Data Creation & Data Analysis
Vernacular data creation for training AI engine or chatbot
Fidel works closely with clients to create vernacular data for training AI engine or chatbot. The data is 8KHz or 16KHz voice data and Fidel puts together teams to create the data. The data is then transcribed and annotated as per client rules or requirements and delivered after stringent QA. Fidel has also developed its own proprietary solution sets to expedite the delivery and strengthen the QA.
The languages handled by Fidel are Asian (Japanese, Chinese, Korean), ASEAN (Vietnamese, Singapore English, Malay) & Indian (Hindi, Tamil, Telugu, Gujarati, Bengali).
Fidel works with clients in the areas of sentiment analysis. Companies often want to finetune its operations by implementing client feedback across business process improvements. For this, client call center data is taken up for analysis, transcripted and then analyzed for sentiments so that clients understand the pain or frustration points of clients while using the website or a particular service. This analysis is used to further finetune operations and reduce the pain points.
Data Analytics and Reporting
Data is a mixture of English and Japanese or other foreign languages. Fidel will get it translated and create a set of multi-lingual reports over the same. In some cases, Fidel will use phonetic trans-literation to convert the data into local language and then display.
Data Conversion Adaptor
While translating ecommerce sites, a common problem is “How to translate the ecommerce product names” because the consumer will search in local language while the product names are in English in the database. In such cases, LinguaSol (group company of Fidel) has developed an adaptor or converter which sits between the search box and the backend software. This adaptor maintains a product dictionary which hosts a combination of phonetic transliterated words and complete translations. Using this approach, without changing the backend software or modifying the database, the problem is solved. So when consumer searches for say “jyagaimo” in the search box, the adaptor will take the same, check with its dictionary and then pick up the “potato” word and send it to the backend database. The results are then shown as it is.