Tech: Text Generation

Sophisticated Natural Language Generation: generate written text from data.

Everyday content production has a lot of use cases, where the underlying information is based on data: Product descriptions are written based on product fact sheets, business reports on some kind of BI data or from a spreadsheet, news are written from information like weather data.

Natural Language Generation turns this data into written content directly, without the need for a human writer.

how nlg works

Three Components to write a text via software

Three components are responsible for the text output:

  1. Your data is the basis for the content itself.
  2. An AI training contributes the information about the desired output, meaning of the data items, phrasing, styling, keywords, translation hints, etc.
  3. The NLG Core Software (our "text-engine") interprets that data together with the AI training and adds grammatical, semantic and lexical information to produce the final output: a human-sounding narrative and styled text. 

NLG Core Features

Any kind of text, any kind of style.

⇠ Text Quality: perfect human sounding text output, including html markup and dynamic keyword density

⇠ Public Availabili

ty: easy setup, no programming, data exchange and training via web app

⇠ Essential Writing: real-time content in 24 languages, millions of articles a day

The process of "data to text"

Step 1: Data Intake

Connect your system via API or use the webinterface

  • Data processing is totally automated via our REST API.
  • Your system delivers the data into the NLG Cloud, each dataset represents one text 
  • The data is being cleansed and analysed and made ready for text processing

Step 2: Editorial Configuration

Define, what is important to your reader, and how it should be written. Define document structure, wordings, etc.

A ruleset is the customization part, that connects data to editorial notions. This allows to write about any topic, in your desired wording. Basically your definition of the output.

  • detecting, what is worthy to write about for each data item
  • selection and prioritization of linguistic direction
  • translation of individual words (like brand or product names)
  • text structure, markup/styles, keywords

Step 3: Grammatical Rendering

Grammatical features of the NLG Core

The NLG Core interprets that ruleset and connects it to grammar, using all grammatical phenomenas of each language.

  • Flexion of nouns and verbs
  • Adding articles and determiners
  • Setting plurals/singulars
  • Sentence structure, synonyms
  • Text length, using keywords …

Step 4: Output Processing

Receive your content.

The text output is then processed to match the output parameters and sent back to you.

  • Rendering markup and HTML styles
  • sent via webhook for real time production
  • provided via API for analysis

Grammatical Features

Automated content needs to be highly dynamic to render your content based on the data: Since you cannot anticipate every single data item, key or value, a grammatical engine is needed to take care of the correct grammar

Some Examples for grammatical features in the NLG Cloud

Setting the Case

Example Input from data: “dog” 

english output

nominative: “The dog”
dative:  “I give the ball to the dog.”
accusative: “I see the dog.”
genitive:  “The dog’s toy”

german output

nominative: “Der Hund”
dative:  “Ich gebe dem Hund seinen Ball.”
accusative: “Ich sehe den Hund.”
genitive:  “Das Spielzeug des Hundes”

Setting the Article

Example Input from data: “smartphone” 

english output

definite: “the smartphone”
indefinite:  “a smartphone”
demonstrative: “this smartphone”
genitive:  “The dog’s toy”

german output

definite: “das Smartphone”
indefinite:  “ein Smartphone”
demonstrative: “dieses Smartphone”

Setting the Pronoun

Example Input from data: “smartphone” 

english output

personal: “it”
demonstrative:  “this smartphone”
demonstrative2: “which smartphone”
which:  “which”

german output
personal: “es”
demonstrative:  “dieses"
demonstrative2: “das”
which:  “welches”

Setting the Plural

Example Input from data: count of HDMI port

english output

singular: “one HDMI port”
plural:  “two HDMI ports”

german output
singular: “ein HDMI Anschluß”
plural:  “zwei HDMI Anschlüße”

Much more Grammar

Some more include:

  • Setting the Preposition
  • Converting Numerals
  • In-sentence references between parts of speech

  • Conjunctions: correctly combining words into lists
  • changing tense (present / past ; e.g. it is /it was)

Question about Pricing?

⇠ We have multiple plans available for you, based on the needs for different areas of text production.