The recent advancement of digital assistant technologies has opened new possibilities in the experiences they can provide. One of them is the ability to converse with a persona, e.g. celebrities, famous imaginary characters, etc. This experience requires that the replies are answered from the point of view of the persona, i.e. the first person. Since the facts about characters are typically found expressed in the third person, there is a need to rephrase them to the first person in order for the assistant not to break character and the experience to remain immersive. However, the automatic solution to such a problem is largely unexplored by the community. In this work, we present a new task for NLP: third person to first person rephrasing. We define the task and analyze its major challenges. We create and publish a novel dataset with 3493 human-annotated pairs of celebrity facts in the third person with their rephrased sentence in the first person. Moreover, we propose a transformer-based pipeline that correctly rephrases 92.0% of sentences compared to 77.0% rephrased by a rule-based baseline system.

Zhang et al. (2020) proposed to formulate few-shot intent classification as natural language inference (NLI) between query utterances and examples in the training set. The method is known as discriminative nearest neighbor classification or DNNC. Inspired by this work, we propose to simplify the NLI-style classification pipeline to be the entailment prediction on the utterance-semantic-label-pair (USLP). The semantic information in the labels can thus been infused into the classification process. Compared with DNNC, our proposed method is more efficient in both training and serving since it is based upon the entailment between query utterance and labels instead of all the training examples. The DNNC method requires more than one example per intent while the USLP approach does not have such constraint. In the 1-shot experiments on the CLINC150 (Larson et al., 2019) dataset, the USLP method outperforms traditional classification approach by >20 points (in-domain ac- curacy). We also find that longer and semantically meaningful labels tend to benefit model performance, however, the benefit shrinks as more training data is available.

We propose a personalized dialogue scenario generation system which transmits efficient and coherent information with a real-time extractive summarization method optimized by an Ising machine. The summarization problem is formulated as a quadratic unconstraint binary optimization (QUBO) problem, which extracts sentences that maximize the sum of the degree of user’s interest in the sentences of documents with the discourse structure of each document and the total utterance time as constraints. To evaluate the proposed method, we constructed a news article corpus with annotations of the discourse structure, users’ profiles, and interests in sentences and topics. The experimental results confirmed that a Digital Annealer, which is a simulated annealing-based Ising machine, can solve our QUBO model in a practical time without violating the constraints using this dataset.

The lack of labeled training data for new features is a common problem in rapidly changing real-world dialog systems. As a solution, we propose a multilingual paraphrase generation model that can be used to generate novel utterances for a target feature and target language. The generated utterances can be used to augment existing training data to improve intent classification and slot labeling models. We evaluate the quality of generated utterances using intrinsic evaluation metrics and by conducting downstream evaluation experiments with English as the source language and nine different target languages. Our method shows promise across languages, even in a zero-shot setting where no seed data is available.

In this paper, we explore how to use a small amount of new data to update a task-oriented semantic parsing model when the desired output for some examples has changed. When making updates in this way, one potential problem that arises is the presence of conflicting data, or out-of-date labels in the original training set. To evaluate the impact of this understudied problem, we propose an experimental setup for simulating changes to a neural semantic parser. We show that the presence of conflicting data greatly hinders learning of an update, then explore several methods to mitigate its effect. Our multi-task and data selection methods lead to large improvements in model accuracy compared to a naive data-mixing strategy, and our best method closes 86% of the accuracy gap between this baseline and an oracle upper bound.

Incremental intent classification requires the assignment of intent labels to partial utterances. However, partial utterances do not necessarily contain enough information to be mapped to the intent class of their complete utterance (correctly and with a certain degree of confidence). Using the final interpretation as the ground truth to measure a classifier’s accuracy during intent classification of partial utterances is thus problematic. We release inCLINC, a dataset of partial and full utterances with human annotations of plausible intent labels for different portions of each utterance, as an upper (human) baseline for incremental intent classification. We analyse the incremental annotations and propose entropy reduction as a measure of human annotators’ convergence on an interpretation (i.e. intent label). We argue that, when the annotators do not converge to one or a few possible interpretations and yet the classifier already identifies the final intent class early on, it is a sign of overfitting that can be ascribed to artefacts in the dataset.

