NLP is currently dominated by general-purpose pretrained language models like RoBERTa, which achieve strong performance on NLU tasks through pretraining on billions of words. But what exact knowledge or skills do Transformer LMs learn from large-scale pretraining that they cannot learn from less data? We adopt four probing methods---classifier probing, information-theoretic probing, unsupervised relative acceptability judgment, and fine-tuning on NLU tasks---and draw learning curves that track the growth of these different measures of linguistic ability with respect to pretraining data volume using the MiniBERTas, a group of RoBERTa models pretrained on 1M, 10M, 100M and 1B words. We find that LMs require only about 10M or 100M words to learn representations that reliably encode most syntactic and semantic features we test. A much larger quantity of data is needed in order to acquire enough commonsense knowledge and other skills required to master typical downstream NLU tasks. The results suggest that, while the ability to encode linguistic features is almost certainly necessary for language understanding, it is likely that other forms of knowledge are the major drivers of recent improvements in language understanding among large pretrained models.
Curió-Edu 7B: Examining Data Selection Impacts in LLM Continued Pretraining
Institute of Computing (IC) University of Campinas (UNICAMP)（计算研究所（IC）坎皮纳斯大学（UNICAMP））；School of Electrical and Computer Engineering (FEEC) University of Campinas (UNICAMP)（电气与计算机工程学院（FEEC）坎皮纳斯大学（UNICAMP））
Allen Institute for Artificial Intelligence（艾伦人工智能研究所）；Harvard John A. Paulson School of Engineering and Applied Sciences（哈佛大学约翰·A·保尔森工程与应用科学学院）；MIT Computer Science and Artificial Intelligence Laboratory（麻省理工学院计算机科学与人工智能实验室）；University of Washington（华盛顿大学）
Contextual word representations derived from large-scale neural language models are successful across a diverse set of NLP tasks, suggesting that they encode useful and transferable features of language. To shed light on the linguistic knowledge they capture, we study the representations produced by several recent pretrained contextualizers (variants of ELMo, the OpenAI transformer language model, and BERT) with a suite of seventeen diverse probing tasks. We find that linear models trained on top of frozen contextual representations are competitive with state-of-the-art task-specific models in…
Chinese Tiny LLM: Pretraining a Chinese-Centric Large Language Model
Multimodal Art Projection Research Community（多模态艺术投影研究社区）；HKUST（香港科技大学）；Fudan University（复旦大学）；Peking University（北京大学）；Shanghai Jiaotong University（上海交通大学）；Nanjing University（南京大学）；University of Waterloo（滑铁卢大学）；Vector Institute（矢量研究所）
With the success of down streaming task using English pre-trained language model, the pre-trained Chinese language model is also necessary to get a better performance of Chinese NLP task. Unlike the English language, Chinese has its special characters such as glyph information. So in this article, we propose the Chinese pre-trained language model StyleBERT which incorporate the following embedding information to enhance the savvy of language model, such as word, pinyin, five stroke and chaizi. The experiments show that the model achieves well performances on a wide range of Chinese NLP tasks.
