Autoregressive language models are the currently dominant paradigm for text generation, however they have some fundamental limitations that cannot be remedied by scale—for example inherently sequential and unidirectional generation. While alternate classes of models have been explored, we have limited mathematical understanding of their fundamental power and limitations. In this paper we focus on Generative Masked Language Models (GMLMs), a non-autoregressive paradigm in which we train a model to fit conditional probabilities of the data distribution via masking, which are subsequently used as inputs to a Markov Chain to draw samples from the model. These models empirically strike a promising speed-quality trade-off as each step can be typically parallelized by decoding the entire sequence in parallel. We develop a mathematical framework for analyzing and improving such models which sheds light on questions of sample complexity and inference speed and quality. Empirically, we adapt the T5 model for iteratively-refined parallel decoding, achieving 2-3x speedup in machine translation with minimal sacrifice in quality compared with autoregressive models. We run careful ablation experiments to give recommendations on key design choices, and make fine-grained observations on the common error modes in connection with our theory. Our mathematical analyses and empirical observations characterize both potentials and limitations of this approach, and can be applied to future works on improving understanding and performance of GMLMs. We released codes for our experiments.

Autoregressive language models have achieved remarkable advancements, yet their potential is often limited by the slow inference speeds associated with sequential token generation. Blockwise parallel decoding (BPD) was proposed by Stern et al. [42] as a method to improve inference speed of language models by simultaneously predicting multiple future tokens, termed block drafts, which are subsequently verified by the autoregressive model. This paper advances the understanding and improvement of block drafts in two ways. First, we analyze token distributions generated across multiple prediction heads. Second, leveraging these insights, we propose algorithms to improve BPD inference speed by refining the block drafts using task-independent \ngram and neural language models as lightweight rescorers. Experiments demonstrate that by refining block drafts of open-sourced Vicuna and Medusa LLMs, the mean accepted token length are increased by 5-25% relative. This results in over a 3x speedup in wall clock time compared to standard autoregressive decoding in open-source 7B and 13B LLMs.

Document and discourse segmentation are two fundamental NLP tasks pertaining to breaking up text into constituents, which are commonly used to help downstream tasks such as information retrieval or text summarization.In this work, we propose three transformer-based architectures and provide comprehensive comparisons with previously proposed approaches on three standard datasets.We establish a new state-of-the-art, reducing in particular the error rates by a large margin in all cases.We further analyze model sizes and find that we can build models with many fewer parameters while keeping good performance, thus facilitating real-world applications.Early life and marriage: Franklin Delano Roosevelt was born on January 30, 1882, in the Hudson Valley town of Hyde Park, New York, to businessman James Roosevelt I and his second wife, Sara Ann Delano.(...) Aides began to refer to her at the time as "the president's girlfriend", and gossip linking the two romantically appeared in the newspapers.(...) Legacy: Roosevelt is widely considered to be one of the most important figures in the history of the United States, as well as one of the most influential figures of the 20th century.(...) Roosevelt has also appeared on several U.S. Postage stamps.

This paper poses a non-linear dynamical system on bipartite graphs and shows its stability under certain conditions. The dynamical system changes the weights on the nodes of the graph in each time step. The underlying weight transformation is non-linear, motivated by information gain in a document retrieval setting. Stability analysis of this problem is therefore more involved than that of PageRank-like algorithms. We show convergence using methods from Lyapunov theory and also provide some examples of how the algorithm performs when ranking keywords and sentences in a set of documents.

A novel distributed language model that has no constraints on the n-gram order and no practical constraints on vocabulary size is presented. This model is scalable and allows for an arbitrarily large corpus to be queried for statistical estimates. Our distributed model is capable of producing n-gram counts on demand. By using a novel heuristic estimate for the interpolation weights of a linearly interpolated model, it is possible to dynamically compute the language model probabilities. The distributed architecture follows the client-server paradigm and allows for each client to request an arbitrary weighted mixture of the corpus. This allows easy adaptation of the language model to particular test conditions. Experiments using the distributed LM for re-ranking N-best lists of a speech recognition system resulted in considerable improvements in word error rate (WER), while integration with a machine translation decoder resulted in significant improvements in translation quality as measured by the BLEU score.