Intent Classification (IC) and Slot Labeling (SL) models, which form the basis of dialogue systems, often encounter noisy data in real-word environments. In this work, we investigate how robust IC/SL models are to noisy data. We collect and publicly release a test-suite for seven common noise types found in production human-to-bot conversations (abbreviations, casing, misspellings, morphological variants, paraphrases, punctuation and synonyms). On this test-suite, we show that common noise types substantially degrade the IC accuracy and SL F1 performance of state-of-the-art BERT-based IC/SL models. By leveraging cross-noise robustness transfer, i.e. training on one noise type to improve robustness on another noise type, we design aggregate data-augmentation approaches that increase the model performance across all seven noise types by +10.8% for IC accuracy and +15 points for SL F1 on average. To the best of our knowledge, this is the first work to present a single IC/SL model that is robust to a wide range of noise phenomena.

In task-oriented dialogue systems, recent dialogue state tracking methods tend to perform one-pass generation of the dialogue state based on the previous dialogue state. The mistakes of these models made at the current turn are prone to be carried over to the next turn, causing error propagation. In this paper, we propose a novel Amendable Generation for Dialogue State Tracking (AG-DST), which contains a two-pass generation process: (1) generating a primitive dialogue state based on the dialogue of the current turn and the previous dialogue state, and (2) amending the primitive dialogue state from the first pass. With the additional amending generation pass, our model is tasked to learn more robust dialogue state tracking by amending the errors that still exist in the primitive dialogue state, which plays the role of reviser in the double-checking process and alleviates unnecessary error propagation. Experimental results show that AG-DST significantly outperforms previous works in two active DST datasets (MultiWOZ 2.2 and WOZ 2.0), achieving new state-of-the-art performances.

Improving user experience of a dialogue system often requires intensive developer effort to read conversation logs, run statistical analyses, and intuit the relative importance of system shortcomings. This paper presents a novel approach to automated analysis of conversation logs that learns the relationship between user-system interactions and overall dialogue quality. Unlike prior work on utterance-level quality prediction, our approach learns the impact of each interaction from the overall user rating without utterance-level annotation, allowing resultant model conclusions to be derived on the basis of empirical evidence and at low cost. Our model identifies interactions that have a strong correlation with the overall dialogue quality in a chatbot setting. Experiments show that the automated analysis from our model agrees with expert judgments, making this work the first to show that such weakly-supervised learning of utterance-level quality prediction is highly achievable.

Personalized dialogue systems are an essential step toward better human-machine interaction. Existing personalized dialogue agents rely on properly designed conversational datasets, which are mostly monolingual (e.g., English), which greatly limits the usage of conversational agents in other languages. In this paper, we propose a multi-lingual extension of Persona-Chat, namely XPersona. Our dataset includes persona conversations in six different languages other than English for evaluating multilingual personalized agents. We experiment with both multilingual and cross-lingual trained baselines and evaluate them against monolingual and translation-pipeline models using both automatic and human evaluation. Experimental results show that the multilingual trained models outperform the translation pipeline and that they are on par with the monolingual models, with the advantage of having a single model across multiple languages. On the other hand, the state-of-the-art cross-lingual trained models achieve inferior performance to the other models, showing that cross-lingual conversation modeling is a challenging task. We hope that our dataset and baselines will accelerate research in multilingual dialogue systems.

Intelligent personal assistants (IPAs) such as Amazon Alexa, Google Assistant and Apple Siri extend their built-in capabilities by supporting voice apps developed by third-party developers. Sometimes the smart assistant is not able to successfully respond to user voice commands (aka utterances). There are many reasons including automatic speech recognition (ASR) error, natural language understanding (NLU) error, routing utterances to an irrelevant voice app or simply that the user is asking for a capability that is not supported yet. The failure to handle a voice command leads to customer frustration. In this paper, we introduce a fallback skill recommendation system to suggest a voice app to a customer for an unhandled voice command. One of the prominent challenges of developing a skill recommender system for IPAs is partial observation. To solve the partial observation problem, we propose collaborative data relabeling (CDR) method. In addition, CDR also improves the diversity of the recommended skills. We evaluate the proposed method both offline and online. The offline evaluation results show that the proposed system outperforms the baselines. The online A/B testing results show significant gain of customer experience metrics.

