Natural Tech
Natural Tech is competing in the ‘Analyse that’ battle of our Startups for News competition to be in with a chance to pitch at the GEN Summit 2018. Natural Tech is developing SentiLecto, a NLU (natural language understanding) engine that can understand a text as well as a native speaker. It lets the user discover actionable insights among large amounts of news, while also automatically generating content. If you would like to vote for Natural Tech, you can do so here.
What did you do prior to launching Natural Tech?
In the last few years, I have developed both my academic profile at Universidad de Buenos Aires (where I teach computational linguistics) and my professional career in the NLP industry in Argentina. Since 2010, I have worked for NLP-software companies, prototyping statistical feature selection and ML-based text classification algorithms, entity-based sentiment analysis, topic detection, full-parsing, question & answering systems, among other challenging projects.
What problem are you trying to address for newsrooms?
Media coverage of news produces huge amount of information. Those interested in media cannot read all that is being produced around breaking news or main topics on a day-to-day basis. Human processing prohibits dealing with a lot of news in a timely manner.
Our product, SentiLecto, is a natural language understanding engine applied to read news in Spanish and Portuguese (and soon in English) from all around the world. It will understand text as well as a native speaker, and it will enable the user to discover actionable insights among large amounts of news.
How are you attempting to solve the problems described above?
SentiLecto can deal with large amounts of news to help journalist get insights:
SentiLecto discovers fine-grained relationships between news covering the same issues and generates semantic graphs for related news (NewsGalaxy & NewsUniverse).
SentiLecto summarises news relying on fact-modelling (an interesting feature for further fact-checking).
By understanding a large amount of news in real-time and detecting new stories, SentiLecto enables alerts on breaking news: NewsFlow.
What sets you apart from your competitors? List three elements.
- SentiLecto can extract facts from complex linguistic models in Spanish and Portuguese. Our NLU solution yields a fine-grained representation of the texts. SentiLecto can identify linguistic features such as passive/active voice transformation, negation scope, anaphora resolution, co-reference chains, ellipsis, modality treatment, semantic features, time reasoning, and accurate verbal frames for all verbs.
- SentiLecto understands natural language just like a native speaker. It can recognise and classify named-entities (NERC) such as places, organisations and people. Also, SentiLecto can identify whether or not an utterance is a real fact (fact mining) through deep understanding of syntax and semantics.
- We offer linguistic models specifically for Spanish and Portuguese: English has traditionally been the target language for NLP frameworks. Companies that own decent NLP solutions for English thought they could just translate their lexical resources into other languages, but it doesn’t work like that. Spanish and Portuguese have more challenging grammar than English and thus specific grammar resources are needed.
What is your business model?
SentiLecto understands, rewrites, and merges news all over the world to generate quality posts in specific verticals. Fine-grained relationships between news covering the same facts and powerful linguistic transformations allow SentiLecto to generate quality posts that are more informative than the very sources they are based on. It can identify and merge missing facts and even come up with original stories from scratch.
SentiLecto is currently being used in this newsroom to automatically generate more than 200 quality posts (including images) from 150 feeds on a daily basis. Every post is enriched with new information, reports, and semantic graphs.
We have also an API-based and licensing business model for our NLU technology (see here).
What are your next steps?
We are planning to:
- Launch the English version of SentiLecto in 2018;
- Expand the team;
- Expand commercial operations in Brazil;
- Rely on human editing for auto-generated posts;
- Incorporate the company in US, Argentina and Brazil.
