First Look: Yseop

I recently caught up with Yseop (“Easy-op”), artificial intelligence software designed to write the way a human would do helping turn analytics and other information into natural language. Yseop was founded in 2008 by a technologist and an angel investor but the product is based on over 20 years of research. Yseop is headquartered in France (Lyon) and has offices in Paris, London, Dallas and New York as well as partners in South America.

Yseop’s patented technology is not semantic software – it’s not about processing unstructured text, doing text analysis. Instead it is about writing. Yseop uses business rules to determine the reasons for things, generates dialogues for ongoing interaction to clarify what is meant and then writes non-repetitive text in multiple languages supported by notes and comments, explanations and recommendations.

Yseop is designed to deliver personalized expertise—knowledge and information that has been applied to a specific person’s context and which is, therefore, tailored to the specific context of the user. This contrasts with descriptions of timely but generic information (the kind of thing that comes back from a search engine) or static knowledge that is well organized but not specific or contextual (like books or reference material).

To this end Yseop has several elements:

• Yseop Rules, an inference engine – a business rules engine that does not require a decision design.
• Yseop Dialog, for generating dynamic questions to users to collect contextual information and diagnose their situation.
• Yseop Text, for writing structured and non-repetitive text.

The basic approach is that data is extracted from various data sources, reasoned against with rules and used to generate text. When this data is incomplete or no structured data source is available or it is insufficiently contextual, Yseop Dialog is used to complete it and drive a more complete set of outputs. Output can be, but are not limited to, PDFs, emails, web pages or even customer-ready PowerPoint presentations. Current customers use Yseop to generate executive summaries that take roll-ups of data and effectively summarize relevant trends or to generate detailed suggestions/scripts in the form of prep-to-meeting reports for account managers based on a wide range of potential account relationships and activity (Examples are available here:.

Users can manage business rules, dialog rules and text generation in either an Eclipse-based environment (Yseop Designer) or in a business-user friendly environment (Yseop Business Studio). Business rules are based on Domain Specific Languages or DSLs that give natural language names to the elements in the data source. Each rule is a simple If-Then construct and many rules can be managed in a folder structure. The DSLs can include functions to perform calculations such as the number of entries or a total for instance. Groups of rules can be activated at different points in a process and at execution time the inference engine will determine which rules should be fired based on the available data.

The Dialog rules have the same structure and also rely on DSLs. The actions in this case are the questions to be asked. These can be fixed text but can also be dynamic, using Yseop’s text generation facilities to customize the question presented to ask for data.

Central to the product are a set of built-ins focused on text generation that are configured using a wizard interface. For instance one generates a sentence from a list of values and allows things like bullet points or a sentence, all or just the top 3, initial and final text elements or transitions between elements and much more. Yseop also has sophisticated built-ins for handling things like a person’s name. Natural sounding text won’t always use someone’s full name but will sometimes use only their first name or a pronoun. Yseop allows you to use a single built in for the person and it will handle things like using a name versus he in a sentence – it will ensure that the full name is used initially, the first name each time a new paragraph starts and a suitable pronoun the rest of the time for example, while ensuring that the choices made are not ambiguous for the reader. These built-ins are critical to the product, making it easy to produce natural looking text. Yseop supports output in multiple languages (English, Spanish, French and German with plans to release Portuguese, Japanese and Chinese next year).

Yseop’s use of business rules to generate dynamic text and interactive dialogues is a well-established use case for rules-based decisioning technology, but its range of built-ins for writing allow it to generate much more natural text.

You can get more information on Yseop here and can try out some of their demos, like one writing your biography from your LinkedIn profile, here. Yseop is one of the vendors in our Decision Management Systems Platform Technology Report and will be included in an upcoming release.