PyTorchKR Papers

최신 논문

총 5,082편
태그·기간 등 상세 필터

카테고리 탐색

로그인 후 이 태그 구독

arXiv:2112.08726공식 코드 · Papers with Code

NeuroLogic A*esque Decoding: Constrained Text Generation with Lookahead Heuristics

The dominant paradigm for neural text generation is left-to-right decoding from autoregressive language models. Constrained or controllable generation under complex lexical constraints, however, requires foresight to plan ahead feasible future paths. Drawing inspiration from the A* search algorithm, we propose NeuroLogic A*esque, a decoding algorithm that incorporates heuristic estimates of future cost. We develop efficient lookahead heuristics that are efficient for large-scale language models, making our method a drop-in replacement for common techniques such as beam search and top-k sampling. To enable constrained generation, we build on NeuroLogic decoding (Lu et al., 2021), combining its flexibility in incorporating logical constraints with A*esque estimates of future constraint satisfaction. Our approach outperforms competitive baselines on five generation tasks, and achieves new state-of-the-art performance on table-to-text generation, constrained machine translation, and keyword-constrained generation. The improvements are particularly notable on tasks that require complex constraint satisfaction or in few-shot or zero-shot settings. NeuroLogic A*esque illustrates the power of decoding for improving and enabling new capabilities of large-scale language models.

arXiv:2112.11642

Joint-training on Symbiosis Networks for Deep Nueral Machine Translation models

Deep encoders have been proven to be effective in improving neural machine translation (NMT) systems, but it reaches the upper bound of translation quality when the number of encoder layers exceeds 18. Worse still, deeper networks consume a lot of memory, making it impossible to train efficiently. In this paper, we present Symbiosis Networks, which include a full network as the Symbiosis Main Network (M-Net) and another shared sub-network with the same structure but less layers as the Symbiotic Sub Network (S-Net). We adopt Symbiosis Networks on Transformer-deep (m-n) architecture and define a particular regularization loss $\mathcal{L}_{\tau}$ between the M-Net and S-Net in NMT. We apply joint-training on the Symbiosis Networks and aim to improve the M-Net performance. Our proposed training strategy improves Transformer-deep (12-6) by 0.61, 0.49 and 0.69 BLEU over the baselines under classic training on WMT'14 EN->DE, DE->EN and EN->FR tasks. Furthermore, our Transformer-deep (12-6) even outperforms classic Transformer-deep (18-6).

arXiv:2112.11640

Self-Distillation Mixup Training for Non-autoregressive Neural Machine Translation

Recently, non-autoregressive (NAT) models predict outputs in parallel, achieving substantial improvements in generation speed compared to autoregressive (AT) models. While performing worse on raw data, most NAT models are trained as student models on distilled data generated by AT teacher models, which is known as sequence-level Knowledge Distillation. An effective training strategy to improve the performance of AT models is Self-Distillation Mixup (SDM) Training, which pre-trains a model on raw data, generates distilled data by the pre-trained model itself and finally re-trains a model on the combination of raw data and distilled data. In this work, we aim to view SDM for NAT models, but find directly adopting SDM to NAT models gains no improvements in terms of translation quality. Through careful analysis, we observe the invalidation is correlated to Modeling Diversity and Confirmation Bias between the AT teacher model and the NAT student models. Based on these findings, we propose an enhanced strategy named SDMRT by adding two stages to classic SDM: one is Pre-Rerank on self-distilled data, the other is Fine-Tune on Filtered teacher-distilled data. Our results outperform baselines by 0.6 to 1.2 BLEU on multiple NAT models. As another bonus, for Iterative Refinement NAT models, our methods can outperform baselines within half iteration number, which means 2X acceleration.

arXiv:2112.11632

Diformer: Directional Transformer for Neural Machine Translation

Autoregressive (AR) and Non-autoregressive (NAR) models have their own superiority on the performance and latency, combining them into one model may take advantage of both. Current combination frameworks focus more on the integration of multiple decoding paradigms with a unified generative model, e.g. Masked Language Model. However, the generalization can be harmful to the performance due to the gap between training objective and inference. In this paper, we aim to close the gap by preserving the original objective of AR and NAR under a unified framework. Specifically, we propose the Directional Transformer (Diformer) by jointly modelling AR and NAR into three generation directions (left-to-right, right-to-left and straight) with a newly introduced direction variable, which works by controlling the prediction of each token to have specific dependencies under that direction. The unification achieved by direction successfully preserves the original dependency assumption used in AR and NAR, retaining both generalization and performance. Experiments on 4 WMT benchmarks demonstrate that Diformer outperforms current united-modelling works with more than 1.5 BLEU points for both AR and NAR decoding, and is also competitive to the state-of-the-art independent AR and NAR models.

