Total: 1059
In the Multi-document summarization (MDS) task, a summary is produced for a given set of documents. A recent line of research introduced the concept of damaging documents, denoting documents that should not be exposed to readers due to various reasons. In the presence of damaging documents, a summarizer is ideally expected to exclude damaging content in its output. Existing metrics evaluate a summary based on aspects such as relevance and consistency with the source documents. We propose to additionally measure the ability of MDS systems to properly handle damaging documents in their input set. To that end, we offer two novel metrics based on lexical similarity and language model likelihood. A set of experiments demonstrates the effectiveness of our metrics in measuring the ability of MDS systems to summarize a set of documents while eliminating damaging content from their summaries.
Abstract Meaning Representation (AMR) parsing aims to extract an abstract semantic graph from a given sentence. The sequence-to-sequence approaches, which linearize the semantic graph into a sequence of nodes and edges and generate the linearized graph directly, have achieved good performance. However, we observed that these approaches suffer from structure loss accumulation during the decoding process, leading to a much lower F1-score for nodes and edges decoded later compared to those decoded earlier. To address this issue, we propose a novel Reverse Graph Linearization (RGL) enhanced framework. RGL defines both default and reverse linearization orders of an AMR graph, where most structures at the back part of the default order appear at the front part of the reversed order and vice versa. RGL incorporates the reversed linearization to the original AMR parser through a two-pass self-distillation mechanism, which guides the model when generating the default linearizations. Our analysis shows that our proposed method significantly mitigates the problem of structure loss accumulation, outperforming the previously best AMR parsing model by 0.8 and 0.5 Smatch scores on the AMR 2.0 and AMR 3.0 dataset, respectively. The code are available at https://github.com/pkunlp-icler/AMR_reverse_graph_linearization.
Song translation requires both translation of lyrics and alignment of music notes so that the resulting verse can be sung to the accompanying melody, which is a challenging problem that has attracted some interests in different aspects of the translation process. In this paper, we propose Lyrics-Melody Translation with Adaptive Grouping (LTAG), a holistic solution to automatic song translation by jointly modeling lyric translation and lyrics-melody alignment. It is a novel encoder-decoder framework that can simultaneously translate the source lyrics and determine the number of aligned notes at each decoding step through an adaptive note grouping module. To address data scarcity, we commissioned a small amount of training data annotated specifically for this task and used large amounts of automatic training data through back-translation. Experiments conducted on an English-Chinese song translation data set show the effectiveness of our model in both automatic and human evaluations.
Transliteration is very important in the Indian language context due to the usage of multiple scripts and the widespread use of romanized inputs. However, few training and evaluation sets are publicly available. We introduce Aksharantar, the largest publicly available transliteration dataset for Indian languages created by mining from monolingual and parallel corpora, as well as collecting data from human annotators. The dataset contains 26 million transliteration pairs for 21 Indic languages from 3 language families using 12 scripts. Aksharantar is 21 times larger than existing datasets and is the first publicly available dataset for 7 languages and 1 language family. We also introduce a test set of 103k word pairs for 19 languages that enables a fine-grained analysis of transliteration models on native origin words, foreign words, frequent words, and rare words. Using the training set, we trained IndicXlit, a multilingual transliteration model that improves accuracy by 15% on the Dakshina test set, and establishes strong baselines on the Aksharantar testset introduced in this work. The models, mining scripts, transliteration guidelines, and datasets are available at https://github.com/AI4Bharat/IndicXlit under open-source licenses.
Transformers have been essential to pretraining success in NLP. While other architectures have been used, downstream accuracy is either significantly worse, or requires attention layers to match standard benchmarks such as GLUE. This work explores pretraining without attention by using recent advances in sequence routing based on state-space models (SSMs). Our proposed model, Bidirectional Gated SSM (BiGS), combines SSM layers with a multiplicative gating architecture that has been effective in simplified sequence modeling architectures. The model learns static layers that do not consider pair-wise interactions. Even so, BiGS is able to match BERT pretraining accuracy on GLUE and can be extended to long-form pretraining of 4096 tokens without approximation. Analysis shows that while the models have similar average accuracy, the approach has different inductive biases than BERT and scales more efficiently to longer sequences.