In this paper, we investigate the problem of training neural machine translation (NMT) systems with a dataset of more than 40 billion bilingual sentence pairs, which is larger than the largest dataset to date by orders of magnitude. Unprecedented challenges emerge in this situation compared to previous NMT work, including severe noise in the data and prohibitively long training time. We propose practical solutions to handle these issues and demonstrate that large-scale pretraining significantly improves NMT performance. We are able to push the BLEU score of WMT17 Chinese-English dataset to 32.…
The Fine Line: Navigating Large Language Model Pretraining with Down-streaming Capability Analysis
Peking University（北京大学）；Tianjin University（天津大学）；University of Melbourne（墨尔本大学）；Multimodal Art Projection Research Community（多模态艺术投影研究社区）；Fudan University（复旦大学）；University of Illinois at Urbana-Champaign（伊利诺伊大学厄巴纳-香槟分校）；Emory University（埃默里大学）；HKUST（香港科技大学）；University of Manchester（曼彻斯特大学）；University of Waterloo（滑铁卢大学）；Vector Institute（矢量研究所）
Pretraining deep language models has led to large performance gains in NLP. Despite this success, Schick and Schütze (2020) recently showed that these models struggle to understand rare words. For static word embeddings, this problem has been addressed by separately learning representations for rare words. In this work, we transfer this idea to pretrained language models: We introduce BERTRAM, a powerful architecture based on BERT that is capable of inferring high-quality embeddings for rare words that are suitable as input representations for deep language models. This is achieved by enabling…
Despite the development of pre-trained language models (PLMs) significantly raise the performances of various Chinese natural language processing (NLP) tasks, the vocabulary for these Chinese PLMs remain to be the one provided by Google Chinese Bert \cite{devlin2018bert}, which is based on Chinese characters. Second, the masked language model pre-training is based on a single vocabulary, which limits its downstream task performances. In this work, we first propose a novel method, \emph{seg\_tok}, to form the vocabulary of Chinese BERT, with the help of Chinese word segmentation (CWS) and subwo…
Clinical notes in electronic health records contain highly heterogeneous writing styles, including non-standard terminology or abbreviations. Using these notes in predictive modeling has traditionally required preprocessing (e.g. taking frequent terms or topic modeling) that removes much of the richness of the source data. We propose a pretrained hierarchical recurrent neural network model that parses minimally processed clinical notes in an intuitive fashion, and show that it improves performance for discharge diagnosis classification tasks on the Medical Information Mart for Intensive Care I…
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…
This technical report briefly describes our JDExplore d-team's submission Vega v1 on the General Language Understanding Evaluation (GLUE) leaderboard, where GLUE is a collection of nine natural language understanding tasks, including question answering, linguistic acceptability, sentiment analysis, text similarity, paraphrase detection, and natural language inference. [Method] We investigate several effective strategies and choose their best combination setting as the training recipes. As for model structure, we employ the vanilla Transformer with disentangled attention as the basic block enco…
We present an efficient method of pretraining large-scale autoencoding language models using training signals generated by an auxiliary model. Originated in ELECTRA, this training strategy has demonstrated sample-efficiency to pretrain models at the scale of hundreds of millions of parameters. In this work, we conduct a comprehensive empirical study, and propose a recipe, namely "Model generated dEnoising TRaining Objective" (METRO), which incorporates some of the best modeling techniques developed recently to speed up, stabilize, and enhance pretrained language models without compromising mod…
In-context learning is a surprising and important phenomenon that emerged when modern language models were scaled to billions of learned parameters. Without modifying a large language model's weights, it can be tuned to perform various downstream natural language tasks simply by including concatenated training examples of these tasks in its input. Though disruptive for many practical applications of large language models, this emergent learning paradigm is not well understood from a theoretical perspective. In this paper, we propose a first-of-its-kind PAC based framework for in-context learna…
Text Mining Unit Barcelona Supercomputing Center（文本挖掘组 巴塞罗那超级计算中心）
This work presents biomedical and clinical language models for Spanish by experimenting with different pretraining choices, such as masking at word and subword level, varying the vocabulary size and testing with domain data, looking for better language representations. Interestingly, in the absence of enough clinical data to train a model from scratch, we applied mixed-domain pretraining and cross-domain transfer approaches to generate a performant bio-clinical model suitable for real-world clinical data. We evaluated our models on Named Entity Recognition (NER) tasks for biomedical documents …
