Knowledge distillation is used, in generative language modeling, to train a smaller student model using the help of a larger teacher model, resulting in improved capabilities for the student model. In this paper, we formulate a more general framework for knowledge distillation where the student learns from the teacher during training, and also learns to ask for the teacher's help at test-time following rules specifying test-time restrictions. Towards this, we first formulate knowledge distillation as an entropy-regularized value optimization problem. Adopting Path Consistency Learning to solve this, leads to a new knowledge distillation algorithm using on-policy and off-policy demonstrations. We extend this using constrained reinforcement learning to a framework that incorporates the use of the teacher model as a test-time reference, within constraints. In this situation, akin to a human learner, the model needs to learn not only the learning material, but also the relative difficulty of different sections to prioritize for seeking teacher help. We examine the efficacy of our method through experiments in translation and summarization tasks, observing trends in accuracy and teacher use, noting that our approach unlocks operating points not available to the popular Speculative Decoding approach.

Fine-tuning large language models (LLMs) with a collection of large and diverse instructions has improved the model's generalization to different tasks, even for unseen tasks. However, most existing instruction datasets include only single instructions, and they struggle to follow complex instructions composed of multiple subtasks. In this work, we propose a novel concept of compositional instructions called chain-of-instructions (CoI), where the output of one instruction becomes an input for the next like a chain. Unlike the conventional practice of solving single instruction tasks, our proposed method encourages a model to solve each subtask step by step until the final answer is reached. CoI-tuning (i.e., fine-tuning with CoI instructions) improves the model's ability to handle instructions composed of multiple subtasks as well as unseen composite tasks such as multilingual summarization. Overall, our study find that simple CoI tuning of existing instruction data can provide consistent generalization to solve more complex, unseen, and longer chains of instructions.

Personalization, the ability to tailor a system to individual users, is an essential factor in user experience with natural language processing (NLP) systems. With the emergence of Large Language Models (LLMs), a key question is how to leverage these models to better personalize user experiences. To personalize a language model's output, a straightforward approach is to incorporate past user data into the language model prompt, but this approach can result in lengthy inputs exceeding limitations on input length and incurring latency and cost issues. Existing approaches tackle such challenges by selectively extracting relevant user data (i.e. selective retrieval) to construct a prompt for downstream tasks. However, retrieval-based methods are limited by potential information loss, lack of more profound user understanding, and cold-start challenges. To overcome these limitations, we propose a novel summary-augmented approach by extending retrieval-augmented personalization with task-aware user summaries generated by LLMs. The summaries can be generated and stored offline, enabling real-world systems with runtime constraints like voice assistants to leverage the power of LLMs. Experiments show our method with 75% less of retrieved user data is on-par or outperforms retrieval augmentation on most tasks in the LaMP personalization benchmark. We demonstrate that offline summarization via LLMs and runtime retrieval enables better performance for personalization on a range of tasks under practical constraints.

Open domain conversational agents can answer a broad range of targeted queries. However, the sequential nature of interaction with these systems makes knowledge exploration a lengthy task which burdens the user with asking a chain of well phrased questions. In this paper, we present a retrieval based system and associated dataset for predicting the next questions that the user might have. Such a system can proactively assist users in knowledge exploration leading to a more engaging dialog. The retrieval system is trained on a dataset which contains ~14K multi-turn information-seeking conversations with a valid follow-up question and a set of invalid candidates. The invalid candidates are generated to simulate various syntactic and semantic confounders such as paraphrases, partial entity match, irrelevant entity, and ASR errors. We use confounder specific techniques to simulate these negative examples on the OR-QuAC dataset and develop a dataset called the Follow-up Query Bank (FQ-Bank). Then, we train ranking models on FQ-Bank and present results comparing supervised and unsupervised approaches. The results suggest that we can retrieve the valid follow-ups by ranking them in higher positions compared to confounders, but further knowledge grounding can improve ranking performance.

