The pre-trained language models have achieved great successes in various natural language understanding (NLU) tasks due to its capacity to capture the deep contextualized information in text by pre-training on large-scale corpora. One of the fundamental components in pre-trained language models is the vocabulary, especially for training multilingual models on many different languages. In the technical report, we present our practices on training multilingual pre-trained language models with BBPE: Byte-Level BPE (i.e., Byte Pair Encoding). In the experiment, we adopted the architecture of NEZHA as the underlying pre-trained language model and the results show that NEZHA trained with byte-level subwords consistently outperforms Google multilingual BERT and vanilla NEZHA by a notable margin in several multilingual NLU tasks. We release the source code of our byte-level vocabulary building tools and the multilingual pre-trained language models.
The University of Texas at Austin（德克萨斯大学奥斯汀分校）；Department of Computer Science（计算机科学系）
The success of pretrained transformer language models (LMs) in natural language processing has led to a wide range of pretraining setups. In particular, these models employ a variety of subword tokenization methods, most notably byte-pair encoding (BPE) (Sennrich et al., 2016; Gage, 1994), the WordPiece method (Schuster and Nakajima, 2012), and unigram language modeling (Kudo, 2018), to segment text. However, to the best of our knowledge, the literature does not contain a direct evaluation of the impact of tokenization on language model pretraining. We analyze differences between BPE and unigr…
Most widely-used pre-trained language models operate on sequences of tokens corresponding to word or subword units. By comparison, token-free models that operate directly on raw text (bytes or characters) have many benefits: they can process text in any language out of the box, they are more robust to noise, and they minimize technical debt by removing complex and error-prone text preprocessing pipelines. Since byte or character sequences are longer than token sequences, past work on token-free models has often introduced new model architectures designed to amortize the cost of operating direc…
Department of Computer Science and Technology, University of Cambridge, U.K.（计算机科学与技术系，剑桥大学，英国）；ALTA Institute, University of Cambridge, U.K.（ALTA研究所，剑桥大学，英国）
Facebook AI Research（脸书人工智能研究院）；New York University（纽约大学）
Almost all existing machine translation models are built on top of character-based vocabularies: characters, subwords or words. Rare characters from noisy text or character-rich languages such as Japanese and Chinese however can unnecessarily take up vocabulary slots and limit its compactness. Representing text at the level of bytes and using the 256 byte set as vocabulary is a potential solution to this issue. High computational cost has however prevented it from being widely deployed or used in practice. In this paper, we investigate byte-level subwords, specifically byte-level BPE (BBPE), w…
Large pretrained language models (LMs) have become the central building block of many NLP applications. Training these models requires ever more computational resources and most of the existing models are trained on English text only. It is exceedingly expensive to train these models in other languages. To alleviate this problem, we introduce a novel method -- called WECHSEL -- to efficiently and effectively transfer pretrained LMs to new languages. WECHSEL can be applied to any model which uses subword-based tokenization and learns an embedding for each subword. The tokenizer of the source mo…
Most of previous work on learning diacritization of the Arabic language relied on training models from scratch. In this paper, we investigate how to leverage pre-trained language models to learn diacritization. We finetune token-free pre-trained multilingual models (ByT5) to learn to predict and insert missing diacritics in Arabic text, a complex task that requires understanding the sentence semantics and the morphological structure of the tokens. We show that we can achieve state-of-the-art on the diacritization task with minimal amount of training and no feature engineering, reducing WER by …
iFLYTEK Research（科大讯飞研究）；State Key Laboratory of Cognitive Intelligence（认知智能国家重点实验室）；Harbin Institute of Technology（哈尔滨工业大学）；Research Center for Social Computing and Information Retrieval (SCIR)（社会计算与信息检索研究中心（SCIR））；iFLYTEK AI Research (Hebei)（科大讯飞AI研究（河北））；University of Maryland（马里兰大学）
Most pre-trained language models (PLMs) construct word representations at subword level with Byte-Pair Encoding (BPE) or its variations, by which OOV (out-of-vocab) words are almost avoidable. However, those methods split a word into subword units and make the representation incomplete and fragile. In this paper, we propose a character-aware pre-trained language model named CharBERT improving on the previous methods (such as BERT, RoBERTa) to tackle these problems. We first construct the contextual word embedding for each token from the sequential character representations, then fuse the repre…
