Keeping the dialogue state in dialogue systems is a notoriously difficult task. We introduce an ontology-based dialogue manage(OntoDM), a dialogue manager that keeps the state of the conversation, provides a basis for anaphora resolution and drives the conversation via domain ontologies. The banking and finance area promises great potential for disambiguating the context via a rich set of products and specificity of proper nouns, named entities and verbs. We used ontologies both as a knowledge base and a basis for the dialogue manager; the knowledge base component and dialogue manager components coalesce in a sense. Domain knowledge is used to track Entities of Interest, i.e. nodes (classes) of the ontology which happen to be products and services. In this way we also introduced conversation memory and attention in a sense. We finely blended linguistic methods, domain-driven keyword ranking and domain ontologies to create ways of domain-driven conversation. Proposed framework is used in our in-house German language banking and finance chatbots. General challenges of German language processing and finance-banking domain chatbot language models and lexicons are also introduced. This work is still in progress, hence no success metrics have been introduced yet.
German language processing is inherently challenging in general, independent of what the specific NLP task is. The main challenge is high variability in word forms due to inflections and compound words. Nouns, adjectives and verbs can be inflected according to gender, number and person. Rich word forms can pose a challenge language understanding components. In this paper, we focus on dialogue management. However, one should keep in mind that input to dialogue management components are provided by natural language understanding components. Another practical issue in everyday written language is the umlaut (mutated vowels). Everyday informal written text includes umlauts replaced by their plain counterparts i.e. "Madchen, uber, schon" rather than "Mädchen, über, schön" etc. Especially in conversational interfaces, usage of umlauts reduce significantly due to English layout keyboards or just being lax about punctuation while typing quickly on a smartphone. In our opinion, umlaut-to-plain vowel replaced words are also a part of chatbot language models. Morphologically rich languages have received considerable attention from many researchers. Many technical papers have been published to highlight the inherent technical difficulties in statistical methods e.g. MT, ASR-TTS, language models, text classification; practical solutions are offered in (Mikolov et al., 2016). We overcome the challenges of rich German morphology using DEMorphy, an open source German morphological analyzer and recognizer . Throughout our work, all
Keeping dialogue state in conversational interfaces is a notoriously difficult task. Dialogue systems, also known as chatbots, virtual assistants and conversational interfaces are already used in a broad set of applications, from psychological support to HR, customer care and 2 entertainment. Dialogue systems can be classified into goal-driven systems (e.g. flight booking, restaurant reservation) vs open-domain systems (e.g. psychological support, language learning and medical aid). As dialogue systems has gained attention, research interest in training natural conversation systems from large volumes of user data has grown. Goal-driven systems admit slot-filling and hand crafted rules, which is reliable but restrictive in the conversation (basically the user has to choose one of available options). Open domain conversational systems, based on generative probabilistic models attracted attention from many researchers, due these limits for goal-oriented systems (Serban et al., 2015;Li et al., 2017).
One problem with conversation is maintaining the dialogue state. This comprises of what the user said and how the chatbot answered, what we’re talking about and which pieces of information are relevant to generating the current answer. Kumar et al. (2017) introduced neural networks with memory and attention (DMN). Done up to here DMN includes episodic memory and an attention module plus to a recurrent encoder decoder. DMN first computes question representation. Then the question representation triggers the attention process. Finally the memory module can reason the answer from all relevant information. However, purely statistical approach has some drawbacks:
• statistical frameworks need huge training sets.
Especially frameworks with many statistical components such as DMN, have a great number of parameters and are vulnerable to sparseness problems. • Anaphora resolution is implicit. The anaphora resolutions go into neural network as implicit parameters, there’s no direct easy way to see how the resolutions worked. Answers come through at least two distinct statistical layers, one encoder and one decoder at least. Thus there is no easy way to understand why a specific answer is generated and how the anaphora resolution contributed to the generation. This paper addresses the dialog management. We will describe domain-driven ways to
• keep the conversation memory, both the user and the bot side • make the anaphora resolution • generate knowledge-based answers • possibly contribute to what to say next • integrate linguistic features into the context NLU and answer generation modules will not be considered in detail in this paper. The focus is on how ontologies can be used to generate natural conversations. However we will present the outputs and presentations for clarity. The goal is here to improve quality of conversations via domain knowledge. This work is still in progress, hence we were not able to include performance metrics yet, given the difficult nature of evaluation of dialogue systems in general.
This paper describes the methodology that is used in our in-house banking and finance chatbots. We chose the banking and finance domain due to rich set of products and specificity of proper nouns, named entities and verbs; high potential for disambiguate the context and drive the conversation. Though development was made on banking and finance domain, framework is applicable to other highly specific
This content is AI-processed based on open access ArXiv data.
