English to Russian Tech/Engineering IT (Information Technology)

Natural language processing – a dream that for a long time only in science fiction films became true. Interpreting human language and behavior is hard, drawing conclusions even harder. NLP, by now being developed for half a century, combines artificial intelligence, linguistics and statistics.

NLP automatically analyzes and channels spoken and textual information. This is the basis for automating processes like compliance control, elaborating new metadata in archived documents or offering new support services using intelligent chatbots (i.e. artificial conversational agents).

NLP is a strong technology, able to detect automatically relevant content and thus supports decision making processes. IT uses machine learning to solve tasks that so far required a human contribution (is a customer communication a regular statement or a complaint, extracting topics of texts, determining tonality of a text, categorizing news articles (e.g. politics/sports…)). So NLP is useful to understand the semantics of human communication in multiple channels (social media, blogs, print or CRM data. It even detects if a man or a woman speaks, how old the person probably is and if it is a discussion, a conversation or an argument.

New services can be built around these capabilities. Is it easy - just plug and play or does a company need help? There are some prerequisites to be met to take the full advantage of the possibilities offered.

First about the data: They have to be stored in an adequate repository (no SQL database in case it is massive, or traditional ECM/document databases). Then about the questions you have to ask yourself: Do I need to categorize information (e.g. type of news article), or to extract “bits and pieces” of information from the content (e.g. name of a person, location, product)? Do I have labeled data (e.g. examples of positive/negative emails from my customer) or is my data unlabeled?

These questions will determine the kind of algorithm that is relevant to apply. Such algorithms can include supervised methods (e.g. logistic regression or support vector machines) or non-supervised ones such as clustering techniques.

An expert in NLP and Machine Learning will be very helpful in selecting both the algorithm and the features it uses (“what to look at”): are words sufficient to represent a text or are some deeper linguistic analysis (part of speech tagging, syntactic parsing, …) needed to “let the data speak”. The progress in deep neural networks is proving extremely interesting and effective in unstructured data analysis and provides exciting perspectives for NLP as well.