Even though several methods have proposed 001 to defend textual neural network (NN) models 002 against black-box adversarial attacks, they of- 003 ten defend against a specific text perturbation 004 strategy and/or require re-training the models 005 from scratch. This leads to a lack of general- 006 ization in practice and redundant computation. 007 In particular, the state-of-the-art transformer 008 models (e.g., BERT, RoBERTa) require great 009 time and computation resources. By borrow- 010 ing an idea from software engineering , in or- 011 der to address these limitations, we propose 012 a novel algorithm, S HIELD , which modifies 013 and re-trains only the last layer of a textual 014 NN, and thus it “patches” and “transforms” 015 the NN into a stochastic weighted ensemble 016 of multi-expert prediction heads. Consider- 017 ing that most of current black-box attacks rely 018 on iterative search mechanisms to optimize 019 their adversarial perturbations, S HIELD con- 020 fuses the attackers by automatically utilizing 021 different weighted ensembles of predictors de- 022 pending on the input. In other words, S HIELD 023 breaks a fundamental assumption of the attack, 024 which is a victim NN model remains constant 025 during an attack. By conducting comprehen- 026 sive experiments, we demonstrate that all of 027 CNN, RNN, BERT, and RoBERTa-based tex- 028 tual NNs, once patched by S HIELD , exhibit 029 a relative enhancement of 15%–70% in accu- 030 racy on average against 14 different black-box 031 attacks, outperforming 6 defensive baselines 032 across 3 public datasets. All codes are to be 033 released. 034

Most recent neural semi-supervised learning algorithms rely on adding small perturbation to either the input vectors or their representations. These methods have been successful on computer vision tasks as the images form a continuous manifold, but are not appropriate for discrete input such as sentence. To adapt these methods to text input, we propose to decompose a neural network $M$ into two components $F$ and $U$ so that $M = U\circ F$. The layers in $F$ are then frozen and only the layers in $U$ will be updated during most time of the training. In this way, $F$ serves as a feature extractor that maps the input to high-level representation and adds systematical noise using dropout. We can then train $U$ using any state-of-the-art SSL algorithms such as $\Pi$-model, temporal ensembling, mean teacher, etc. Furthermore, this gradually unfreezing schedule also prevents a pretrained model from catastrophic forgetting. The experimental results demonstrate that our approach provides improvements when compared to state of the art methods especially on short texts.

Symbol sequences are estimated using a computer-implemented method including detecting one or more candidates of a target symbol sequence from a speech-to-text data, extracting a related portion of each candidate from the speech-to-text data, detecting repetition of at least a partial sequence of each candidate within the related portion of the corresponding candidate, labeling the detected repetition with a repetition indication, and estimating whether each candidate is the target symbol sequence, using the corresponding related portion including the repetition indication of each of the candidates.

User interactions with personal assistants like Alexa, Google Home and Siri are typically initiated by a wake term or wakeword. Several personal assistants feature "follow-up" modes that allow users to make additional interactions without the need of a wakeword. For the system to only respond when appropriate, and to ignore speech not intended for it, utterances must be classified as device-directed or non-device-directed. State-of-the-art systems have largely used acoustic features for this task, while others have used only lexical features or have added LM-based lexical features. We propose a directedness classifier that combines semantic lexical features with a lightweight acoustic feature and show it is effective in classifying directedness. The mixed-domain lexical and acoustic feature model is able to achieve 14% relative reduction of EER over a state-of-the-art acoustic-only baseline model. Finally, we successfully apply transfer learning and semi-supervised learning to the model to improve accuracy even further.