arXiv:2112.10898

Load-balanced Gather-scatter Patterns for Sparse Deep Neural Networks

Deep neural networks (DNNs) have been proven to be effective in solving many real-life problems, but its high computation cost prohibits those models from being deployed to edge devices. Pruning, as a method to introduce zeros to model weights, has shown to be an effective method to provide good trade-offs between model accuracy and computation efficiency, and is a widely-used method to generate compressed models. However, the granularity of pruning makes important trade-offs. At the same sparsity level, a coarse-grained structured sparse pattern is more efficient on conventional hardware but results in worse accuracy, while a fine-grained unstructured sparse pattern can achieve better accuracy but is inefficient on existing hardware. On the other hand, some modern processors are equipped with fast on-chip scratchpad memories and gather/scatter engines that perform indirect load and store operations on such memories. In this work, we propose a set of novel sparse patterns, named gather-scatter (GS) patterns, to utilize the scratchpad memories and gather/scatter engines to speed up neural network inferences. Correspondingly, we present a compact sparse format. The proposed set of sparse patterns, along with a novel pruning methodology, address the load imbalance issue and result in models with quality close to unstructured sparse models and computation efficiency close to structured sparse models. Our experiments show that GS patterns consistently make better trade-offs between accuracy and computation efficiency compared to conventional structured sparse patterns. GS patterns can reduce the runtime of the DNN components by two to three times at the same accuracy levels. This is confirmed on three different deep learning tasks and popular models, namely, GNMT for machine translation, ResNet50 for image recognition, and Japser for acoustic speech recognition.

arXiv:2112.10543

Spiral Language Modeling

In almost all text generation applications, word sequences are constructed in a left-to-right (L2R) or right-to-left (R2L) manner, as natural language sentences are written either L2R or R2L. However, we find that the natural language written order is not essential for text generation. In this paper, we propose Spiral Language Modeling (SLM), a general approach that enables one to construct natural language sentences beyond the L2R and R2L order. SLM allows one to form natural language text by starting from an arbitrary token inside the result text and expanding the rest tokens around the selected ones. It makes the decoding order a new optimization objective besides the language model perplexity, which further improves the diversity and quality of the generated text. Furthermore, SLM makes it possible to manipulate the text construction process by selecting a proper starting token. SLM also introduces generation orderings as additional regularization to improve model robustness in low-resource scenarios. Experiments on 8 widely studied Neural Machine Translation (NMT) tasks show that SLM is constantly effective with up to 4.7 BLEU increase comparing to the conventional L2R decoding approach.

arXiv:2112.10255

Task-Oriented Multi-User Semantic Communications

While semantic communications have shown the potential in the case of single-modal single-users, its applications to the multi-user scenario remain limited. In this paper, we investigate deep learning (DL) based multi-user semantic communication systems for transmitting single-modal data and multimodal data, respectively. We will adopt three intelligent tasks, including, image retrieval, machine translation, and visual question answering (VQA) as the transmission goal of semantic communication systems. We will then propose a Transformer based unique framework to unify the structure of transmitters for different tasks. For the single-modal multi-user system, we will propose two Transformer based models, named, DeepSC-IR and DeepSC-MT, to perform image retrieval and machine translation, respectively. In this case, DeepSC-IR is trained to optimize the distance in embedding space between images and DeepSC-MT is trained to minimize the semantic errors by recovering the semantic meaning of sentences. For the multimodal multi-user system, we develop a Transformer enabled model, named, DeepSC-VQA, for the VQA task by extracting text-image information at the transmitters and fusing it at the receiver. In particular, a novel layer-wise Transformer is designed to help fuse multimodal data by adding connection between each of the encoder and decoder layers. Numerical results will show that the proposed models are superior to traditional communications in terms of the robustness to channels, computational complexity, transmission delay, and the task-execution performance at various task-specific metrics.