User intent discovery is a key step in developing a Natural Language Understanding (NLU) module at the core of any modern Conversational AI system. Typically, human experts review a representative sample of user input data to discover new intents, which is subjective, costly, and error-prone. In this work, we aim to assist the NLU developers by presenting a novel method for discovering new intents at scale given a corpus of utterances. Our method utilizes supervised contrastive learning to leverage information from a domain-relevant, already labeled dataset and identifies new intents in the corpus at hand using unsupervised K-means clustering. Our method outperforms the state-of-the-art by a large margin up to 2% and 13% on two benchmark datasets, measured by clustering accuracy. Furthermore, we apply our method on a large dataset from the travel domain to demonstrate its effectiveness on a real-world use case.

Large-scale pretrained transformer models have demonstrated state-of-the-art (SOTA) performance in a variety of NLP tasks. Nowadays, numerous pretrained models are available in different model flavors and different languages, and can be easily adapted to one’s downstream task. However, only a limited number of models are available for dialogue tasks, and in particular, goal-oriented dialogue tasks. In addition, the available pretrained models are trained on general domain language, creating a mismatch between the pretraining language and the downstream domain launguage. In this contribution, we present CS-BERT, a BERT model pretrained on millions of dialogues in the customer service domain. We evaluate CS-BERT on several downstream customer service dialogue tasks, and demonstrate that our in-domain pretraining is advantageous compared to other pretrained models in both zero-shot experiments as well as in finetuning experiments, especially in a low-resource data setting.

Large-scale conversation models are turning to leveraging external knowledge to improve the factual accuracy in response generation. Considering the infeasibility to annotate the external knowledge for large-scale dialogue corpora, it is desirable to learn the knowledge selection and response generation in an unsupervised manner. In this paper, we propose PLATO-KAG (Knowledge-Augmented Generation), an unsupervised learning approach for end-to-end knowledge-grounded conversation modeling. For each dialogue context, the top-k relevant knowledge elements are selected and then employed in knowledge-grounded response generation. The two components of knowledge selection and response generation are optimized jointly and effectively under a balanced objective. Experimental results on two publicly available datasets validate the superiority of PLATO-KAG.

Dialogue state tracking models play an important role in a task-oriented dialogue system. However, most of them model the slot types conditionally independently given the input. We discover that it may cause the model to be confused by slot types that share the same data type. To mitigate this issue, we propose TripPy-MRF and TripPy-LSTM that models the slots jointly. Our results show that they are able to alleviate the confusion mentioned above, and they push the state-of-the-art on dataset MultiWoz 2.1 from 58.7 to 61.3.

For each goal-oriented dialog task of interest, large amounts of data need to be collected for end-to-end learning of a neural dialog system. Collecting that data is a costly and time-consuming process. Instead, we show that we can use only a small amount of data, supplemented with data from a related dialog task. Naively learning from related data fails to improve performance as the related data can be inconsistent with the target task. We describe a meta-learning based method that selectively learns from the related dialog task data. Our approach leads to significant accuracy improvements in an example dialog task.

Query rewrite (QR) is an emerging component in conversational AI systems, reducing user defect. User defect is caused by various reasons, such as errors in the spoken dialogue system, users’ slips of the tongue or their abridged language. Many of the user defects stem from personalized factors, such as user’s speech pattern, dialect, or preferences. In this work, we propose a personalized search-based QR framework, which focuses on automatic reduction of user defect. We build a personalized index for each user, which encompasses diverse affinity layers to reflect personal preferences for each user in the conversational AI. Our personalized QR system contains retrieval and ranking layers. Supported by user feedback based learning, training our models does not require hand-annotated data. Experiments on personalized test set showed that our personalized QR system is able to correct systematic and user errors by utilizing phonetic and semantic inputs.

