Narrative summarization aims to produce a distilled version of a narrative to describe its most salient events and characters. Summarizing a narrative is challenging as it requires an understanding of event causality and character behaviors. To encourage research in this direction, we propose NarraSum, a large-scale narrative summarization dataset. It contains 122K narrative documents, which are collected from plot descriptions of movies and TV episodes with diverse genres, and their corresponding abstractive summaries. Experiments show that there is a large performance gap between humans and the state-of-the-art summarization models on NarraSum. We hope that this dataset will promote future research in summarization, as well as broader studies of natural language understanding and generation. The dataset is available at https://github.com/zhaochaocs/narrasum.
Detecting sarcasm and verbal irony from people's subjective statements is crucial to understanding their intended meanings and real sentiments and positions in social scenarios. This paper describes the X-PuDu system that participated in SemEval-2022 Task 6, iSarcasmEval - Intended Sarcasm Detection in English and Arabic, which aims at detecting intended sarcasm in various settings of natural language understanding. Our solution finetunes pre-trained language models, such as ERNIE-M and DeBERTa, under the multilingual settings to recognize the irony from Arabic and English texts. Our system ranked second out of 43, and ninth out of 32 in Task A: one-sentence detection in English and Arabic; fifth out of 22 in Task B: binary multi-label classification in English; first out of 16, and fifth out of 13 in Task C: sentence-pair detection in English and Arabic.
Public conversations on Twitter comprise many pertinent topics including disasters, protests, politics, propaganda, sports, climate change, epidemics/pandemic outbreaks, etc., that can have both regional and global aspects. Spatial discourse analysis rely on geographical data. However, today less than 1% of tweets are geotagged; in both cases--point location or bounding place information. A major issue with tweets is that Twitter users can be at location A and exchange conversations specific to location B, which we call the Location A/B problem. The problem is considered solved if location entities can be classified as either origin locations (Location As) or non-origin locations (Location Bs). In this work, we propose a simple yet effective framework--the True Origin Model--to address the problem that uses machine-level natural language understanding to identify tweets that conceivably contain their origin location information. The model achieves promising accuracy at country (80%), state (67%), city (58%), county (56%) and district (64%) levels with support from a Location Extraction Model as basic as the CoNLL-2003-based RoBERTa. We employ a tweet contexualizer (locBERT) which is one of the core components of the proposed model, to investigate multiple tweets' distributions for understanding Twitter users' tweeting behavior in terms of mentioning origin and non-origin locations. We also highlight a major concern with the currently regarded gold standard test set (ground truth) methodology, introduce a new data set, and identify further research avenues for advancing the area.
Most spoken language understanding systems use a pipeline approach composed of an automatic speech recognition interface and a natural language understanding module. This approach forces hard decisions when converting continuous inputs into discrete language symbols. Instead, we propose a representation model to encode speech in rich bidirectional encodings that can be used for downstream tasks such as intent prediction. The approach uses a masked language modelling objective to learn the representations, and thus benefits from both the left and right contexts. We show that the performance of the resulting encodings before fine-tuning is better than comparable models on multiple datasets, and that fine-tuning the top layers of the representation model improves the current state of the art on the Fluent Speech Command dataset, also in a low-data regime, when a limited amount of labelled data is used for training. Furthermore, we propose class attention as a spoken language understanding module, efficient both in terms of speed and number of parameters. Class attention can be used to visually explain the predictions of our model, which goes a long way in understanding how the model makes predictions. We perform experiments in English and in Dutch.
Recent work has demonstrated substantial gains in pre-training large-language models (LLMs) followed by supervised fine-tuning on the downstream task. In this paper, we evaluate the performance of the GPT-neo model using $6$ commonsense reasoning benchmark tasks. We aim to examine the performance of smaller models using the GPT-neo models against several larger model baselines such as GPT-$3$, Llama-$2$, MPT and Falcon. Upon fine-tuning with the appropriate set of hyperparameters, our model achieves competitive accuracy on several tasks. We also investigate and substantiate our results using attention-head visualization to better understand the model performance. Finally, we conduct various robustness tests using various methods to gauge the model performance under numerous settings.
Natural language understanding (NLU) has two core tasks: intent classification and slot filling. The success of pre-training language models resulted in a significant breakthrough in the two tasks. One of the promising solutions called BERT can jointly optimize the two tasks. We note that BERT-based models convert each complex token into multiple sub-tokens by wordpiece algorithm, which generates a mismatch between the lengths of the tokens and the labels. This leads to BERT-based models do not do well in label prediction which limits model performance improvement. Many existing models can be compatible with this issue but some hidden semantic information is discarded in the fine-tuning process. We address the problem by introducing a novel joint method on top of BERT which explicitly models the multiple sub-tokens features after wordpiece tokenization, thereby contributing to the two tasks. Our method can well extract the contextual features from complex tokens by the proposed sub-words attention adapter (SAA), which preserves overall utterance information. Additionally, we propose an intent attention adapter (IAA) to obtain the full sentence features to aid users to predict intent. Experimental results confirm that our proposed model is significantly improved on two public benchmark datasets. In particular, the slot filling F1 score is improved from 96.1 to 98.2 (2.1% absolute) on the Airline Travel Information Systems (ATIS) dataset.