In this paper, we argue that n-gram language models are not sufficient to address word reordering required for Machine Translation. We propose a new distortion model that can be used with existing phrase-based SMT decoders to address those n-gram language model limitations. We present empirical results in Arabic to English Machine Translation that show statistically significant improvements when our proposed model is used. We also propose a novel metric to measure word order similarity (or difference) between any pair of languages based on word alignments.

We approximate Arabic's rich morphology by a model that a word consists of a sequence of morphemes in the pattern prefix*-stem-suffix* (* denotes zero or more occurrences of a morpheme). Our method is seeded by a small manually segmented Arabic corpus and uses it to bootstrap an unsupervised algorithm to build the Arabic word segmenter from a large unsegmented Arabic corpus. The algorithm uses a trigram language model to determine the most probable morpheme sequence for a given input. The language model is initially estimated from a small manually segmented corpus of about 110,000 words. To improve the segmentation accuracy, we use an unsupervised algorithm for automatically acquiring new stems from a 155 million word unsegmented corpus, and re-estimate the model parameters with the expanded vocabulary and training corpus. The resulting Arabic word segmentation system achieves around 97% exact match accuracy on a test corpus containing 28,449 word tokens. We believe this is a state-of-the-art performance and the algorithm can be used for many highly inflected languages provided that one can create a small manually segmented corpus of the language of interest.

Human evaluations of machine translation are extensive but expensive. Human evaluations can take months to finish and involve human labor that can not be reused. We propose a method of automatic machine translation evaluation that is quick, inexpensive, and language-independent, that correlates highly with human evaluation, and that has little marginal cost per run. We present this method as an automated understudy to skilled human judges which substitutes for them when there is need for quick or frequent evaluations.

This paper describes results of an experiment with 9 different DARPA Communicator Systems who participated in the June 2000 data collection. All systems supported travel planning and utilized some form of mixed-initiative interaction. However they varied in several critical dimensions: (1) They targeted different back-end databases for travel information; (2) The used different modules for ASR,NLU,TTS and dialog management. We describe the experimental design, the approach to data collection, the metrics collected, and results comparing the systems. 1.

This paper presents a linear programming approach to discriminative training. We first define a measure of discrimination of an arbitrary conditional probability model on a set of labeled training data. We consider maximizing discrimination on a parametric family of exponential models that arises naturally in the maximum entropy framework. We show that this optimization problem is globally convex in R/sup n/, and is moreover piecewise linear on R/sup n/. We propose a solution that involves solving a series of linear programming problems. We provide a characterization of global optimizers. We compare this framework with those of minimum classification error and maximum entropy.

We describe our development work on a telephonebased conversational system in the domain of mutual fund transactions. This system uses several components including robust large vocabulary continuous speech recognition, natural language understanding, dialog management, and text-to-speech synthesis technologies.

In natural language, some sequences of words are very frequent.A classical language model, like n-gram, does not adequately take into account such sequences, because it underestimates their probabilities.A better approach consists in modeling word sequences as if they were individual dictionary elements.Sequences are considered as additional entries of the word lexicon, on which language models are computed.In this paper, we present two methods for automatically determining frequent phrases in unlabeled corpora of written sentences.These methods are based on information theoretic criteria which insure a high statistical consistency.Our models reach their local optimum since they minimize the perplexity.One procedure is based only on the n-gram language model to extract word sequences.The second one is based on a class n-gram model trained on 233 classes extracted from the eight grammatical classes of French.Experimental tests, in terms of perplexity and recognition rate, are carried out on a vocabulary of 20000 words and a corpus of 43 million words extracted from the "Le Monde" newspaper.Our models reduce perplexity by more than 20% compared with n-gram (nR3) and multigram models.In terms of recognition rate, our models outperform n-gram and multigram models.

We consider translating natural language sentences into a formal language using direct translation models built automatically from training data. Direct translation models have three components: an arbitrary prior conditional probability distribution, features that capture correlations between automatically determined key phrases or sets of words in both languages, and weights associated with these features. The features and the weights are selected using a training corpus of matched pairs of source and target language sentences to maximize the entropy or a new discrimination measure of the resulting conditional probability model. We report results in the air travel information system domain and compare the two methods of training.

We present a new approach to natural language understanding (NLU) based on the source-channel paradigm, and apply it to ARPA's Air Travel Information Service (ATIS) domain. The model uses techniques similar to those used by IBM in statistical machine translation. The parameters are trained using the exact match algorithm; a hierarchy of models is used to facilitate the bootstrapping of more complex models from simpler models.