Singapore University of Technology and Design（新加坡科技设计大学）
Neural word segmentation research has benefited from large-scale raw texts by leveraging them for pretraining character and word embeddings. On the other hand, statistical segmentation research has exploited richer sources of external information, such as punctuation, automatic segmentation and POS. We investigate the effectiveness of a range of external training sources for neural word segmentation by building a modular segmentation model, pretraining the most important submodule using rich external sources. Results show that such pretraining significantly improves the model, leading to accur…
College of Information Engineering, Beijing Institute of Petrochemical Technology（北京石油化工学院信息工程学院）；School of computer science, Beijing University of Posts and Telecommunications（北京邮电大学计算机学院）
Though pre-trained language models such as Bert and XLNet, have rapidly advanced the state-of-the-art on many NLP tasks, they implicit semantics only relying on surface information between words in corpus. Intuitively, background knowledge influences the efficacy of understanding. Inspired by this common sense, we focus on improving model pretraining by leveraging explicit knowledge. Different from recent research that optimize pretraining model by knowledge masking strategies, we propose a simple but general method to combine explicit knowledge with pretraining. To be specific, we first match…
Raising Bars, Not Parameters: LilMoo Compact Language Model for Hindi
Bonn-Aachen International Center for Information Technology (b-it) / CAISA Lab（波恩-亚琛国际信息科技中心（b-it）/ CAISA实验室）；Lamarr Institute for Machine Learning and Artificial Intelligence（拉玛尔人工智能与机器学习研究所）；Bonn Sustainable AI Lab（波恩可持续AI实验室）；Helmholtz-Institut für Strahlen- und Kernphysik（海德堡辐射与核物理研究所）
The ever-growing volume of data of user-generated content on social media provides a nearly unlimited corpus of unlabeled data even in languages where resources are scarce. In this paper, we demonstrate that state-of-the-art results on two Thai social text categorization tasks can be realized by pretraining a language model on a large noisy Thai social media corpus of over 1.26 billion tokens and later fine-tuned on the downstream classification tasks. Due to the linguistically noisy and domain-specific nature of the content, our unique data preprocessing steps designed for Thai social media w…
Pretraining Data Detection for Large Language Models: A Divergence-based Calibration Method
CAS Key Lab of Network Data Science and Technology, ICT, CAS, Beijing, China（中国科学院信息科技研究所网络数据科学与技术重点实验室）；University of Chinese Academy of Sciences, Beijing, China（中国科学院大学北京校区）；Zhongguancun Laboratory, Beijing, China（中关村实验室）；University of Amsterdam, Amsterdam, The Netherlands（阿姆斯特丹大学）
Recently, Yuan et al. (2016) have shown the effectiveness of using Long Short-Term Memory (LSTM) for performing Word Sense Disambiguation (WSD). Their proposed technique outperformed the previous state-of-the-art with several benchmarks, but neither the training data nor the source code was released. This paper presents the results of a reproduction study of this technique using only openly available datasets (GigaWord, SemCore, OMSTI) and software (TensorFlow). From them, it emerged that state-of-the-art results can be obtained with much less data than hinted by Yuan et al. All code and train…
DatedGPT: Preventing Lookahead Bias in Large Language Models with Time-Aware Pretraining
Department of Finance, CUHK Business School, The Chinese University of Hong Kong（香港中文大学商学院金融系，香港中文大学）；Centre for Artificial Intelligence, University College London（伦敦大学学院人工智能中心）
University of Mannheim（曼海姆大学）；University of Cambridge（剑桥大学）
Unsupervised pretraining models have been shown to facilitate a wide range of downstream NLP applications. These models, however, retain some of the limitations of traditional static word embeddings. In particular, they encode only the distributional knowledge available in raw text corpora, incorporated through language modeling objectives. In this work, we complement such distributional knowledge with external lexical knowledge, that is, we integrate the discrete knowledge on word-level semantic similarity into pretraining. To this end, we generalize the standard BERT model to a multi-task le…
New York University（纽约大学）；Google AI Language（谷歌人工智能语言团队）；Harvard University（哈佛大学）；Johns Hopkins University（约翰斯·霍普金斯大学）；Brown University（布朗大学）
We introduce a set of nine challenge tasks that test for the understanding of function words. These tasks are created by structurally mutating sentences from existing datasets to target the comprehension of specific types of function words (e.g., prepositions, wh-words). Using these probing tasks, we explore the effects of various pretraining objectives for sentence encoders (e.g., language modeling, CCG supertagging and natural language inference (NLI)) on the learned representations. Our results show that pretraining on language modeling performs the best on average across our probing tasks,…
Central South University（中南大学）；Tencent AI Lab（腾讯人工智能实验室）
