Ciprian Chelba

Ciprian Chelba

Ciprian Chelba is a Research Scientist with Google. Previously he worked as a Researcher in the Speech Technology Group at Microsoft Research.

His research interests are in statistical modeling of natural language and speech. Recent projects include: Google Audio Indexing; indexing, ranking and snippeting of speech content; Language Modeling for Google Search by Voice, and Android IME predictive keyboard.

Authored Publications
Sort By
  • Title
  • Title, descending
  • Year
  • Year, descending
    Preview abstract The paper presents an approach to semantic grounding of language models (LMs) that conceptualizes the LM as a conditional model generating text given a desired semantic message. It embeds the LM in an auto-encoder by feeding its output to a semantic parser whose output is in the same representation domain as the input message. Compared to a baseline that generates text using greedy search, we demonstrate two techniques that improve the fluency and semantic accuracy of the generated text: The first technique samples multiple candidate text sequences from which the semantic parser chooses. The second trains the language model while keeping the semantic parser frozen to improve the semantic accuracy of the auto-encoder. We carry out experiments on the English WebNLG 3.0 data set, using BLEU to measure the fluency of generated text and standard parsing metrics to measure semantic accuracy. We show that our proposed approaches significantly improve on the greedy search baseline. Human evaluation corroborates the results of the automatic evaluation experiments. View details
    Preview abstract Motivated by the fact that most of the information relevant to the prediction of target tokens is drawn from the source sentence S=s1,...,sS, we propose truncating the target-side context used for incremental predictions by making a Markov (N-gram) assumption. Experiments on WMT EnDe and EnFr data sets show that the N-gram masked self-attention model loses very little in BLEU score for N values in the range 4,...,8, depending on the task. View details
    Preview abstract The paper investigates the feasibility of confidence estimation for neural machine translation models operating at the high end of the performance spectrum. As a side product of the data annotation process necessary for building such models we propose sentence level accuracy $SACC$ as a simple, self-explanatory evaluation metric for quality of translation. Experiments on two different annotator pools, one comprised of non-expert (crowd-sourced) and one of expert (professional) translators show that $SACC$ can vary greatly depending on the translation proficiency of the annotators, despite the fact that both pools are about equally reliable according to Krippendorff's alpha metric; the relatively low values of inter-annotator agreement confirm the expectation that sentence-level binary labeling $good$ / $needs\ work$ for translation out of context is very hard. For an English-Spanish translation model operating at $SACC = 0.89$ according to a non-expert annotator pool we can derive a confidence estimate that labels 0.5-0.6 of the $good$ translations in an ``in-domain" test set with 0.95 Precision. Switching to an expert annotator pool decreases $SACC$ dramatically: $0.61$ for English-Spanish, measured on the exact same data as above. This forces us to lower the CE model operating point to 0.9 Precision while labeling correctly about 0.20-0.25 of the $good$ translations in the data. We find surprising the extent to which CE depends on the level of proficiency of the annotator pool used for labeling the data. This leads to an important recommendation we wish to make when tackling CE modeling in practice: it is critical to match the end-user expectation for translation quality in the desired domain with the demands of annotators assigning binary quality labels to CE training data. View details
    Preview abstract Noise and domain are important aspects of data quality for neural machine translation. Existing research focus separately on domain-data selection, clean-data selection, or their static combination, leaving the dynamic interaction across them not explicitly examined. This paper introduces a ``co-curricular learning'' method to compose dynamic domain-data selection with dynamic clean-data selection, for transfer learning across both capabilities. We apply an EM-style optimization procedure to further refine the ``co-curriculum''. Experiment results and analysis with two domains demonstrate the viability of the method and the properties of data scheduled by the co-curriculum. View details
    Preview abstract Recent work in Neural Machine Translation (NMT) has shown significant quality gains from noised-beam decoding during back-translation, a method to generate synthetic parallel data. We show that the main role of such synthetic noise is not to diversify the source side, as previously suggested, but simply to indicate to the model that the given source is synthetic. We propose a simpler alternative to noising techniques, consisting of tagging back-translated source sentences with an extra token. Our results on WMT outperform noised back-translation in English-Romanian and match performance on English-German, re-defining state-of-the-art in the former. View details
    Denoising Neural Machine Translation Training with Trusted Data and Online Data Selection
    Wei Wang
    Taro Watanabe
    Macduff Hughes
    Tetsuji Nakagawa