arXiv:2112.10668공식 코드 · Papers with Code

Few-shot Learning with Multilingual Language Models

Large-scale generative language models such as GPT-3 are competitive few-shot learners. While these models are known to be able to jointly represent many different languages, their training data is dominated by English, potentially limiting their cross-lingual generalization. In this work, we train multilingual generative language models on a corpus covering a diverse set of languages, and study their few- and zero-shot learning capabilities in a wide range of tasks. Our largest model with 7.5 billion parameters sets new state of the art in few-shot learning in more than 20 representative languages, outperforming GPT-3 of comparable size in multilingual commonsense reasoning (with +7.4% absolute accuracy improvement in 0-shot settings and +9.4% in 4-shot settings) and natural language inference (+5.4% in each of 0-shot and 4-shot settings). On the FLORES-101 machine translation benchmark, our model outperforms GPT-3 on 171 out of 182 directions with 32 training examples, while surpassing the official supervised baseline in 45 directions. We conduct an in-depth analysis of different multilingual prompting approaches, showing in particular that strong few-shot learning performance across languages can be achieved via cross-lingual transfer through both templates and demonstration examples. Finally, we evaluate our models in social value tasks such as hate speech detection in five languages and find it has limitations similar to comparable sized GPT-3 models.

arXiv:2112.09526공식 코드 · Papers with Code

Challenge Dataset of Cognates and False Friend Pairs from Indian Languages

Cognates are present in multiple variants of the same text across different languages (e.g., "hund" in German and "hound" in English language mean "dog"). They pose a challenge to various Natural Language Processing (NLP) applications such as Machine Translation, Cross-lingual Sense Disambiguation, Computational Phylogenetics, and Information Retrieval. A possible solution to address this challenge is to identify cognates across language pairs. In this paper, we describe the creation of two cognate datasets for twelve Indian languages, namely Sanskrit, Hindi, Assamese, Oriya, Kannada, Gujarati, Tamil, Telugu, Punjabi, Bengali, Marathi, and Malayalam. We digitize the cognate data from an Indian language cognate dictionary and utilize linked Indian language Wordnets to generate cognate sets. Additionally, we use the Wordnet data to create a False Friends' dataset for eleven language pairs. We also evaluate the efficacy of our dataset using previously available baseline cognate detection approaches. We also perform a manual evaluation with the help of lexicographers and release the curated gold-standard dataset with this paper.

arXiv:2112.08914공식 코드 · Papers with Code

Characterizing and addressing the issue of oversmoothing in neural autoregressive sequence modeling

Neural autoregressive sequence models smear the probability among many possible sequences including degenerate ones, such as empty or repetitive sequences. In this work, we tackle one specific case where the model assigns a high probability to unreasonably short sequences. We define the oversmoothing rate to quantify this issue. After confirming the high degree of oversmoothing in neural machine translation, we propose to explicitly minimize the oversmoothing rate during training. We conduct a set of experiments to study the effect of the proposed regularization on both model distribution and decoding performance. We use a neural machine translation task as the testbed and consider three different datasets of varying size. Our experiments reveal three major findings. First, we can control the oversmoothing rate of the model by tuning the strength of the regularization. Second, by enhancing the oversmoothing loss contribution, the probability and the rank of <eos> token decrease heavily at positions where it is not supposed to be. Third, the proposed regularization impacts the outcome of beam search especially when a large beam is used. The degradation of translation quality (measured in BLEU) with a large beam significantly lessens with lower oversmoothing rate, but the degradation compared to smaller beam sizes remains to exist. From these observations, we conclude that the high degree of oversmoothing is the main reason behind the degenerate case of overly probable short sequences in a neural autoregressive model.

arXiv:2112.08682

Isometric MT: Neural Machine Translation for Automatic Dubbing

Automatic dubbing (AD) is among the machine translation (MT) use cases where translations should match a given length to allow for synchronicity between source and target speech. For neural MT, generating translations of length close to the source length (e.g. within +-10% in character count), while preserving quality is a challenging task. Controlling MT output length comes at a cost to translation quality, which is usually mitigated with a two step approach of generating N-best hypotheses and then re-ranking based on length and quality. This work introduces a self-learning approach that allows a transformer model to directly learn to generate outputs that closely match the source length, in short Isometric MT. In particular, our approach does not require to generate multiple hypotheses nor any auxiliary ranking function. We report results on four language pairs (English - French, Italian, German, Spanish) with a publicly available benchmark. Automatic and manual evaluations show that our method for Isometric MT outperforms more complex approaches proposed in the literature.

arXiv:2112.08548

Isochrony-Aware Neural Machine Translation for Automatic Dubbing

We introduce the task of isochrony-aware machine translation which aims at generating translations suitable for dubbing. Dubbing of a spoken sentence requires transferring the content as well as the speech-pause structure of the source into the target language to achieve audiovisual coherence. Practically, this implies correctly projecting pauses from the source to the target and ensuring that target speech segments have roughly the same duration of the corresponding source speech segments. In this work, we propose implicit and explicit modeling approaches to integrate isochrony information into neural machine translation. Experiments on English-German/French language pairs with automatic metrics show that the simplest of the considered approaches works best. Results are confirmed by human evaluations of translations and dubbed videos.