The size of the vocabulary is a central design choice in large pretrained language models, with respect to both performance and memory requirements. Typically, subword tokenization algorithms such as byte pair encoding and WordPiece are used. In this work, we investigate the compatibility of tokenizations for multilingual static and contextualized embedding spaces and propose a measure that reflects the compatibility of tokenizations across languages. Our goal is to prevent incompatible tokenizations, e.g., "wine" (word-level) in English vs.\ "v i n" (character-level) in French, which make it …
RIKEN AIP（理化学研究所先进智能项目中心）；Tohoku University（东北大学）；Heidelberg Institute for Theoretical Studies gGmbH（海德堡理论研究所）
Pretrained contextual and non-contextual subword embeddings have become available in over 250 languages, allowing massively multilingual NLP. However, while there is no dearth of pretrained embeddings, the distinct lack of systematic evaluations makes it difficult for practitioners to choose between them. In this work, we conduct an extensive evaluation comparing non-contextual subword embeddings, namely FastText and BPEmb, and a contextual representation method, namely BERT, on multilingual named entity recognition and part-of-speech tagging. We find that overall, a combination of BERT, BPEmb…
National Institute of Information and Communications Technology（信息通信技术研究所）
Multilingual pretrained models are effective for machine translation and cross-lingual processing because they contain multiple languages in one model. However, they are pretrained after their tokenizers are fixed; therefore it is difficult to change the vocabulary after pretraining. When we extend the pretrained models to new languages, we must modify the tokenizers simultaneously. In this paper, we add new subwords to the SentencePiece tokenizer to apply a multilingual pretrained model to new languages (Inuktitut in this paper). In our experiments, we segmented Inuktitut sentences into subwo…
This paper describes the UMDSub system that participated in Task 2 of SemEval-2018. We developed a system that predicts an emoji given the raw text in a English tweet. The system is a Multi-channel Convolutional Neural Network based on subword embeddings for the representation of tweets. This model improves on character or word based methods by about 2\%. Our system placed 21st of 48 participating systems in the official evaluation.
Multilingual pretrained representations generally rely on subword segmentation algorithms to create a shared multilingual vocabulary. However, standard heuristic algorithms often lead to sub-optimal segmentation, especially for languages with limited amounts of data. In this paper, we take two major steps towards alleviating this problem. First, we demonstrate empirically that applying existing subword regularization methods(Kudo, 2018; Provilkov et al., 2020) during fine-tuning of pre-trained multilingual representations improves the effectiveness of cross-lingual transfer. Second, to take fu…
AIPHES, Heidelberg Institute for Theoretical Studies（海德堡理论研究所 AIPHES）
We present BPEmb, a collection of pre-trained subword unit embeddings in 275 languages, based on Byte-Pair Encoding (BPE). In an evaluation using fine-grained entity typing as testbed, BPEmb performs competitively, and for some languages bet- ter than alternative subword approaches, while requiring vastly fewer resources and no tokenization. BPEmb is available at https://github.com/bheinzerling/bpemb
In this paper, we investigate how the output representation of an end-to-end neural network affects multilingual automatic speech recognition (ASR). We study different representations including character-level, byte-level, byte pair encoding (BPE), and byte-level byte pair encoding (BBPE) representations, and analyze their strengths and weaknesses. We focus on developing a single end-to-end model to support utterance-based bilingual ASR, where speakers do not alternate between two languages in a single utterance but may change languages across utterances. We conduct our experiments on English …
OFA: A Framework of Initializing Unseen Subword Embeddings for Efficient Large-scale Multilingual Continued Pretraining
Center for Information and Language Processing, LMU Munich（慕尼黑大学信息与语言处理中心）；Munich Center for Machine Learning (MCML)（慕尼黑机器学习中心）；Bosch Center for Artificial Intelligence（博世人工智能中心）
Multilingual search can be achieved with subword tokenization. The accuracy of traditional TF-IDF approaches depend on manually curated tokenization, stop words and stemming rules, whereas subword TF-IDF (STF-IDF) can offer higher accuracy without such heuristics. Moreover, multilingual support can be incorporated inherently as part of the subword tokenization model training. XQuAD evaluation demonstrates the advantages of STF-IDF: superior information retrieval accuracy of 85.4% for English and over 80% for 10 other languages without any heuristics-based preprocessing. The software to reprodu…
We describe an LSTM-based model which we call Byte-to-Span (BTS) that reads text as bytes and outputs span annotations of the form [start, length, label] where start positions, lengths, and labels are separate entries in our vocabulary. Because we operate directly on unicode bytes rather than language-specific words or characters, we can analyze text in many languages with a single model. Due to the small vocabulary size, these multilingual models are very compact, but produce results similar to or better than the state-of- the-art in Part-of-Speech tagging and Named Entity Recognition that us…