Time is one of the crucial factors in real-world question answering (QA) problems. However, language models have difficulty understanding the relationships between time specifiers, such as ‘after’ and ‘before’, and numbers, since existing QA datasets do not include sufficient time expressions. To address this issue, we propose a Time-Context aware Question Answering (TCQA) framework. We suggest a Time-Context dependent Span Extraction (TCSE) task, and build a time-context dependent data generation framework for model training. Moreover, we present a metric to evaluate the time awareness of the QA model using TCSE. The TCSE task consists of a question and four sentence candidates classified as correct or incorrect based on time and context. The model is trained to extract the answer span from the sentence that is both correct in time and context. The model trained with TCQA outperforms baseline models up to 8.5 of the F1-score in the TimeQA dataset. Our dataset and code are available at https://github.com/sonjbin/TCQA
Efficiency is a key property to foster inclusiveness and reduce environmental costs, especially in an era of LLMs. In this work, we provide a comprehensive evaluation of efficiency for MT evaluation metrics. Our approach involves replacing computation-intensive transformers with lighter alternatives and employing linear and quadratic approximations for alignment algorithms on top of LLM representations. We evaluate six (reference-free and reference-based) metrics across three MT datasets and examine 16 lightweight transformers. In addition, we look into the training efficiency of metrics like COMET by utilizing adapters. Our results indicate that (a) TinyBERT provides the optimal balance between quality and efficiency, (b) CPU speed-ups are more substantial than those on GPU; (c) WMD approximations yield no efficiency gains while reducing quality and (d) adapters enhance training efficiency (regarding backward pass speed and memory requirements) as well as, in some cases, metric quality. These findings can help to strike a balance between evaluation speed and quality, which is essential for effective NLG systems. Furthermore, our research contributes to the ongoing efforts to optimize NLG evaluation metrics with minimal impact on performance. To our knowledge, ours is the most comprehensive analysis of different aspects of efficiency for MT metrics conducted so far.
Opinion summarization is the task of creating summaries capturing popular opinions from user reviews. In this paper, we introduce Geodesic Summarizer (GeoSumm), a novel system to perform unsupervised extractive opinion summarization. GeoSumm consists of an encoder-decoder based representation learning model that generates topical representations of texts. These representations capture the underlying semantics of the text as a distribution over learnable latent units. GeoSumm generates these topical representations by performing dictionary learning over pre-trained text representations at multiple layers of the decoder. We then use these topical representations to quantify the importance of review sentences using a novel approximate geodesic distance-based scoring mechanism. We use the importance scores to identify popular opinions in order to compose general and aspect-specific summaries. Our proposed model, GeoSumm, achieves strong performance on three opinion summarization datasets. We perform additional experiments to analyze the functioning of our model and showcase the generalization ability of GeoSumm across different domains.
Recent research has shown that static word embeddings can encode words’ frequencies. However, little has been studied about this behavior. In the present work, we study how frequency and semantic similarity relate to one another in static word embeddings, and we assess the impact of this relationship on embedding-based bias metrics. We find that Skip-gram, GloVe and FastText embeddings tend to produce higher similarity between high-frequency words than between other frequency combinations. We show that the association between frequency and similarity also appears when words are randomly shuffled, and holds for different hyperparameter settings. This proves that the patterns we find are neither due to real semantic associations nor to specific parameters choices, and are an artifact produced by the word embeddings. To illustrate how frequencies can affect the measurement of biases related to gender, ethnicity, and affluence, we carry out a controlled experiment that shows that biases can even change sign or reverse their order when word frequencies change.
Counterfactual Data Augmentation (CDA) is a commonly used technique for improving robustness in natural language classifiers. However, one fundamental challenge is how to discover meaningful counterfactuals and efficiently label them, with minimal human labeling cost. Most existing methods either completely rely on human-annotated labels, an expensive process which limits the scale of counterfactual data, or implicitly assume label invariance, which may mislead the model with incorrect labels. In this paper, we present a novel framework that utilizes counterfactual generative models to generate a large number of diverse counterfactuals by actively sampling from regions of uncertainty, and then automatically label them with a learned auxiliary classifier. Our key insight is that we can more correctly label the generated counterfactuals by training a pairwise classifier that interpolates the relationship between the original example and the counterfactual. We demonstrate that with a small amount of human-annotated counterfactual data (10%), we can generate a counterfactual augmentation dataset with learned labels, that provides an 18-20% improvement in robustness and a 14-21% reduction in errors on 6 out-of-domain datasets, comparable to that of a fully human-annotated counterfactual dataset for both sentiment classification and question paraphrase tasks.