arXiv:2112.09097공식 코드 · Papers with Code

Learning and Analyzing Generation Order for Undirected Sequence Models

Undirected neural sequence models have achieved performance competitive with the state-of-the-art directed sequence models that generate monotonically from left to right in machine translation tasks. In this work, we train a policy that learns the generation order for a pre-trained, undirected translation model via reinforcement learning. We show that the translations decoded by our learned orders achieve higher BLEU scores than the outputs decoded from left to right or decoded by the learned order from Mansimov et al. (2019) on the WMT'14 German-English translation task. On examples with a maximum source and target length of 30 from De-En, WMT'16 English-Romanian, and WMT'21 English-Chinese translation tasks, our learned order outperforms all heuristic generation orders on four out of six tasks. We next carefully analyze the learned order patterns via qualitative and quantitative analysis. We show that our policy generally follows an outer-to-inner order, predicting the left-most and right-most positions first, and then moving toward the middle while skipping less important words at the beginning. Furthermore, the policy usually predicts positions for a single syntactic constituent structure in consecutive steps. We believe our findings could provide more insights on the mechanism of undirected generation models and encourage further research in this direction. Our code is publicly available at https://github.com/jiangycTarheel/undirected-generation

arXiv:2112.08789공식 코드 · Papers with Code

Harnessing Cross-lingual Features to Improve Cognate Detection for Low-resource Languages

Cognates are variants of the same lexical form across different languages; for example 'fonema' in Spanish and 'phoneme' in English are cognates, both of which mean 'a unit of sound'. The task of automatic detection of cognates among any two languages can help downstream NLP tasks such as Cross-lingual Information Retrieval, Computational Phylogenetics, and Machine Translation. In this paper, we demonstrate the use of cross-lingual word embeddings for detecting cognates among fourteen Indian Languages. Our approach introduces the use of context from a knowledge graph to generate improved feature representations for cognate detection. We, then, evaluate the impact of our cognate detection mechanism on neural machine translation (NMT), as a downstream task. We evaluate our methods to detect cognates on a challenging dataset of twelve Indian languages, namely, Sanskrit, Hindi, Assamese, Oriya, Kannada, Gujarati, Tamil, Telugu, Punjabi, Bengali, Marathi, and Malayalam. Additionally, we create evaluation datasets for two more Indian languages, Konkani and Nepali. We observe an improvement of up to 18% points, in terms of F-score, for cognate detection. Furthermore, we observe that cognates extracted using our method help improve NMT quality by up to 2.76 BLEU. We also release our code, newly constructed datasets and cross-lingual models publicly.

arXiv:2112.08288공식 코드 · Papers with Code

Improving Both Domain Robustness and Domain Adaptability in Machine Translation

We consider two problems of NMT domain adaptation using meta-learning. First, we want to reach domain robustness, i.e., we want to reach high quality on both domains seen in the training data and unseen domains. Second, we want our systems to be adaptive, i.e., making it possible to finetune systems with just hundreds of in-domain parallel sentences. We study the domain adaptability of meta-learning when improving the domain robustness of the model. In this paper, we propose a novel approach, RMLNMT (Robust Meta-Learning Framework for Neural Machine Translation Domain Adaptation), which improves the robustness of existing meta-learning models. More specifically, we show how to use a domain classifier in curriculum learning and we integrate the word-level domain mixing model into the meta-learning framework with a balanced sampling strategy. Experiments on English$\rightarrow$German and English$\rightarrow$Chinese translation show that RMLNMT improves in terms of both domain robustness and domain adaptability in seen and unseen domains. Our source code is available at https://github.com/lavine-lmu/RMLNMT.