Large-scale auto-regressive models have achieved great success in dialogue response generation, with the help of Transformer layers. However, these models do not learn a representative latent space of the sentence distribution, making it hard to control the generation. Recent works have tried on learning sentence representations using Transformer-based framework, but do not model the context-response relationship embedded in the dialogue datasets. In this work, we aim to construct a robust sentence representation learning model, that is specifically designed for dialogue response generation, with Transformer-based encoder-decoder structure. An utterance-level contrastive learning is proposed, encoding predictive information in each context representation for its corresponding response. Extensive experiments are conducted to verify the robustness of the proposed representation learning mechanism. By using both reference-based and reference-free evaluation metrics, we provide detailed analysis on the generated sentences, demonstrating the effectiveness of our proposed model.

Attention-based pre-trained language models such as GPT-2 brought considerable progress to end-to-end dialogue modelling. However, they also present considerable risks for task-oriented dialogue, such as lack of knowledge grounding or diversity. To address these issues, we introduce modified training objectives for language model finetuning, and we employ massive data augmentation via back-translation to increase the diversity of the training data. We further examine the possibilities of combining data from multiples sources to improve performance on the target dataset. We carefully evaluate our contributions with both human and automatic methods. Our model substantially outperforms the baseline on the MultiWOZ data and shows competitive performance with state of the art in both automatic and human evaluation.

Graph-to-text generation aims to generate fluent texts from graph-based data. In this paper, we investigate two recent pretrained language models (PLMs) and analyze the impact of different task-adaptive pretraining strategies for PLMs in graph-to-text generation. We present a study across three graph domains: meaning representations, Wikipedia knowledge graphs (KGs) and scientific KGs. We show that approaches based on PLMs BART and T5 achieve new state-of-the-art results and that task-adaptive pretraining strategies improve their performance even further. We report new state-of-the-art BLEU scores of 49.72 on AMR-LDC2017T10, 59.70 on WebNLG, and 25.66 on AGENDA datasets - a relative improvement of 31.8%, 4.5%, and 42.4%, respectively, with our models generating significantly more fluent texts than human references. In an extensive analysis, we identify possible reasons for the PLMs’ success on graph-to-text tasks. Our findings suggest that the PLMs benefit from similar facts seen during pretraining or fine-tuning, such that they perform well even when the input graph is reduced to a simple bag of node and edge labels.

Natural Language Generation (NLG) for task-oriented dialogue systems focuses on communicating specific content accurately, fluently, and coherently. While these attributes are crucial for a successful dialogue, it is also desirable to simultaneously accomplish specific stylistic goals, such as response length, point-of-view, descriptiveness, sentiment, formality, and empathy. In this work, we focus on stylistic control and evaluation for schema-guided NLG, with joint goals of achieving both semantic and stylistic control. We experiment in detail with various controlled generation methods for large pretrained language models: specifically, conditional training, guided fine-tuning, and guided decoding. We discuss their advantages and limitations, and evaluate them with a broad range of automatic and human evaluation metrics. Our results show that while high style accuracy and semantic correctness are easier to achieve for more lexically-defined styles with conditional training, stylistic control is also achievable for more semantically complex styles using discriminator-based guided decoding methods. The results also suggest that methods that are more scalable (with less hyper-parameters tuning) and that disentangle context generation and stylistic variations are more effective at achieving semantic correctness and style accuracy.

Entity tags in human-machine dialog are integral to natural language understanding (NLU) tasks in conversational assistants. However, current systems struggle to accurately parse spoken queries with the typical use of text input alone, and often fail to understand the user intent. Previous work in linguistics has identified a cross-language tendency for longer speech pauses surrounding nouns as compared to verbs. We demonstrate that the linguistic observation on pauses can be used to improve accuracy in machine-learnt language understanding tasks. Analysis of pauses in French and English utterances from a commercial voice assistant shows the statistically significant difference in pause duration around multi-token entity span boundaries compared to within entity spans. Additionally, in contrast to text-based NLU, we apply pause duration to enrich contextual embeddings to improve shallow parsing of entities. Results show that our proposed novel embeddings improve the relative error rate by up to 8% consistently across three domains for French, without any added annotation or alignment costs to the parser.

