Make Conversational AI real – and ubiquitous
Our accessible technology is the fruit of extensive R&D into some of the most fundamental problems in Conversational AI. Our focus lies at the intersection of Artificial Intelligence, Natural Language Processing, and Linguistics, which is seeing the birth of a growing number of fields in everything NLU and Dialogue-related.
Technology
Deep Semantic NLU
Our industry-ready semantic representation is a new approach to the unpredictable nature of human language. It balances expressiveness and annotation cost and utilises structured state-of-the-art Machine Learning.
Dynamic Dialogue State Tracking
Remembering what is said in a conversation is at the core of any conversation. Our innovative take on solving memory and combining real world data, databases and NLP pushes the boundaries of what’s possible.
Dialogue Management (Symbolic AI)
One of the biggest determinants of real conversational abilities is interpretable and reasoned decision making. Our system-wide goal tracking and interpretable dialogue-capable policy present a leap forward.
Natural Language Generation
Receiving precise feedback or information in natural language is crucial for the next generation of interfaces. Our Natural Language Generation provides the best responses based on your structured data, databases and APIs.
Data Pipeline
Training data for Machine Learning models is key. Our proprietary data-pipeline handles everything end-to-end, so that you don’t have to. We generate, annotate, inspect and augment the data with our in-house technology and human-in-the-loop approaches.
Model Intelligence
Understanding the model is key to providing the best capabilities for users. Our unique analytics suite provides in-depth analysis of the models deployed in production and during development, leading to precise improvements and insights.
Research at Wluper
Our research philosophy is to question the ‘norm’ and other such assumptions until we identify every artificial limit. Now that we have identified the walls of the ‘box’, we can start thinking outside of it, surfing the vast ocean of ideas, diving in for extraordinary results.
Universal NLU
Why train a new NLU for every topic of conversation? We are in the process of developing the next generation of NLU that works universally on any topic without extensive re-training.
Imagine an off-the-shelf NLU which works perfectly for you and only requires a single configuration.
Non-linear Dialogue State Tracking
Why must human-machine dialogues be constrained to rigid, pre-defined paths? We are integrating recursive flows to add extended flexibility and robustness to your dialogue experience.
Imagine conversing freely, oh and jumping between topics in between, while a machine still understands and keeps up with you.
Reasoning
Why only focus on the power of neural systems and ignore the higher-level cognition offered by symbolic reasoning systems? We integrate DL models with first-order logic to empower virtual agents with neural-symbolic reasoning.
Imagine a system that just gets what you are saying or what you want and can still do the maths.
Research Papers
June 2023
How Are Idioms Processed Inside Transformer Language Models?
*SEM (ACL) 2023
Idioms such as “call it a day” and “piece of cake,” are prevalent in natural language. How do Transformer language models process idioms? This study examines this question by analysing three models - BERT, Multilingual BERT, and DistilBERT. We compare the embeddings of idiomatic and literal expressions across all layers
December 2022
Connecting the Semantic Dots:
Zero-shot Learning with Self-Aligning Autoencoders and a New Contrastive-Loss for Negative Sampling
IEEE ICMLA 2022
We introduce a novel zero-shot learning (ZSL)method, known as ‘self-alignment training’, and use it to train avanilla autoencoder which is then evaluated on four prominentZSL Tasks CUB, SUN, AWA1&2. Despite being a far simplermodel than the competition, our method achieved results on parwith SOTA.
April 2021
MATILDA - Multi-AnnoTator multi-language Interactive Light-weight Dialogue Annotator
EACL 2021
Dialogue Systems are becoming ubiquitous in various forms and shapes - virtual assistants (Siri, Alexa, etc.), chat-bots, customer sup-port, chit-chat systems just to name a few.The advances in language models and their publication have democratised advanced NLP.However, data remains a crucial bottleneck.Our contribution to this essential pillar
November 2020
Common Sense or World Knowledge? Investigating Adapter-Based Knowledge Injection into Pretrained Transformers
EMNLP 2020
Following the major success of neural language models (LMs) such as BERT or GPT-2 on a variety of language understanding tasks, recent work focused on injecting (structured) knowledge from external resources into these models
September 2019
LIDA: Lightweight Interactive Dialogue Annotator
EMNLP 2019 - Hong Kong, China
Dialogue systems have the potential to change how people interact with machines but are highly dependent on the quality of the data used to train them. It is therefore important to develop good dialogue annotation tools which can improve the speed and quality of dialogue data annotation.
October 2018
Evolutionary Data Measures: Understanding the Difficulty of Text Classification Tasks
CoNLL 2018 - Brussels, Belgium
Classification tasks are usually analysed and improved through new model architectures or hyperparameter optimisation but the underlying properties of datasets are discovered on an ad-hoc basis as errors occur. However, understanding the properties of the data is crucial in perfecting models.