    Third Conference on Machine Translation (WMT18) (2018)
    Preview abstract Measuring domain relevance of data and identifying or selecting well-fit domain data for machine translation (MT) is a well-studied topic, but denoising is not yet. Denoising is concerned with a different type of data quality and tries to reduce the negative impact of data noise on MT training, in particular, neural MT (NMT) training. This paper generalizes methods for measuring and selecting data for domain MT and applies them to denoising NMT training. The proposed approach uses trusted data and a denoising curriculum realized by online data selection. Intrinsic and extrinsic evaluations of the approach show its significant effectiveness for NMT to train on data with severe noise. View details
    Preview abstract Model compression is essential for serving large deep neural nets on devices with limited resources or applications that require real-time responses. For advanced NLP problems, a neural language model usually consists of recurrent layers (e.g., using LSTM cells), an embedding matrix for representing input tokens, and a softmax layer for generating output tokens. For problems with a very large vocabulary size, the embedding and the softmax matrices can account for more than half of the model size. For instance, the bigLSTM model achieves state-of-the-art performance on the One-Billion-Word (OBW) dataset with around 800k vocabulary, and its word embedding and softmax matrices use more than 6GBytes space, and are responsible for over 90\% of the model parameters. In this paper, we propose GroupReduce, a novel compression method for neural language models, based on vocabulary-partition (block) based low-rank matrix approximation and the inherent frequency distribution of tokens (the power-law distribution of words). We start by grouping words into c blocks based on their frequency, and then refine the clustering iteratively by constructing weighted low-rank approximation for each block, where the weights are based the frequencies of the words in the block. The experimental results show our method can significantly outperform traditional compression methods such as low-rank approximation and pruning. On the OBW dataset, our method achieved 6.6x compression rate for the embedding and softmax matrices, and when combined with quantization, our method can achieve 26x compression rate without losing prediction accuracy. View details
    Preview abstract We investigate the effective memory depth of RNN models by using them for $n$-gram language model (LM) smoothing. Experiments on a small corpus (UPenn Treebank, one million words of training data and 10k vocabulary) have found the LSTM cell with dropout to be the best model for encoding the $n$-gram state when compared with feed-forward and vanilla RNN models. When preserving the sentence independence assumption the LSTM $n$-gram matches the LSTM LM performance for $n=9$ and slightly outperforms it for $n=13$. When allowing dependencies across sentence boundaries, the LSTM $13$-gram almost matches the perplexity of the unlimited history LSTM LM. LSTM $n$-gram smoothing also has the desirable property of improving with increasing $n$-gram order, unlike the Katz or Kneser-Ney back-off estimators. Using multinomial distributions as targets in training instead of the usual one-hot target is only slightly beneficial for low $n$-gram orders. Experiments on the One Billion Words benchmark show that the results hold at larger scale. Building LSTM $n$-gram LMs may be appealing for some practical situations: the state in a $n$-gram LM can be succinctly represented with $(n-1)*4$ bytes storing the identity of the words in the context and batches of $n$-gram contexts can be processed in parallel. On the downside, the $n$-gram context encoding computed by the LSTM is discarded, making the model more expensive than a regular recurrent LSTM LM. View details
    Language Modeling in the Era of Abundant Data
    AI With the Best online conference. (2017)
    Preview abstract Overview of N-gram language modeling on large amounts of data, anchored in the reality of the speech recognition team at Google. View details
    Sparse Non-negative Matrix Language Modeling: Maximum Entropy Flexibility on the Cheap
    The 18th Annual Conference of the International Speech Communication Association, Stockholm, Sweden, pp. 2725-2729 (to appear)
    Preview abstract We present a new method for estimating the sparse non-negative model (SNM) by using a small amount of held-out data and the multinomial loss that is natural for language modeling; we validate it experimentally against the previous estimation method which uses leave-one-out on training data and a binary loss function and show that it performs equally well. Being able to train on held-out data is very important in practical situations where training data is mismatched from held-out/test data. We find that fairly small amounts of held-out data (on the order of 30-70 thousand words) are sufficient for training the adjustment model, which is the only model component estimated using gradient descent; the bulk of model parameters are relative frequencies counted on training data. A second contribution is a comparison between SNM and the related class of Maximum Entropy language models. While much cheaper computationally, we show that SNM achieves slightly better perplexity results for the same feature set and same speech recognition accuracy on voice search and short message dictation. View details