Neural network models have been very successful at achieving high accuracy on natural language inference (NLI) tasks. However, as demonstrated in recent literature, when tested on some simple adversarial examples, most of the models suffer a significant drop in performance. This raises the concern about the robustness of NLI models. In this paper, we propose to make NLI models robust by incorporating external knowledge to the attention mechanism using a simple transformation. We apply the new attention to two popular types of NLI models: one is Transformer encoder, and the other is a decomposable model, and show that our method can significantly improve their robustness. Moreover, when combined with BERT pretraining, our method achieves the human-level performance on the adversarial SNLI data set.

Complex deep learning models now achieve state of the art performance for many document retrieval tasks. The best models process the query or claim jointly with the document. However for fast scalable search it is desirable to have document embeddings which are independent of the claim. In this paper we show that knowledge distillation can be used to encourage a model that generates claim independent document encodings to mimic the behavior of a more complex model which generates claim dependent encodings. We explore this approach in document retrieval for a fact extraction and verification task. We show that by using the soft labels from a complex cross attention teacher model, the performance of claim independent student LSTM or CNN models is improved across all the ranking metrics. The student models we use are 12x faster in runtime and 20x smaller in number of parameters than the teacher

In this work we focus on confidence modeling for neural network based text classification and sequence to sequence models in the context of Natural Language Understanding (NLU) tasks. For most applications, the confidence of a neural network model in it's output is computed as a function of the posterior probability, determined via a softmax layer. In this work, we show that such scores can be poorly calibrated [1]. We propose new ensemble and gradient based features that predict model uncertainty and confidence. We evaluate the impact of these features through a gradient boosted decision tree (GBDT) framework to produce calibrated confidence scores. We demonstrate that the performance of our proposed approach surpasses the baseline across multiple tasks. Moreover, we show that this method produces confidence scores which are better suited for Out-Of-Distribution(OOD) classification when compared to the baseline.

Generating paraphrases that are lexically similar but semantically different is a challenging task. Paraphrases of this form can be used to augment data sets for various NLP tasks such as machine reading comprehension and question answering with non-trivial negative examples. In this article, we propose a deep variational model to generate paraphrases conditioned on a label that specifies whether the paraphrases are semantically related or not. We also present new training recipes and KL regularization techniques that improve the performance of variational paraphrasing models. Our proposed model demonstrates promising results in enhancing the generative power of the model by employing label-dependent generation on paraphrasing datasets.

Optimal selection of a subset of items from a given set is a hard problem that requires combinatorial optimization. In this paper, we propose a subset selection algorithm that is trainable with gradient-based methods yet achieves near-optimal performance via submodular optimization. We focus on the task of identifying a relevant set of sentences for claim verification in the context of the FEVER task. Conventional methods for this task look at sentences on their individual merit and thus do not optimize the informativeness of sentences as a set. We show that our proposed method which builds on the idea of unfolding a greedy algorithm into a computational graph allows both interpretability and gradient-based training. The proposed differentiable greedy network (DGN) outperforms discrete optimization algorithms as well as other baseline methods in terms of precision and recall.

Recurrent neural networks have become increasingly popular for the task of language modeling achieving impressive gains in state-of-the-art speech recognition and natural language processing (NLP) tasks. Recurrent models exploit word dependencies over a much longer context window (as retained by the history states) than what is feasible with n-gram language models. However the training criterion of choice for recurrent language models continues to be the local conditional likelihood of generating the current word given the (possibly long) word context, thus making local decisions at each word. This locally-conditional design fundamentally limits the ability of the model in exploiting whole sentence structures. In this paper, we present our initial results at whole sentence neural language models which assign a probability to the entire word sequence. We extend the previous work on whole sentence maximum entropy models to recurrent language models while using Noise Contrastive Estimation (NCE) for training, as these sentence models are fundamentally unnormalizable. We present results on a range of tasks: from sequence identification tasks such as, palindrome detection to large vocabulary automatic speech recognition (LVCSR) and demonstrate the modeling power of this approach.

