scholarly journals Cross Entropy of Neural Language Models at Infinity—A New Bound of the Entropy Rate

Entropy ◽  
2018 ◽  
Vol 20 (11) ◽  
pp. 839 ◽  
Author(s):  
Shuntaro Takahashi ◽  
Kumiko Tanaka-Ishii
Keyword(s):  
Natural Language ◽  
State Of The Art ◽  
Training Data ◽  
Language Models ◽  
Cross Entropy ◽  
Entropy Rate ◽  
Natural Language Text ◽  
Power Law Decay ◽  
The Cross ◽  
Two Parameters

Neural language models have drawn a lot of attention for their strong ability to predict natural language text. In this paper, we estimate the entropy rate of natural language with state-of-the-art neural language models. To obtain the estimate, we consider the cross entropy, a measure of the prediction accuracy of neural language models, under the theoretically ideal conditions that they are trained with an infinitely large dataset and receive an infinitely long context for prediction. We empirically verify that the effects of the two parameters, the training data size and context length, on the cross entropy consistently obey a power-law decay with a positive constant for two different state-of-the-art neural language models with different language datasets. Based on the verification, we obtained 1.12 bits per character for English by extrapolating the two parameters to infinity. This result suggests that the upper bound of the entropy rate of natural language is potentially smaller than the previously reported values.

scholarly journals Attending to Entities for Better Text Understanding

2020 ◽  
Vol 34 (05) ◽  
pp. 7554-7561
Author(s):  
Pengxiang Cheng ◽  
Katrin Erk
Keyword(s):  
Large Scale ◽  
Human Performance ◽  
State Of The Art ◽  
Syntactic Structure ◽  
Semantic Knowledge ◽  
Training Data ◽  
Language Models ◽  
Long Distance ◽  
Future Directions ◽  
Text Understanding

Recent progress in NLP witnessed the development of large-scale pre-trained language models (GPT, BERT, XLNet, etc.) based on Transformer (Vaswani et al. 2017), and in a range of end tasks, such models have achieved state-of-the-art results, approaching human performance. This clearly demonstrates the power of the stacked self-attention architecture when paired with a sufficient number of layers and a large amount of pre-training data. However, on tasks that require complex and long-distance reasoning where surface-level cues are not enough, there is still a large gap between the pre-trained models and human performance. Strubell et al. (2018) recently showed that it is possible to inject knowledge of syntactic structure into a model through supervised self-attention. We conjecture that a similar injection of semantic knowledge, in particular, coreference information, into an existing model would improve performance on such complex problems. On the LAMBADA (Paperno et al. 2016) task, we show that a model trained from scratch with coreference as auxiliary supervision for self-attention outperforms the largest GPT-2 model, setting the new state-of-the-art, while only containing a tiny fraction of parameters compared to GPT-2. We also conduct a thorough analysis of different variants of model architectures and supervision configurations, suggesting future directions on applying similar techniques to other problems.

scholarly journals Pre-Trained Transformer-Based Language Models for Sundanese

2021 ◽  
Author(s):  
Wilson Wongso ◽  
Henry Lucky ◽  
Derwin Suhartono
Keyword(s):  
Natural Language ◽  
Text Classification ◽  
Training Data ◽  
Language Models ◽  
Classification Task ◽  
Language Understanding ◽  
Training Corpus ◽  
Low Resource ◽  
Corpus Size ◽  
Fine Tune

Abstract The Sundanese language has over 32 million speakers worldwide, but the language has reaped little to no benefits from the recent advances in natural language understanding. Like other low-resource languages, the only alternative is to fine-tune existing multilingual models. In this paper, we pre-trained three monolingual Transformer-based language models on Sundanese data. When evaluated on a downstream text classification task, we found that most of our monolingual models outperformed larger multilingual models despite the smaller overall pre-training data. In the subsequent analyses, our models benefited strongly from the Sundanese pre-training corpus size and do not exhibit socially biased behavior. We released our models for other researchers and practitioners to use.

scholarly journals Densely Supervised Hierarchical Policy-Value Network for Image Paragraph Generation

2019 ◽  
Author(s):  
Siying Wu ◽  
Zheng-Jun Zha ◽  
Zilei Wang ◽  
Houqiang Li ◽  
Feng Wu
Keyword(s):  
Natural Language ◽  
State Of The Art ◽  
Cross Entropy ◽  
Image Captioning ◽  
Value Network ◽  
Entropy Loss ◽  
Fine Grained ◽  
Performance Improvements ◽  
Single Sentence ◽  
Multiple State