Legal documents are typically long and written in legalese, which makes it particularly difficult for laypeople to understand their rights and duties. While natural language understanding technologies can be valuable in supporting such understanding in the legal domain, the limited availability of datasets annotated for deontic modalities in the legal domain, due to the cost of hiring experts and privacy issues, is a bottleneck. To this end, we introduce, LEXDEMOD, a corpus of English contracts annotated with deontic modality expressed with respect to a contracting party or agent along with the modal triggers. We benchmark this dataset on two tasks: (i) agent-specific multi-label deontic modality classification, and (ii) agent-specific deontic modality and trigger span detection using Transformer-based (Vaswani et al., 2017) language models. Transfer learning experiments show that the linguistic diversity of modal expressions in LEXDEMOD generalizes reasonably from lease to employment and rental agreements. A small case study indicates that a model trained on LEXDEMOD can detect red flags with high recall. We believe our work offers a new research direction for deontic modality detection in the legal domain.
Contextual synonym knowledge is crucial for those similarity-oriented tasks whose core challenge lies in capturing semantic similarity between entities in their contexts, such as entity linking and entity matching. However, most Pre-trained Language Models (PLMs) lack synonym knowledge due to inherent limitations of their pre-training objectives such as masked language modeling (MLM). Existing works which inject synonym knowledge into PLMs often suffer from two severe problems: (i) Neglecting the ambiguity of synonyms, and (ii) Undermining semantic understanding of original PLMs, which is caused by inconsistency between the exact semantic similarity of the synonyms and the broad conceptual relevance learned from the original corpus. To address these issues, we propose PICSO, a flexible framework that supports the injection of contextual synonym knowledge from multiple domains into PLMs via a novel entity-aware Adapter which focuses on the semantics of the entities (synonyms) in the contexts. Meanwhile, PICSO stores the synonym knowledge in additional parameters of the Adapter structure, which prevents it from corrupting the semantic understanding of the original PLM. Extensive experiments demonstrate that PICSO can dramatically outperform the original PLMs and the other knowledge and synonym injection models on four different similarity-oriented tasks. In addition, experiments on GLUE prove that PICSO also benefits general natural language understanding tasks. Codes and data will be public.
Conversations emerge as the primary media for exchanging ideas and conceptions. From the listener's perspective, identifying various affective qualities, such as sarcasm, humour, and emotions, is paramount for comprehending the true connotation of the emitted utterance. However, one of the major hurdles faced in learning these affect dimensions is the presence of figurative language, viz. irony, metaphor, or sarcasm. We hypothesize that any detection system constituting the exhaustive and explicit presentation of the emitted utterance would improve the overall comprehension of the dialogue. To this end, we explore the task of Sarcasm Explanation in Dialogues, which aims to unfold the hidden irony behind sarcastic utterances. We propose MOSES, a deep neural network, which takes a multimodal (sarcastic) dialogue instance as an input and generates a natural language sentence as its explanation. Subsequently, we leverage the generated explanation for various natural language understanding tasks in a conversational dialogue setup, such as sarcasm detection, humour identification, and emotion recognition. Our evaluation shows that MOSES outperforms the state-of-the-art system for SED by an average of ~2% on different evaluation metrics, such as ROUGE, BLEU, and METEOR. Further, we observe that leveraging the generated explanation advances three downstream tasks for affect classification - an average improvement of ~14% F1-score in the sarcasm detection task and ~2% in the humour identification and emotion recognition task. We also perform extensive analyses to assess the quality of the results.
Can we teach natural language understanding models to track their beliefs through intermediate points in text? We propose a representation learning framework called breakpoint modeling that allows for learning of this type. Given any text encoder and data marked with intermediate states (breakpoints) along with corresponding textual queries viewed as true/false propositions (i.e., the candidate beliefs of a model, consisting of information changing through time) our approach trains models in an efficient and end-to-end fashion to build intermediate representations that facilitate teaching and direct querying of beliefs at arbitrary points alongside solving other end tasks. To show the benefit of our approach, we experiment with a diverse set of NLU tasks including relational reasoning on CLUTRR and narrative understanding on bAbI. Using novel belief prediction tasks for both tasks, we show the benefit of our main breakpoint transformer, based on T5, over conventional representation learning approaches in terms of processing efficiency, prediction accuracy and prediction consistency, all with minimal to no effect on corresponding QA end tasks. To show the feasibility of incorporating our belief tracker into more complex reasoning pipelines, we also obtain SOTA performance on the three-tiered reasoning challenge for the TRIP benchmark (around 23-32% absolute improvement on Tasks 2-3).
