In human-computer conversations, extracting entities such as names, street addresses and email addresses from speech is a challenging task. In this paper, we study the impact of fine-tuning pre-trained speech encoders on extracting spoken entities in human-readable form directly from speech without the need for text transcription. We illustrate that such a direct approach optimizes the encoder to transcribe only the entity relevant portions of speech ignoring the superfluous portions such as carrier phrases, or spell name entities. In the context of dialog from an enterprise virtual agent, we demonstrate that the 1-step approach outperforms the typical 2-step approach which first generates lexical transcriptions followed by text-based entity extraction for identifying spoken entities.

Spoken language understanding (SLU) extracts the intended mean- ing from a user utterance and is a critical component of conversational virtual agents. In enterprise virtual agents (EVAs), language understanding is substantially challenging. First, the users are infrequent callers who are unfamiliar with the expectations of a pre-designed conversation flow. Second, the users are paying customers of an enterprise who demand a reliable, consistent and efficient user experience when resolving their issues. In this work, we describe a general and robust framework for intent and entity extraction utilizing a hybrid of statistical and rule-based approaches. Our framework includes confidence modeling that incorporates information from all components in the SLU pipeline, a critical addition for EVAs to en- sure accuracy. Our focus is on creating accurate and scalable SLU that can be deployed rapidly for a large class of EVA applications with little need for human intervention.

Complex natural language understanding modules in dialog systems have a richer understanding of user utterances, and thus are critical in providing a better user experience. However, these models are often created from scratch, for specific clients and use cases and require the annotation of large datasets. This encourages the sharing of annotated data across multiple clients. To facilitate this we introduce the idea of intent features: domain and topic agnostic properties of intents that can be learnt from the syntactic cues only, and hence can be shared. We introduce a new neural network architecture, the Global-Local model, that shows significant improvement over strong baselines for identifying these features in a deployed, multi-intent natural language understanding module, and more generally in a classification setting where a part of an utterance has to be classified utilizing the whole context.

Designing robust conversation systems with great customer experience requires a team of design experts to think of all probable ways a customer can interact with the system and then author responses for each use case individually. The responses are authored from scratch for each new client and application even though similar responses have been created in the past. This happens largely because the responses are encoded using domain specific set of intents and entities. In this paper, we present preliminary work to define a dialog act schema to merge and map responses from different domains and applications using a consistent domain-independent representation. These representations are stored and maintained using an Elasticsearch system to facilitate generation of responses through a search and retrieval process. We experimented generating different surface realizations for a response given a desired information state of the dialog.

Current state-of-the-art models for named entity recognition (NER) are neural models with a conditional random field (CRF) as the final layer. Entities are represented as per-token labels with a special structure in order to decode them into spans. Current work eschews prior knowledge of how the span encoding scheme works and relies on the CRF learning which transitions are illegal and which are not to facilitate global coherence. We find that by constraining the output to suppress illegal transitions we can train a tagger with a cross-entropy loss twice as fast as a CRF with differences in F1 that are statistically insignificant, effectively eliminating the need for a CRF. We analyze the dynamics of tag co-occurrence to explain when these constraints are most effective and provide open source implementations of our tagger in both PyTorch and TensorFlow.

This paper describes our system (SLT-Interactions) for the CoNLL 2018 shared task: Multilingual Parsing from Raw Text to Universal Dependencies. Our system performs three main tasks: word segmentation (only for few treebanks), POS tagging and parsing. While segmentation is learned separately, we use neural stacking for joint learning of POS tagging and parsing tasks. For all the tasks, we employ simple neural network architectures that rely on long short-term memory (LSTM) networks for learning task-dependent features. At the basis of our parser, we use an arc-standard algorithm with Swap action for general non-projective parsing. Additionally, we use neural stacking as a knowledge transfer mechanism for cross-domain parsing of low resource domains. Our system shows substantial gains against the UDPipe baseline, with an average improvement of 4.18% in LAS across all languages. Overall, we are placed at the 12th position on the official test sets.

With the resurgence of chat-based dialog systems in consumer and enterprise applications, there has been much success in developing data-driven and rule-based natural language models to understand human intent. Since these models require large amounts of data and in-domain knowledge, expanding an equivalent service into new markets is disrupted by language barriers that inhibit dialog automation. This paper presents a user study to evaluate the utility of out-of-the-box machine translation technology to (1) rapidly bootstrap multilingual spoken dialog systems and (2) enable existing human analysts to understand foreign language utterances. We additionally evaluate the utility of machine translation in human assisted environments, where a portion of the traffic is processed by analysts. In English→Spanish experiments, we observe a high potential for dialog automation, as well as the potential for human analysts to process foreign language utterances with high accuracy.

With the increasing number of communication platforms that offer variety of ways of connecting two interlocutors, there is a resurgence of chat-based dialog systems. These systems, typically known as chatbots have been successfully applied in a range of consumer and enterprise applications. A key technology in such chat-bots is robust natural language understanding (NLU) which can significantly influence and impact the efficacy of the conversation and ultimately the user-experience. While NLU is far from perfect, this paper illustrates the role of underspecification and its impact on successful dialog completion.

The purpose of the current investigation is to predict post-editor profiles based on user behaviour and demographics using machine learning techniques to gain a better understanding of post-editor styles. Our study extracts process unit features from the CasMaCat LS14 database from the CRITT Translation Process Research Database (TPR-DB). The analysis has two main research goals: We create n-gram models based on user activity and part-of-speech sequences to automatically cluster post-editors, and we use discriminative classifier models to characterize post-editors based on a diverse range of translation process features. The classification and clustering of participants resulting from our study suggest this type of exploration could be used as a tool to develop new translation tool features or customization possibilities.

In simultaneous interpreting, human experts incrementally construct and extend partial hypotheses about the source speaker’s message, and start to verbalize a corresponding message in the target language, based on a partial translation – which may have to be corrected occasionally. They commence the target utterance in the hope that they will be able to finish understanding the source speaker’s message and determine its translation in time for the unfolding delivery. Of course, both incremental understanding and translation by humans can be garden-pathed, although experts are able to optimize their delivery so as to balance the goals of minimal latency, translation quality and high speech fluency with few corrections. We investigate the temporal properties of both translation input and output to evaluate the tradeoff between low latency and translation quality. In addition, we estimate the improvements that can be gained with a tempo-elastic speech synthesizer.

In previous work we introduced the idea of supertagging as a means of improving the efficiency of a lexicalized grammar parser. In this paper, we present supertagging in conjunction with a lightweight dependency analyzer as a robust and efficient partial parser. The present work is significant for two reasons. First, we have vastly improved our results; 92% accurate for supertag disambiguation using lexical information, larger training corpus and smoothing techniques. Second, we show how supertagging can be used for partial parsing and provide detailed evaluation results for detecting noun chunks, verb chunks, preposition phrase attachment and a variety of other linguistic constructions. Using supertag representation, we achieve a recall rate of 93.0% and a precision rate of 91.8% for noun chunking, improving on the best known result for noun chunking.

There are currently two philosophies for building grammars and parsers – Statistically induced grammars and Wide-coverage grammars. One way to combine the strengths of both approaches is to have a wide-coverage grammar with a heuristic component which is domain independent but whose contribution is tuned to particular domains. In this paper, we discuss a three-stage approach to disambiguation in the context of a lexicalized grammar, using a variety of domain independent heuristic techniques. We present a training algorithm which uses hand-bracketed treebank parses to set the weights of these heuristics. We compare the performance of our grammar against the performance of the IBM statistical grammar, using both untrained and trained weights for the heuristics.