Code-switching (CSW) text generation has been receiving increasing attention as a solution to address data scarcity. In light of this growing interest, we need more comprehensive studies comparing different augmentation approaches. In this work, we compare three popular approaches: lexical replacements, linguistic theories, and back-translation (BT), in the context of Egyptian Arabic-English CSW. We assess the effectiveness of the approaches on machine translation and the quality of augmentations through human evaluation. We show that BT and CSW predictive-based lexical replacement, being trained on CSW parallel data, perform best on both tasks. Linguistic theories and random lexical replacement prove to be effective in the lack of CSW parallel data, where both approaches achieve similar results.
In-context learning (ICL) suffers from oversensitivity to the prompt, making it unreliable in real-world scenarios. We study the sensitivity of ICL with respect to multiple perturbation types. First, we find that label bias obscures the true sensitivity, and therefore prior work may have significantly underestimated ICL sensitivity. Second, we observe a strong negative correlation between ICL sensitivity and accuracy: predictions sensitive to perturbations are less likely to be correct. Motivated by these findings, we propose SenSel, a few-shot selective prediction method that abstains from sensitive predictions. Experiments on ten classification datasets show that SenSel consistently outperforms two commonly used confidence-based and entropy-based baselines on abstention decisions.
The widespread existence of wrongly labeled instances is a challenge to distantly supervised relation extraction. Most of the previous works are trained in a bag-level setting to alleviate such noise. However, sentence-level training better utilizes the information than bag-level training, as long as combined with effective noise alleviation. In this work, we propose a novel Transitive Instance Weighting mechanism integrated with the self-distilled BERT backbone, utilizing information in the intermediate outputs to generate dynamic instance weights for denoised sentence-level training. By down-weighting wrongly labeled instances and discounting the weights of easy-to-fit ones, our method can effectively tackle wrongly labeled instances and prevent overfitting. Experiments on both held-out and manual datasets indicate that our method achieves state-of-the-art performance and consistent improvements over the baselines.
Recent approaches to word sense disambiguation (WSD) utilize encodings of the sense gloss (definition), in addition to the input context, to improve performance. In this work we demonstrate that this approach can be adapted for use in multiword expression (MWE) identification by training models which use gloss and context information to filter MWE candidates produced by a rule-based extraction pipeline. Our approach substantially improves precision, outperforming the state-of-the-art in MWE identification on the DiMSUM dataset by up to 1.9 F1 points and achieving competitive results on the PARSEME 1.1 English dataset. Our models also retain most of their WSD performance, showing that a single model can be used for both tasks. Finally, building on similar approaches using Bi-encoders for WSD, we introduce a novel Poly-encoder architecture which improves MWE identification performance.
Incremental learning plays a pivotal role in the context of online knowledge discovery, as it encourages large models (LM) to learn and refresh knowledge continuously. Many approaches have been proposed to simultaneously preserve knowledge from previous tasks while learning new concepts in online NLP applications. In this paper, we primarily focus on learning a more generalized embedding space that could be better transferred to various downstream sequence tasks. The key idea is to learn from both task-agnostic and task-specific embedding aspects so that the inherent challenge of catastrophic forgetting that arises in incremental learning scenarios can be addressed with a more generalized solution. We propose a dual contrastive learning (DCL) based framework to foster the transferability of representations across different tasks, it consists of two key components: firstly, we utilize global contrastive learning that intertwines a task-agnostic strategy for promoting a generalized embedding space; secondly, considering the domain shift from unseen distributions can compromise the quality of learned embeddings. We further incorporate a task-specific attention mechanism to enhance the adaptability of task-specific weight for various emerging tasks and ultimately reduce errors in generic representations. Experiments over various text datasets demonstrate that our work achieves superior performance and outperforms the current state-of-the-art methods.
