dataset. MULTI3NLU++ extends the English-only NLU++ dataset to include manual translations into a range of high, medium and low resource languages (Spanish, Marathi, Turkish and Amharic), in two domains (banking and hotels). MULTI3NLU++ inherits the multi-intent property of NLU++, where an utterance may be labelled with multiple intents, providing a more realistic representation of a user's goals and aligning with the more complex tasks that commercial systems aim to model. We use MULTI3NLU++ to benchmark state-of-the-art multilingual language models as well as Machine Translation and Question Answering systems for the NLU task of intent detection for TOD systems in the multilingual setting. The results demonstrate the challenging nature of the dataset, particularly in the low-resource language setting.
End-to-end (E2E) task-oriented dialogue (ToD) systems are prone to fall into the so-called 'likelihood trap', resulting in generated responses which are dull, repetitive, and often inconsistent with dialogue history. Comparing ranked lists of multiple generated responses against the 'gold response' (from training data) reveals a wide diversity in response quality, with many good responses placed lower in the ranked list. The main challenge, addressed in this work, is then how to reach beyond greedily generated system responses, that is, how to obtain and select such high-quality responses from the list of overgenerated responses at inference without availability of the gold response. To this end, we propose a simple yet effective reranking method which aims to select high-quality items from the lists of responses initially overgenerated by the system. The idea is to use any sequence-level (similarity) scoring function to divide the semantic space of responses into high-scoring versus low-scoring partitions. At training, the high-scoring partition comprises all generated responses whose similarity to the gold response is higher than the similarity of the greedy response to the gold response. At inference, the aim is to estimate the probability that each overgenerated response belongs to the high-scoring partition, given only previous dialogue history. We validate the robustness and versatility of our proposed method on the standard MultiWOZ dataset: our methods improve a state-of-the-art E2E ToD system by 2.4 BLEU, 3.2 ROUGE, and 2.8 METEOR scores, achieving new peak results. Additional experiments on the BiTOD dataset and human evaluation further ascertain the generalisability and effectiveness of the proposed framework.
Bilingual lexicon induction (BLI) with limited bilingual supervision is a crucial yet challenging task in multilingual NLP. Current state-of-the-art BLI methods rely on the induction of cross-lingual word embeddings (CLWEs) to capture cross-lingual word similarities; such CLWEs are obtained 1) via traditional static models (e.g., VecMap), or 2) by extracting type-level CLWEs from multilingual pretrained language models (mPLMs), or 3) through combining the former two options. In this work, we propose a novel semi-supervised post-hoc reranking method termed BLICEr (BLI with Cross-Encoder Reranking), applicable to any precalculated CLWE space, which improves their BLI capability. The key idea is to 'extract' cross-lingual lexical knowledge from mPLMs, and then combine it with the original CLWEs. This crucial step is done via 1) creating a word similarity dataset, comprising positive word pairs (i.e., true translations) and hard negative pairs induced from the original CLWE space, and then 2) fine-tuning an mPLM (e.g., mBERT or XLM-R) in a cross-encoder manner to predict the similarity scores. At inference, we 3) combine the similarity score from the original CLWE space with the score from the BLI-tuned cross-encoder. BLICEr establishes new state-of-the-art results on two standard BLI benchmarks spanning a wide spectrum of diverse languages: it substantially outperforms a series of strong baselines across the board. We also validate the robustness of BLICEr with different CLWEs.
Acquiring factual knowledge with Pretrained Language Models (PLMs) has attracted increasing attention, showing promising performance in many knowledge-intensive tasks. Their good performance has led the community to believe that the models do possess a modicum of reasoning competence rather than merely memorising the knowledge. In this paper, we conduct a comprehensive evaluation of the learnable deductive (also known as explicit) reasoning capability of PLMs. Through a series of controlled experiments, we posit two main findings. (i) PLMs inadequately generalise learned logic rules and perform inconsistently against simple adversarial surface form edits. (ii) While the deductive reasoning fine-tuning of PLMs does improve their performance on reasoning over unseen knowledge facts, it results in catastrophically forgetting the previously learnt knowledge. Our main results suggest that PLMs cannot yet perform reliable deductive reasoning, demonstrating the importance of controlled examinations and probing of PLMs' reasoning abilities; we reach beyond (misleading) task performance, revealing that PLMs are still far from human-level reasoning capabilities, even for simple deductive tasks.
Pretrained multilingual language models (LMs) can be successfully transformed into multilingual sentence encoders (SEs; e.g., LaBSE, xMPNET) via additional fine-tuning or model distillation on parallel data. However, it remains uncertain how to best leverage their knowledge to represent sub-sentence lexical items (i.e., words and phrases) in cross-lingual lexical tasks. In this work, we probe these SEs for the amount of cross-lingual lexical knowledge stored in their parameters, and compare them against the original multilingual LMs. We also devise a novel method to expose this knowledge by additionally fine-tuning multilingual models through inexpensive contrastive learning procedure, requiring only a small amount of word translation pairs. We evaluate our method on bilingual lexical induction (BLI), cross-lingual lexical semantic similarity, and cross-lingual entity linking, and report substantial gains on standard benchmarks (e.g., +10 Precision@1 points in BLI), validating that the SEs such as LaBSE can be 'rewired' into effective cross-lingual lexical encoders. Moreover, we show that resulting representations can be successfully interpolated with static embeddings from cross-lingual word embedding spaces to further boost the performance in lexical tasks. In sum, our approach provides an effective tool for exposing and harnessing multilingual lexical knowledge 'hidden' in multilingual sentence encoders.
Word translation or bilingual lexicon induction (BLI) is a key cross-lingual task, aiming to bridge the lexical gap between different languages. In this work, we propose a robust and effective two-stage contrastive learning framework for the BLI task. At Stage C1, we propose to refine standard cross-lingual linear maps between static word embeddings (WEs) via a contrastive learning objective; we also show how to integrate it into the self-learning procedure for even more refined cross-lingual maps. In Stage C2, we conduct BLI-oriented contrastive fine-tuning of mBERT, unlocking its word translation capability. We also show that static WEs induced from the `C2-tuned' mBERT complement static WEs from Stage C1. Comprehensive experiments on standard BLI datasets for diverse languages and different experimental setups demonstrate substantial gains achieved by our framework. While the BLI method from Stage C1 already yields substantial gains over all state-of-the-art BLI methods in our comparison, even stronger improvements are met with the full two-stage framework: e.g., we report gains for 112/112 BLI setups, spanning 28 language pairs.
Visual question answering (VQA) is one of the crucial vision-and-language tasks. Yet, the bulk of research until recently has focused only on the English language due to the lack of appropriate evaluation resources. Previous work on cross-lingual VQA has reported poor zero-shot transfer performance of current multilingual multimodal Transformers and large gaps to monolingual performance, attributed mostly to misalignment of text embeddings between the source and target languages, without providing any additional deeper analyses. In this work, we delve deeper and address different aspects of cross-lingual VQA holistically, aiming to understand the impact of input data, fine-tuning and evaluation regimes, and interactions between the two modalities in cross-lingual setups. 1) We tackle low transfer performance via novel methods that substantially reduce the gap to monolingual English performance, yielding +10 accuracy points over existing transfer methods. 2) We study and dissect cross-lingual VQA across different question types of varying complexity, across different multilingual multi-modal Transformers, and in zero-shot and few-shot scenarios. 3) We further conduct extensive analyses on modality biases in training data and models, aimed to further understand why zero-shot performance gaps remain for some question types and languages. We hope that the novel methods and detailed analyses will guide further progress in multilingual VQA.