Humans make appropriate responses not only based on previous dialogue utterances but also on implicit background knowledge such as common sense. Although neural response generation models seem to produce human-like responses, they are mostly end-to-end and not generating intermediate grounds between a dialogue history and responses. This work aims to study if and how we can train an RG model that talks with itself to generate implicit knowledge before making responses. We further investigate can such models identify when to generate implicit background knowledge and when it is not necessary. Experimental results show that compared with models that directly generate responses given a dialogue history, self-talk models produce better-quality responses according to human evaluation on grammaticality, coherence, and engagingness. And models that are trained to identify when to self-talk further improves the response quality. Analysis on generated implicit knowledge shows that models mostly use the knowledge appropriately in the responses.

Knowledge Grounded Conversation Models are usually based on a selection/retrieval module and a generation module, trained separately or simultaneously, with or without having access to a ‘gold’ knowledge option. With the introduction of large pre-trained generative models, the selection and generation part have become more and more entangled, shifting the focus towards enhancing knowledge incorporation (from multiple sources) instead of trying to pick the best knowledge option. These approaches however depend on knowledge labels and/or a separate dense retriever for their best performance. In this work we study the unsupervised selection abilities of pre-trained generative models (e.g. BART) and show that by adding a score-and-aggregate module between encoder and decoder, they are capable of learning to pick the proper knowledge through minimising the language modelling loss (i.e. without having access to knowledge labels). Trained as such, our model - K-Mine - shows competitive selection and generation performance against models that benefit from knowledge labels and/or separate dense retriever.

Endowing a task-oriented dialogue system with adaptiveness to user personality can greatly help improve the performance of a dialogue task. However, such a dialogue system can be practically challenging to implement, because it is unclear how user personality influences dialogue task performance. To explore the relationship between user personality and dialogue task performance, we enrolled participants via crowdsourcing to first answer specified personality questionnaires and then chat with a dialogue system to accomplish assigned tasks. A rule-based dialogue system on the prevalent Multi-Domain Wizard-of-Oz (MultiWOZ) task was used. A total of 211 participants’ personalities and their 633 dialogues were collected and analyzed. The results revealed that sociable and extroverted people tended to fail the task, whereas neurotic people were more likely to succeed. We extracted features related to user dialogue behaviors and performed further analysis to determine which kind of behavior influences task performance. As a result, we identified that average utterance length and slots per utterance are the key features of dialogue behavior that are highly correlated with both task performance and user personality.

Code-mixed language plays a crucial role in communication in multilingual societies. Though the recent growth of web users has greatly boosted the use of such mixed languages, the current generation of dialog systems is primarily monolingual. This increase in usage of code-mixed language has prompted dialog systems in a similar language. We present our work in Code-Mixed Dialog Generation, an unexplored task in code-mixed languages, generating utterances in code-mixed language rather than a single language that is more often just English. We present a new synthetic corpus in code-mix for dialogs, CM-DailyDialog, by converting an existing English-only dialog corpus to a mixed Hindi-English corpus. We then propose a baseline approach where we show the effectiveness of using mBART like multilingual sequence-to-sequence transformers for code-mixed dialog generation. Our best performing dialog models can conduct coherent conversations in Hindi-English mixed language as evaluated by human and automatic metrics setting new benchmarks for the Code-Mixed Dialog Generation task.

Most prior work on task-oriented dialogue systems is restricted to supporting domain APIs. However, users may have requests that are out of the scope of these APIs. This work focuses on identifying such user requests. Existing methods for this task mainly rely on fine-tuning pre-trained models on large annotated data. We propose a novel method, REDE, based on adaptive representation learning and density estimation. REDE can be applied to zero-shot cases, and quickly learns a high-performing detector with only a few shots by updating less than 3K parameters. We demonstrate REDE’s competitive performance on DSTC9 data and our newly collected test set.