arXiv:2112.08152

Faster Nearest Neighbor Machine Translation

$k$NN based neural machine translation ($k$NN-MT) has achieved state-of-the-art results in a variety of MT tasks. One significant shortcoming of $k$NN-MT lies in its inefficiency in identifying the $k$ nearest neighbors of the query representation from the entire datastore, which is prohibitively time-intensive when the datastore size is large. In this work, we propose \textbf{Faster $k$NN-MT} to address this issue. The core idea of Faster $k$NN-MT is to use a hierarchical clustering strategy to approximate the distance between the query and a data point in the datastore, which is decomposed into two parts: the distance between the query and the center of the cluster that the data point belongs to, and the distance between the data point and the cluster center. We propose practical ways to compute these two parts in a significantly faster manner. Through extensive experiments on different MT benchmarks, we show that \textbf{Faster $k$NN-MT} is faster than Fast $k$NN-MT \citep{meng2021fast} and only slightly (1.2 times) slower than its vanilla counterpart while preserving model performance as $k$NN-MT. Faster $k$NN-MT enables the deployment of $k$NN-MT models on real-world MT services.

arXiv:2112.08191

Lesan -- Machine Translation for Low Resource Languages

Millions of people around the world can not access content on the Web because most of the content is not readily available in their language. Machine translation (MT) systems have the potential to change this for many languages. Current MT systems provide very accurate results for high resource language pairs, e.g., German and English. However, for many low resource languages, MT is still under active research. The key challenge is lack of datasets to build these systems. We present Lesan, an MT system for low resource languages. Our pipeline solves the key bottleneck to low resource MT by leveraging online and offline sources, a custom OCR system for Ethiopic and an automatic alignment module. The final step in the pipeline is a sequence to sequence model that takes parallel corpus as input and gives us a translation model. Lesan's translation model is based on the Transformer architecture. After constructing a base model, back translation, is used to leverage monolingual corpora. Currently Lesan supports translation to and from Tigrinya, Amharic and English. We perform extensive human evaluation and show that Lesan outperforms state-of-the-art systems such as Google Translate and Microsoft Translator across all six pairs. Lesan is freely available and has served more than 10 million translations so far. At the moment, there are only 217 Tigrinya and 15,009 Amharic Wikipedia articles. We believe that Lesan will contribute towards democratizing access to the Web through MT for millions of people.

arXiv:2112.08185공식 코드 · Papers with Code

Learning Cross-Lingual IR from an English Retriever

We present DR.DECR (Dense Retrieval with Distillation-Enhanced Cross-Lingual Representation), a new cross-lingual information retrieval (CLIR) system trained using multi-stage knowledge distillation (KD). The teacher of DR.DECR relies on a highly effective but computationally expensive two-stage inference process consisting of query translation and monolingual IR, while the student, DR.DECR, executes a single CLIR step. We teach DR.DECR powerful multilingual representations as well as CLIR by optimizing two corresponding KD objectives. Learning useful representations of non-English text from an English-only retriever is accomplished through a cross-lingual token alignment algorithm that relies on the representation capabilities of the underlying multilingual encoders. In both in-domain and zero-shot out-of-domain evaluation, DR.DECR demonstrates far superior accuracy over direct fine-tuning with labeled CLIR data. It is also the best single-model retriever on the XOR-TyDi benchmark at the time of this writing.

arXiv:2112.08087공식 코드 · Papers with Code

Cognition-aware Cognate Detection

Automatic detection of cognates helps downstream NLP tasks of Machine Translation, Cross-lingual Information Retrieval, Computational Phylogenetics and Cross-lingual Named Entity Recognition. Previous approaches for the task of cognate detection use orthographic, phonetic and semantic similarity based features sets. In this paper, we propose a novel method for enriching the feature sets, with cognitive features extracted from human readers' gaze behaviour. We collect gaze behaviour data for a small sample of cognates and show that extracted cognitive features help the task of cognate detection. However, gaze data collection and annotation is a costly task. We use the collected gaze behaviour data to predict cognitive features for a larger sample and show that predicted cognitive features, also, significantly improve the task performance. We report improvements of 10% with the collected gaze features, and 12% using the predicted gaze features, over the previously proposed approaches. Furthermore, we release the collected gaze behaviour data along with our code and cross-lingual models.

arXiv:2112.07660공식 코드 · Papers with Code

Massive-scale Decoding for Text Generation using Lattices

Conditional neural text generation models generate high-quality outputs, but often concentrate around a mode when what we really want is a diverse set of options. We present a search algorithm to construct lattices encoding a massive number of generation options. First, we restructure decoding as a best-first search, which explores the space differently than beam search and improves efficiency by avoiding pruning paths. Second, we revisit the idea of hypothesis recombination: we can identify pairs of similar generation candidates during search and merge them as an approximation. On both summarization and machine translation, we show that our algorithm encodes thousands of diverse options that remain grammatical and high-quality into one lattice. This algorithm provides a foundation for building downstream generation applications on top of massive-scale diverse outputs.