Image paragraph generation aims to describe an image with a paragraph in natural language. Compared to image captioning with a single sentence, paragraph generation provides more expressive and fine-grained description for storytelling. Existing approaches mainly optimize paragraph generator towards minimizing word-wise cross entropy loss, which neglects linguistic hierarchy of paragraph and results in ``sparse" supervision for generator learning. In this paper, we propose a novel Densely Supervised Hierarchical Policy-Value (DHPV) network for effective paragraph generation. We design new hierarchical supervisions consisting of hierarchical rewards and values at both sentence and word levels. The joint exploration of hierarchical rewards and values provides dense supervision cues for learning effective paragraph generator. We propose a new hierarchical policy-value architecture which exploits compositionality at token-to-token and sentence-to-sentence levels simultaneously and can preserve the semantic and syntactic constituent integrity. Extensive experiments on the Stanford image-paragraph benchmark have demonstrated the effectiveness of the proposed DHPV approach with performance improvements over multiple state-of-the-art methods.

scholarly journals QASC: A Dataset for Question Answering via Sentence Composition

2020 ◽  
Vol 34 (05) ◽  
pp. 8082-8090
Author(s):  
Tushar Khot ◽  
Peter Clark ◽  
Michal Guerquin ◽  
Peter Jansen ◽  
Ashish Sabharwal
Keyword(s):  
Common Sense ◽  
Human Performance ◽  
Question Answering ◽  
State Of The Art ◽  
Multiple Choice ◽  
Training Data ◽  
Language Models ◽  
Current State ◽  
New Concepts ◽  
Large Corpus

Composing knowledge from multiple pieces of texts is a key challenge in multi-hop question answering. We present a multi-hop reasoning dataset, Question Answering via Sentence Composition (QASC), that requires retrieving facts from a large corpus and composing them to answer a multiple-choice question. QASC is the first dataset to offer two desirable properties: (a) the facts to be composed are annotated in a large corpus, and (b) the decomposition into these facts is not evident from the question itself. The latter makes retrieval challenging as the system must introduce new concepts or relations in order to discover potential decompositions. Further, the reasoning model must then learn to identify valid compositions of these retrieved facts using common-sense reasoning. To help address these challenges, we provide annotation for supporting facts as well as their composition. Guided by these annotations, we present a two-step approach to mitigate the retrieval challenges. We use other multiple-choice datasets as additional training data to strengthen the reasoning model. Our proposed approach improves over current state-of-the-art language models by 11% (absolute). The reasoning and retrieval problems, however, remain unsolved as this model still lags by 20% behind human performance.

Improving word prediction for augmentative communication by using idiolects and sociolects

2014 ◽  
Vol 3 (2) ◽  
pp. 137-154 ◽  
Author(s):  
Wessel Stoop ◽  
Antal van den Bosch
Keyword(s):  
Data Collection ◽  
State Of The Art ◽  
Training Data ◽  
Language Models ◽  
Prediction System ◽  
Word Prediction ◽  
The People ◽  
The Social ◽  
Statistical Language Models ◽  
Assistive Communication

Word prediction, or predictive editing, has a long history as a tool for augmentative and assistive communication. Improvements in the state-of-the-art can still be achieved, for instance by training personalized statistical language models. We developed the word prediction system Soothsayer. The main innovation of Soothsayer is that it not only uses idiolects, the language of one individual person, as training data, but also sociolects, the language of the social circle around that person. We use Twitter for data collection and experimentation. The idiolect models are based on individual Twitter feeds, the sociolect models are based on the tweets of a particular person and the tweets of the people he often communicates with. The sociolect approach achieved the best results. For a number of users, more than 50% of the keystrokes could have been saved if they had used Soothsayer.

scholarly journals Text: An R-package for Analyzing and Visualizing Human Language Using Natural Language Processing and Deep Learning

2021 ◽  
Author(s):  
Oscar Nils Erik Kjell ◽  
H. Andrew Schwartz ◽  
Salvatore Giorgi
Keyword(s):  
Deep Learning ◽  
Natural Language Processing ◽  
Natural Language ◽  
Language Processing ◽  
Rating Scale ◽  
State Of The Art ◽  
R Package ◽  
Language Models ◽  
Categorical Variables ◽  
Human Language

The language that individuals use for expressing themselves contains rich psychological information. Recent significant advances in Natural Language Processing (NLP) and Deep Learning (DL), namely transformers, have resulted in large performance gains in tasks related to understanding natural language such as machine translation. However, these state-of-the-art methods have not yet been made easily accessible for psychology researchers, nor designed to be optimal for human-level analyses. This tutorial introduces text (www.r-text.org), a new R-package for analyzing and visualizing human language using transformers, the latest techniques from NLP and DL. Text is both a modular solution for accessing state-of-the-art language models and an end-to-end solution catered for human-level analyses. Hence, text provides user-friendly functions tailored to test hypotheses in social sciences for both relatively small and large datasets. This tutorial describes useful methods for analyzing text, providing functions with reliable defaults that can be used off-the-shelf as well as providing a framework for the advanced users to build on for novel techniques and analysis pipelines. The reader learns about six methods: 1) textEmbed: to transform text to traditional or modern transformer-based word embeddings (i.e., numeric representations of words); 2) textTrain: to examine the relationships between text and numeric/categorical variables; 3) textSimilarity and 4) textSimilarityTest: to computing semantic similarity scores between texts and significance test the difference in meaning between two sets of texts; and 5) textProjection and 6) textProjectionPlot: to examine and visualize text within the embedding space according to latent or specified construct dimensions (e.g., low to high rating scale scores).