Chinese BERT models achieve remarkable progress in dealing with grammatical errors of word substitution. However, they fail to handle word insertion and deletion because BERT assumes the existence of a word at each position. To address this, we present a simple and effective Chinese pretrained model. The basic idea is to enable the model to determine whether a word exists at a particular position. We achieve this by introducing a special token \texttt{[null]}, the prediction of which stands for the non-existence of a word. In the training stage, we design pretraining tasks such that the model …
The standard BERT adopts subword-based tokenization, which may break a word into two or more wordpieces (e.g., converting "lossless" to "loss" and "less"). This will bring inconvenience in following situations: (1) what is the best way to obtain the contextual vector of a word that is divided into multiple wordpieces? (2) how to predict a word via cloze test without knowing the number of wordpieces in advance? In this work, we explore the possibility of developing BERT-style pretrained model over a vocabulary of words instead of wordpieces. We call such word-level BERT model as WordBERT. We tr…
CSIRO Data61（联邦科学与工业研究组织Data61）；University of Sydney（悉尼大学）；Digital Health CRC（数字健康合作研究中心）
Word vectors and Language Models (LMs) pretrained on a large amount of unlabelled data can dramatically improve various Natural Language Processing (NLP) tasks. However, the measure and impact of similarity between pretraining data and target task data are left to intuition. We propose three cost-effective measures to quantify different aspects of similarity between source pretraining and target task data. We demonstrate that these measures are good predictors of the usefulness of pretrained models for Named Entity Recognition (NER) over 30 data pairs. Results also suggest that pretrained LMs …
The lack of annotated data in many languages is a well-known challenge within the field of multilingual natural language processing (NLP). Therefore, many recent studies focus on zero-shot transfer learning and joint training across languages to overcome data scarcity for low-resource languages. In this work we (i) perform a comprehensive comparison of state-ofthe-art multilingual word and sentence encoders on the tasks of named entity recognition (NER) and part of speech (POS) tagging; and (ii) propose a new method for creating multilingual contextualized word embeddings, compare it to multip…
Google（谷歌）；University of Edinburgh（爱丁堡大学）；Cantab Research Ltd（Cantab研究有限公司）
We propose a new benchmark corpus to be used for measuring progress in statistical language modeling. With almost one billion words of training data, we hope this benchmark will be useful to quickly evaluate novel language modeling techniques, and to compare their contribution when combined with other advanced techniques. We show performance of several well-known types of language models, with the best results achieved with a recurrent neural network based language model. The baseline unpruned Kneser-Ney 5-gram model achieves perplexity 67.6; a combination of techniques leads to 35% reduction …
We present a new approach for pretraining a bi-directional transformer model that provides significant performance gains across a variety of language understanding problems. Our model solves a cloze-style word reconstruction task, where each word is ablated and must be predicted given the rest of the text. Experiments demonstrate large performance gains on GLUE and new state of the art results on NER as well as constituency parsing benchmarks, consistent with the concurrently introduced BERT model. We also present a detailed analysis of a number of factors that contribute to effective pretrain…
The One Billion Word Benchmark is a dataset derived from the WMT 2011 News Crawl, commonly used to measure language modeling ability in natural language processing. We train models solely on Common Crawl web scrapes partitioned by year, and demonstrate that they perform worse on this task over time due to distributional shift. Analysis of this corpus reveals that it contains several examples of harmful text, as well as outdated references to current events. We suggest that the temporal nature of news and its distribution shift over time makes it poorly suited for measuring language modeling ab…
This study is an attempt to build a contemporary linguistic corpus for Arabic language. The corpus produced, is a text corpus includes more than five million newspaper articles. It contains over a billion and a half words in total, out of which, there is about three million unique words. The data were collected from newspaper articles in ten major news sources from eight Arabic countries, over a period of fourteen years. The corpus was encoded with two types of encoding, namely: UTF-8, and Windows CP-1256. Also it was marked with two mark-up languages, namely: SGML, and XML.
Neural networks have excelled at many NLP tasks, but there remain open questions about the performance of pretrained distributed word representations and their interaction with weight initialization and other hyperparameters. We address these questions empirically using attention-based sequence-to-sequence models for natural language inference (NLI). Specifically, we compare three types of embeddings: random, pretrained (GloVe, word2vec), and retrofitted (pretrained plus WordNet information). We show that pretrained embeddings outperform both random and retrofitted ones in a large NLI corpus. …