Pre-trained language models (PLMs) are known to improve the generalization performance of natural language understanding models by leveraging large amounts of data during the pre-training phase. However, the out-of-distribution (OOD) generalization problem remains a challenge in many NLP tasks, limiting the real-world deployment of these methods. This paper presents the first attempt at creating a unified benchmark named GLUE-X for evaluating OOD robustness in NLP models, highlighting the importance of OOD robustness and providing insights on how to measure the robustness of a model and how to improve it. The benchmark includes 13 publicly available datasets for OOD testing, and evaluations are conducted on 8 classic NLP tasks over 21 popularly used PLMs, including GPT-3 and GPT-3.5. Our findings confirm the need for improved OOD accuracy in NLP tasks, as significant performance degradation was observed in all settings compared to in-distribution (ID) accuracy.
Pre-trained Language Models (PLMs) which are trained on large text corpus via self-supervised learning method, have yielded promising performance on various tasks in Natural Language Processing (NLP). However, though PLMs with huge parameters can effectively possess rich knowledge learned from massive training text and benefit downstream tasks at the fine-tuning stage, they still have some limitations such as poor reasoning ability due to the lack of external knowledge. Research has been dedicated to incorporating knowledge into PLMs to tackle these issues. In this paper, we present a comprehensive review of Knowledge Enhanced Pre-trained Language Models (KE-PLMs) to provide a clear insight into this thriving field. We introduce appropriate taxonomies respectively for Natural Language Understanding (NLU) and Natural Language Generation (NLG) to highlight these two main tasks of NLP. For NLU, we divide the types of knowledge into four categories: linguistic knowledge, text knowledge, knowledge graph (KG), and rule knowledge. The KE-PLMs for NLG are categorized into KG-based and retrieval-based methods. Finally, we point out some promising future directions of KE-PLMs.
Many recent perturbation studies have found unintuitive results on what does and does not matter when performing Natural Language Understanding (NLU) tasks in English. Coding properties, such as the order of words, can often be removed through shuffling without impacting downstream performances. Such insight may be used to direct future research into English NLP models. As many improvements in multilingual settings consist of wholesale adaptation of English approaches, it is important to verify whether those studies replicate or not in multilingual settings. In this work, we replicate a study on the importance of local structure, and the relative unimportance of global structure, in a multilingual setting. We find that the phenomenon observed on the English language broadly translates to over 120 languages, with a few caveats.
While a substantial body of prior work has explored adversarial example generation for natural language understanding tasks, these examples are often unrealistic and diverge from the real-world data distributions. In this work, we introduce a two-stage adversarial example generation framework (NaturalAdversaries), for designing adversaries that are effective at fooling a given classifier and demonstrate natural-looking failure cases that could plausibly occur during in-the-wild deployment of the models. At the first stage a token attribution method is used to summarize a given classifier's behaviour as a function of the key tokens in the input. In the second stage a generative model is conditioned on the key tokens from the first stage. NaturalAdversaries is adaptable to both black-box and white-box adversarial attacks based on the level of access to the model parameters. Our results indicate these adversaries generalize across domains, and offer insights for future research on improving robustness of neural text classification models.
Personal assistants, automatic speech recognizers and dialogue understanding systems are becoming more critical in our interconnected digital world. A clear example is air traffic control (ATC) communications. ATC aims at guiding aircraft and controlling the airspace in a safe and optimal manner. These voice-based dialogues are carried between an air traffic controller (ATCO) and pilots via very-high frequency radio channels. In order to incorporate these novel technologies into ATC (low-resource domain), large-scale annotated datasets are required to develop the data-driven AI systems. Two examples are automatic speech recognition (ASR) and natural language understanding (NLU). In this paper, we introduce the ATCO2 corpus, a dataset that aims at fostering research on the challenging ATC field, which has lagged behind due to lack of annotated data. The ATCO2 corpus covers 1) data collection and pre-processing, 2) pseudo-annotations of speech data, and 3) extraction of ATC-related named entities. The ATCO2 corpus is split into three subsets. 1) ATCO2-test-set corpus contains 4 hours of ATC speech with manual transcripts and a subset with gold annotations for named-entity recognition (callsign, command, value). 2) The ATCO2-PL-set corpus consists of 5281 hours of unlabeled ATC data enriched with automatic transcripts from an in-domain speech recognizer, contextual information, speaker turn information, signal-to-noise ratio estimate and English language detection score per sample. Both available for purchase through ELDA at http://catalog.elra.info/en-us/repository/browse/ELRA-S0484. 3) The ATCO2-test-set-1h corpus is a one-hour subset from the original test set corpus, that we are offering for free at https://www.atco2.org/data. We expect the ATCO2 corpus will foster research on robust ASR and NLU not only in the field of ATC communications but also in the general research community.