The Blavatnik School of Computer Science, Tel Aviv University（特拉维夫大学布拉瓦特尼克计算机科学学院）；Facebook AI Research（Facebook人工智能研究院）
Many NLP models operate over sequences of subword tokens produced by hand-crafted tokenization rules and heuristic subword induction algorithms. A simple universal alternative is to represent every computerized text as a sequence of bytes via UTF-8, obviating the need for an embedding layer since there are fewer token types (256) than dimensions. Surprisingly, replacing the ubiquitous embedding layer with one-hot representations of each byte does not hurt performance; experiments on byte-to-byte machine translation from English to 10 different languages show a consistent improvement in BLEU, r…
State-of-the-art models in natural language processing rely on separate rigid subword tokenization algorithms, which limit their generalization ability and adaptation to new settings. In this paper, we propose a new model inductive bias that learns a subword tokenization end-to-end as part of the model. To this end, we introduce a soft gradient-based subword tokenization module (GBST) that automatically learns latent subword representations from characters in a data-driven fashion. Concretely, GBST enumerates candidate subword blocks and learns to score them in a position-wise fashion using a …
Token free approaches have been successfully applied to a series of word and span level tasks. In this work, we compare a byte-level (ByT5) and a wordpiece based (mT5) sequence to sequence model on the 51 languages of the MASSIVE multilingual semantic parsing dataset. We examine multiple experimental settings: (i) zero-shot, (ii) full gold data and (iii) zero-shot with synthetic data. By leveraging a state-of-the-art label projection method for machine translated examples, we are able to reduce the gap in exact match accuracy to only 5 points with respect to a model trained on gold data from a…
Subwords have become the standard units of text in NLP, enabling efficient open-vocabulary models. With algorithms like byte-pair encoding (BPE), subword segmentation is viewed as a preprocessing step applied to the corpus before training. This can lead to sub-optimal segmentations for low-resource languages with complex morphologies. We propose a subword segmental language model (SSLM) that learns how to segment words while being trained for autoregressive language modelling. By unifying subword segmentation and language modelling, our model learns subwords that optimise LM performance. We tr…
Karlsruhe Institute of Technology（卡尔斯鲁厄理工学院）；Carnegie Mellon University（卡内基梅隆大学）
This paper proposes a novel approach to create an unit set for CTC based speech recognition systems. By using Byte Pair Encoding we learn an unit set of an arbitrary size on a given training text. In contrast to using characters or words as units this allows us to find a good trade-off between the size of our unit set and the available training data. We evaluate both Crossword units, that may span multiple word, and Subword units. By combining this approach with decoding methods using a separate language model we are able to achieve state of the art results for grapheme based CTC systems.
Subword tokenization is a commonly used input pre-processing step in most recent NLP models. However, it limits the models' ability to leverage end-to-end task learning. Its frequency-based vocabulary creation compromises tokenization in low-resource languages, leading models to produce suboptimal representations. Additionally, the dependency on a fixed vocabulary limits the subword models' adaptability across languages and domains. In this work, we propose a vocabulary-free neural tokenizer by distilling segmentation information from heuristic-based subword tokenization. We pre-train our char…
We present two end-to-end models: Audio-to-Byte (A2B) and Byte-to-Audio (B2A), for multilingual speech recognition and synthesis. Prior work has predominantly used characters, sub-words or words as the unit of choice to model text. These units are difficult to scale to languages with large vocabularies, particularly in the case of multilingual processing. In this work, we model text via a sequence of Unicode bytes, specifically, the UTF-8 variable length byte sequence for each character. Bytes allow us to avoid large softmaxes in languages with large vocabularies, and share representations in …
How Important Is Tokenization in French Medical Masked Language Models?
LIA, Avignon University（阿维尼翁大学计算机科学与人工智能实验室）；Zenidoc（Zenidoc公司）；Nantes Université（南特大学）；École Centrale Nantes（南特中央理工学院）；CNRS（法国国家科学研究中心）；LS2N, UMR 6004（南特数字科学实验室）
How can we effectively develop speech technology for languages where no transcribed data is available? Many existing approaches use no annotated resources at all, yet it makes sense to leverage information from large annotated corpora in other languages, for example in the form of multilingual bottleneck features (BNFs) obtained from a supervised speech recognition system. In this work, we evaluate the benefits of BNFs for subword modeling (feature extraction) in six unseen languages on a word discrimination task. First we establish a strong unsupervised baseline by combining two existing meth…