Feed forward Neural Network Language Models (NNLM) have shown consistent gains over backoff word n-gram models in a variety of tasks. However, backoff n-gram models still remain dominant in applications with real time decoding requirements as word probabilities can be computed orders of magnitude faster than the NNLM. In this paper, we present a combination of techniques that allows us to speed up the probability computation from a neural net language model to make it comparable to the word n-gram model without any approximations. We present results on state of the art systems for Broadcast news transcription and conversational speech which demonstrate the speed improvements in real time factor and probability computation while retaining the WER gains from NNLM.

This paper investigates the effectiveness of knowledge distillation in the context of multilingual models. We show that with knowledge distillation, Long Short-Term Memory(LSTM) models can be used to train standard feed-forward Deep Neural Network (DNN) models for a variety of low-resource languages. We then examine how the agreement between the teacher's best labels and the original labels affects the student model's performance. Next, we show that knowledge distillation can be easily applied to semi-supervised learning to improve model performance. We also propose a promising data selection method to filter un-transcribed data. Then we focus on knowledge transfer among DNN models with multilingual features derived from CNN+DNN, LSTM, VGG, CTC and attention models. We show that a student model equipped with better input features not only learns better from the teacher's labels, but also outperforms the teacher. Further experiments suggest that by learning from each other, the original ensemble of various models is able to evolve into a new ensemble with even better combined performance.

End-to-end models (or sequence-to-sequence models) based on deep neural networks have recently become popular within the machine learning community. These techniques are also increasingly used in automatic speech recognition as an alternative to the state-of-the-art, hybrid HMM-DNN (hidden Markov model, deep neural network) system. The end-to-end systems contain a purely neural architecture that eliminates the need of any time alignment between the input acoustic feature vector sequence and output phone sequence. In this paper, we present progress within the IBM Watson Multimodal Group on end-to-end models for spoken language processing. We present our work on two types of end-to-end models applied to speech-to-text and keyword search tasks, namely, 1) recurrent neural networks (RNNs) based on connectionist temporal classification loss, and 2) attention-based encoder-decoder RNNs. We present results on several languages (such as Pashto, Mongolian, Javanese, Amharic, Guarani, Dholuo, Igbo, and Georgian) from the Intelligence Advanced Research Projects Activity funded Babel Program. We also present a detailed analysis of some salient characteristics of these models compared with the state-of-the-art HMM-DNN hybrid systems, and also discuss future challenges in using such models for spoken language processing.

End-to-end (E2E) systems have achieved competitive results compared to conventional hybrid hidden Markov model (HMM)-deep neural network based automatic speech recognition (ASR) systems. Such E2E systems are attractive due to the lack of dependence on alignments between input acoustic and output grapheme or HMM state sequence during training. This paper explores the design of an ASR-free end-to-end system for text query-based keyword search (KWS) from speech trained with minimal supervision. Our E2E KWS system consists of three sub-systems. The first sub-system is a recurrent neural network (RNN)-based acoustic auto-encoder trained to reconstruct the audio through a finite-dimensional representation. The second sub-system is a character-level RNN language model using embeddings learned from a convolutional neural network. Since the acoustic and text query embeddings occupy different representation spaces, they are input to a third feed-forward neural network that predicts whether the query occurs in the acoustic utterance or not. This E2E ASR-free KWS system performs respectably despite lacking a conventional ASR system and trains much faster.

While recurrent neural network language models based on Long Short Term Memory (LSTM) have shown good gains in many automatic speech recognition tasks, Convolutional Neural Network (CNN) language models are relatively new and have not been studied in-depth. In this paper we present an empirical comparison of LSTM and CNN language models on English broadcast news and various conversational telephone speech transcription tasks. We also present a new type of CNN language model that leverages dilated causal convolution to efficiently exploit long range history. We propose a novel criterion for training language models that combines word and class prediction in a multi-task learning framework. We apply this criterion to train word and character based LSTM language models and CNN language models and show that it improves performance. Our results also show that CNN and LSTM language models are complementary and can be combined to obtain further gains.