The advances in language-based Artificial Intelligence (AI) technologies applied to build educational applications can present AI for social-good opportunities with a broader positive impact. Across many disciplines, enhancing the quality of mathematics education is crucial in building critical thinking and problem-solving skills at younger ages. Conversational AI systems have started maturing to a point where they could play a significant role in helping students learn fundamental math concepts. This work presents a task-oriented Spoken Dialogue System (SDS) built to support play-based learning of basic math concepts for early childhood education. The system has been evaluated via real-world deployments at school while the students are practicing early math concepts with multimodal interactions. We discuss our efforts to improve the SDS pipeline built for math learning, for which we explore utilizing MathBERT representations for potential enhancement to the Natural Language Understanding (NLU) module. We perform an end-to-end evaluation using real-world deployment outputs from the Automatic Speech Recognition (ASR), Intent Recognition, and Dialogue Manager (DM) components to understand how error propagation affects the overall performance in real-world scenarios.
Pre-trained multilingual language models play an important role in cross-lingual natural language understanding tasks. However, existing methods did not focus on learning the semantic structure of representation, and thus could not optimize their performance. In this paper, we propose Multi-level Multilingual Knowledge Distillation (MMKD), a novel method for improving multilingual language models. Specifically, we employ a teacher-student framework to adopt rich semantic representation knowledge in English BERT. We propose token-, word-, sentence-, and structure-level alignment objectives to encourage multiple levels of consistency between source-target pairs and correlation similarity between teacher and student models. We conduct experiments on cross-lingual evaluation benchmarks including XNLI, PAWS-X, and XQuAD. Experimental results show that MMKD outperforms other baseline models of similar size on XNLI and XQuAD and obtains comparable performance on PAWS-X. Especially, MMKD obtains significant performance gains on low-resource languages.
The full power of human language-based communication cannot be realized without negation. All human languages have some form of negation. Despite this, negation remains a challenging phenomenon for current natural language understanding systems. To facilitate the future development of models that can process negation effectively, we present CONDAQA, the first English reading comprehension dataset which requires reasoning about the implications of negated statements in paragraphs. We collect paragraphs with diverse negation cues, then have crowdworkers ask questions about the implications of the negated statement in the passage. We also have workers make three kinds of edits to the passage -- paraphrasing the negated statement, changing the scope of the negation, and reversing the negation -- resulting in clusters of question-answer pairs that are difficult for models to answer with spurious shortcuts. CONDAQA features 14,182 question-answer pairs with over 200 unique negation cues and is challenging for current state-of-the-art models. The best performing model on CONDAQA (UnifiedQA-v2-3b) achieves only 42% on our consistency metric, well below human performance which is 81%. We release our dataset, along with fully-finetuned, few-shot, and zero-shot evaluations, to facilitate the development of future NLP methods that work on negated language.
In recent years, pretrained models revolutionized the paradigm of natural language understanding (NLU), where we append a randomly initialized classification head after the pretrained backbone, e.g. BERT, and finetune the whole model. As the pretrained backbone makes a major contribution to the improvement, we naturally expect a good pretrained classification head can also benefit the training. However, the final-layer output of the backbone, i.e. the input of the classification head, will change greatly during finetuning, making the usual head-only pretraining (LP-FT) ineffective. In this paper, we find that parameter-efficient tuning makes a good classification head, with which we can simply replace the randomly initialized heads for a stable performance gain. Our experiments demonstrate that the classification head jointly pretrained with parameter-efficient tuning consistently improves the performance on 9 tasks in GLUE and SuperGLUE.
Debiasing language models from unwanted behaviors in Natural Language Understanding tasks is a topic with rapidly increasing interest in the NLP community. Spurious statistical correlations in the data allow models to perform shortcuts and avoid uncovering more advanced and desirable linguistic features. A multitude of effective debiasing approaches has been proposed, but flexibility remains a major issue. For the most part, models must be retrained to find a new set of weights with debiased behavior. We propose a new debiasing method in which we identify debiased pruning masks that can be applied to a finetuned model. This enables the selective and conditional application of debiasing behaviors. We assume that bias is caused by a certain subset of weights in the network; our method is, in essence, a mask search to identify and remove biased weights. Our masks show equivalent or superior performance to the standard counterparts, while offering important benefits. Pruning masks can be stored with high efficiency in memory, and it becomes possible to switch among several debiasing behaviors (or revert back to the original biased model) at inference time. Finally, it opens the doors to further research on how biases are acquired by studying the generated masks. For example, we observed that the early layers and attention heads were pruned more aggressively, possibly hinting towards the location in which biases may be encoded.