Database Tuning using Natural Language Processing

2021 ◽  
Vol 50 (3) ◽  
pp. 27-28
Author(s):  
Immanuel Trummer
Keyword(s):  
Natural Language Processing ◽  
Natural Language ◽  
Language Processing ◽  
Training Data ◽  
Language Models ◽  
Learning Approaches ◽  
Training Samples ◽  
Starting Point ◽  
Training Cost ◽  
Transformer Model

Introduction. We have seen significant advances in the state of the art in natural language processing (NLP) over the past few years [20]. These advances have been driven by new neural network architectures, in particular the Transformer model [19], as well as the successful application of transfer learning approaches to NLP [13]. Typically, training for specific NLP tasks starts from large language models that have been pre-trained on generic tasks (e.g., predicting obfuscated words in text [5]) for which large amounts of training data are available. Using such models as a starting point reduces task-specific training cost as well as the number of required training samples by orders of magnitude [7]. These advances motivate new use cases for NLP methods in the context of databases.

scholarly journals Concept Recognition as a Machine Translation Problem

2020 ◽  
Author(s):  
Mayla R Boguslav ◽  
Negacy D Hailu ◽  
Michael Bada ◽  
William A Baumgartner ◽  
Lawrence E Hunter
Keyword(s):  
Machine Learning ◽  
Machine Translation ◽  
Language Processing ◽  
State Of The Art ◽  
Training Data ◽  
Language Models ◽  
Alternative Methods ◽  
Automated Assignment ◽  
Concept Recognition ◽  
Alternative Approaches

AbstractBackgroundAutomated assignment of specific ontology concepts to mentions in text is a critical task in biomedical natural language processing, and the subject of many open shared tasks. Although the current state of the art involves the use of neural network language models as a post-processing step, the very large number of ontology classes to be recognized and the limited amount of gold-standard training data has impeded the creation of end-to-end systems based entirely on machine learning. Recently, Hailu et al. recast the concept recognition problem as a type of machine translation and demonstrated that sequence-to-sequence machine learning models had the potential to outperform multi-class classification approaches. Here we systematically characterize the factors that contribute to the accuracy and efficiency of several approaches to sequence-to-sequence machine learning.ResultsWe report on our extensive studies of alternative methods and hyperparameter selections. The results not only identify the best-performing systems and parameters across a wide variety of ontologies but also illuminate about the widely varying resource requirements and hyperparameter robustness of alternative approaches. Analysis of the strengths and weaknesses of such systems suggest promising avenues for future improvements as well as design choices that can increase computational efficiency with small costs in performance. Bidirectional Encoder Representations from Transformers for Biomedical Text Mining (BioBERT) for span detection (as previously found) along with the Open-source Toolkit for Neural Machine Translation (OpenNMT) for concept normalization achieve state-of-the-art performance for most ontologies in CRAFT Corpus. This approach uses substantially fewer computational resources, including hardware, memory, and time than several alternative approaches.ConclusionsMachine translation is a promising avenue for fully machine-learning-based concept recognition that achieves state-of-the-art results on the CRAFT Corpus, evaluated via a direct comparison to previous results from the 2019 CRAFT Shared Task. Experiments illuminating the reasons for the surprisingly good performance of sequence-to-sequence methods targeting ontology identifiers suggest that further progress may be possible by mapping to alternative target concept representations. All code and models can be found at: https://github.com/UCDenver-ccp/Concept-Recognition-as-Translation.

scholarly journals Complementary Auxiliary Classifiers for Label-Conditional Text Generation

2020 ◽  
Vol 34 (05) ◽  
pp. 8303-8310
Author(s):  
Yuan Li ◽  
Chunyuan Li ◽  
Yizhe Zhang ◽  
Xiujun Li ◽  
Guoqing Zheng ◽  
...  
Keyword(s):  
Natural Language ◽  
State Of The Art ◽  
Language Models ◽  
Future Research ◽  
Text Generation ◽  
Trade Off

Learning to generate text with a given label is a challenging task because natural language sentences are highly variable and ambiguous. It renders difficulties in trade-off between sentence quality and label fidelity. In this paper, we present CARA to alleviate the issue, where two auxiliary classifiers work simultaneously to ensure that (1) the encoder learns disentangled features and (2) the generator produces label-related sentences. Two practical techniques are further proposed to improve the performance, including annealing the learning signal from the auxiliary classifier, and enhancing the encoder with pre-trained language models. To establish a comprehensive benchmark fostering future research, we consider a suite of four datasets, and systematically reproduce three representative methods. CARA shows consistent improvement over the previous methods on the task of label-conditional text generation, and achieves state-of-the-art on the task of attribute transfer.

Sign in / Sign up

Export Citation Format

Share Document