Community Question-Answering (CQA) portals serve as a valuable tool for helping users within an organization. However, making them accessible to non-English-speaking users continues to be a challenge. Translating questions can broaden the community’s reach, benefiting individuals with similar inquiries in various languages. Translating questions using Neural Machine Translation (NMT) poses more challenges, especially in noisy environments, where the grammatical correctness of the questions is not monitored. These questions may be phrased as statements by non-native speakers, with incorrect subject-verb order and sometimes even missing question marks. Creating a synthetic parallel corpus from such data is also difficult due to its noisy nature. To address this issue, we propose a training methodology that fine-tunes the NMT system only using source-side data. Our approach balances adequacy and fluency by utilizing a loss function that combines BERTScore and Masked Language Model (MLM) Score. Our method surpasses the conventional Maximum Likelihood Estimation (MLE) based fine-tuning approach, which relies on synthetic target data, by achieving a 1.9 BLEU score improvement. Our model exhibits robustness while we add noise to our baseline, and still achieve 1.1 BLEU improvement and large improvements on TER and BLEURT metrics. Our proposed methodology is model-agnostic and is only necessary during the training phase. We make the codes and datasets publicly available at https://www.iitp.ac.in/~ai-nlp-ml/resources.html#DomainAdapt for facilitating further research.
Semi-Markov CRF has been proposed as an alternative to the traditional Linear Chain CRF for text segmentation tasks such as Named Entity Recognition (NER). Unlike CRF, which treats text segmentation as token-level prediction, Semi-CRF considers segments as the basic unit, making it more expressive. However, Semi-CRF suffers from two major drawbacks: (1) quadratic complexity over sequence length, as it operates on every span of the input sequence, and (2) inferior performance compared to CRF for sequence labeling tasks like NER. In this paper, we introduce Filtered Semi-Markov CRF, a variant of Semi-CRF that addresses these issues by incorporating a filtering step to eliminate irrelevant segments, reducing complexity and search space. Our approach is evaluated on several NER benchmarks, where it outperforms both CRF and Semi-CRF while being significantly faster. The implementation of our method is available on Github.
Model pruning methods reduce memory requirements and inference time of large-scale pre-trained language models after deployment. However, the actual pruning procedure is computationally intensive, involving repeated training and pruning until the required sparsity is achieved. This paper combines data pruning with movement pruning for Neural Machine Translation (NMT) to enable efficient fine-pruning. We design a dataset pruning strategy by leveraging cross-entropy scores of individual training instances. We conduct pruning experiments on the task of machine translation from Romanian-to-English and Turkish-to-English, and demonstrate that selecting hard-to-learn examples (top-k) based on training cross-entropy scores outperforms other dataset pruning methods. We empirically demonstrate that data pruning reduces the overall steps required for convergence and the training time of movement pruning. Finally, we perform a series of experiments to tease apart the role of training data during movement pruning and uncover new insights to understand the interplay between data and model pruning in the context of NMT.
One challenge in speech translation is that plenty of spoken content is long-form, but short units are necessary for obtaining high-quality translations. To address this mismatch, we adapt large language models (LLMs) to split long ASR transcripts into segments that can be independently translated so as to maximize the overall translation quality. We overcome the tendency of hallucination in LLMs by incorporating finite-state constraints during decoding; these eliminate invalid outputs without requiring additional training. We discover that LLMs are adaptable to transcripts containing ASR errors through prompt-tuning or fine-tuning. Relative to a state-of-the-art automatic punctuation baseline, our best LLM improves the average BLEU by 2.9 points for English–German, English–Spanish, and English–Arabic TED talk translation in 9 test sets, just by improving segmentation.
Temporal Knowledge Graph Completion (TKGC) under the extrapolation setting aims to predict the missing entity from a fact in the future, posing a challenge that aligns more closely with real-world prediction problems. Existing research mostly encodes entities and relations using sequential graph neural networks applied to recent snapshots. However, these approaches tend to overlook the ability to skip irrelevant snapshots according to entity-related relations in the query and disregard the importance of explicit temporal information. To address this, we propose our model, Re-Temp (Relation-Aware Temporal Representation Learning), which leverages explicit temporal embedding as input and incorporates skip information flow after each timestamp to skip unnecessary information for prediction. Additionally, we introduce a two-phase forward propagation method to prevent information leakage. Through the evaluation on six TKGC (extrapolation) datasets, we demonstrate that our model outperforms all eight recent state-of-the-art models by a significant margin.
Recently, Target-oriented Multimodal Sentiment Classification (TMSC) has gained significant attention among scholars. However, current multimodal models have reached a performance bottleneck. To investigate the causes of this problem, we perform extensive empirical evaluation and in-depth analysis of the datasets to answer the following questions: **Q1**: Are the modalities equally important for TMSC? **Q2**: Which multimodal fusion modules are more effective? **Q3**: Do existing datasets adequately support the research? Our experiments and analyses reveal that the current TMSC systems primarily rely on the textual modality, as most of targets’ sentiments can be determined *solely* by text. Consequently, we point out several directions to work on for the TMSC task in terms of model design and dataset construction. The code and data can be found in https://github.com/Junjie-Ye/RethinkingTMSC.
Conversational agents have become ubiquitous in assisting with daily tasks, and are expected to possess human-like features. One such feature is lexical entrainment (LE), a phenomenon in which speakers in human-human conversations tend to naturally and subconsciously align their lexical choices with those of their interlocutors, leading to more successful and engaging conversations. As an example, if a digital assistant replies “Your appointment for Jinling Noodle Pub is at 7 pm” to the question “When is my reservation for Jinling Noodle Bar today?”, it may feel as though the assistant is trying to correct the speaker, whereas a response of “Your reservation for Jinling Noodle Baris at 7 pm” would likely be perceived as more positive. This highlights the importance of LE in establishing a shared terminology for maximum clarity and reducing ambiguity in conversations. However, we demonstrate in this work that current response generation models do not adequately address this crucial human-like phenomenon. To address this, we propose a new dataset, named MultiWOZ-ENTR, and a measure for LE for conversational systems. Additionally, we suggest a way to explicitly integrate LE into conversational systems with two new tasks, a LE extraction task and a LE generation task. We also present two baseline approaches for the LE extraction task, which aim to detect LE expressions from dialogue contexts
We show that dialogue models can detect errors in their own messages, by calculating the likelihood of replies that are indicative of poor messages. For example, if an agent believes its partner is likely to respond “I don’t understand” to a candidate message, that message may not make sense, so an alternative message should be chosen. We evaluate our approach on a dataset from the game Diplomacy, which contains long dialogues richly grounded in the game state, on which existing models make many errors. We first show that hand-crafted replies can be effective for the task of detecting nonsense in applications as complex as Diplomacy. We then design AutoReply, an algorithm to search for such discriminative replies automatically, given a small number of annotated dialogue examples. We find that AutoReply-generated replies outperform handcrafted replies and perform on par with supervised learning approaches.
The adoption of voice assistants like Alexa or Siri has grown rapidly, allowing users to instantly access information via voice search. Query suggestion is a standard feature of screen-based search experiences, allowing users to explore additional topics. However, this is not trivial to implement in voice-based settings. To enable this, we tackle the novel task of suggesting questions with compact and natural voice hints to allow users to ask follow-up questions. We define the task, ground it in syntactic theory and outline linguistic desiderata for spoken hints. We propose baselines and an approach using sequence-to-sequence Transformers to generate spoken hints from a list of questions. Using a new dataset of 6681 input questions and human written hints, we evaluated the models with automatic metrics and human evaluation. Results show that a naive approach of concatenating suggested questions creates poor voice hints. Our approach, which applies a linguistically-motivated pretraining task was strongly preferred by humans for producing the most natural hints.
Attention mechanisms have become a crucial aspect of deep learning, particularly in natural language processing (NLP) tasks. However, in tasks such as constituency parsing, attention mechanisms can lack the directional information needed to form sentence spans. To address this issue, we propose a Bidirectional masked and N-gram span Attention (BNA) model, which is designed by modifying the attention mechanisms to capture the explicit dependencies between each word and enhance the representation of the output span vectors. The proposed model achieves state-of-the-art performance on the Penn Treebank and Chinese Penn Treebank datasets, with F1 scores of 96.47 and 94.15, respectively. Ablation studies and analysis show that our proposed BNA model effectively captures sentence structure by contextualizing each word in a sentence through bidirectional dependencies and enhancing span representation.