If you have already adopted AI in your small or mid-size organization, congratulations. If not, there is no better time than now. The urgency of adopting and immediately implementing some form of AI should be a top priority. If not, there are inherent risks. One, you will become laggard and most likely obsolete, given the super-cycle of innovation we are currently experiencing—the costs associated with “not” implementing AI, including going out of business.

Implementing AI is quite different from other organization-wide strategies because it involves highly specific characteristics and expert resource pools that SMBs might not have access to.

As one of the current five subcategories of AI, Natural Language Processing or NLP has been around for quite some time. In fact, AI has been around since the 1940s. It has attempted to gain traction (start and restart for decades) and finally broke through in 2006 after a cold winter in the mid-1980s. With the explosion of data in the past two decades and hardware computing power, companies have figured out how to finally leverage AI. Companies that don’t begin to implement AI, in its simplest form, truly become obsolete, and for stakeholders, it’s non-negotiable. The challenge is curating a vision of what AI will accomplish, the overall cost of implementation, the risks associated, which include horrible user-experiences, and getting 100% buy-in company-wide. I

In this post, I will attempt to discuss implementation techniques in the purest form of AI, a subfield called Natural Language Processing or (NLP), which uses computers to process human language. It’s ubiquitous now, but implementation has some risks. This process enables computer systems to understand, interpret, translate, and generate human-like language in both spoken and written forms. Intuitive understanding of human language at a general level is still beyond the capabilities of computers, but NLP is advancing rapidly as it learns from the explosion of data clusters that identify patterns used by humans in language.

Given that we all have a creative side, check out these two sites that are using NLP in an advanced form: authrors.ai and primer.ai.

NLP techniques include but are not limited to, counting words quickly, counting word occurrences, for example in the narrative reporting of quarterly results from a financial institution, (e.g., today’s customer feedback), while these short bursts of words are tricky to analyze, I think the results are spectacular if implemented correctly.

Another technique is the Hidden Markov Model or (HMM). This model uses sequential data that can predict words to complete a sentence for that matter or an algorithm that recognizes characters on a street sign. One example of this model is used for autonomous driving where the self-driving car recognizes the letters on the street sign, or variation thereof. The time will come when only driverless vehicles are allowed in city centers within a 1-mile radius, and if a human is found driving in that one-mile radius, he/she will be reprimanded… save for another post.

Another technique that has dominated NLP since 2015 is neural networks. It is rapidly expanding, including sentiment analysis, that you can find pattern recognition of words and phrases, machine translation, text generation, and text classification. Neural networks consist of hidden layers used to weigh and process information by performing calculations to make sense of the data. These hidden layers come in various forms.

These are the most common techniques, but we will surely find many other uses cases for processing NLP, which have already been discovered, such as identifying major turning points in a novel. Is this book a page-turner, for example. Or you can discover instant character analysis and so forth. Another example will be script writing. Producers are always concerned with budget, so an NLP powered algorithm will have the ability to understand the costs of a scene, and make recommendations for a more cost-effective budget.

I mentioned earlier that enterprises already have great success with implementing all five stages of AI, including robotics in health and computer-vision in film. But the implementation for smaller to mid-size companies becomes a far greater challenge. Resources and talent are two common constraints, besides budget.

Here are factors you might consider when attempting to implement NLP into your SMB organization:

Variation in scope and complexity: NLP projects vary enormously in scope and complexity, from a few hours of solitary work to install a chatbot, to highly complex efforts by a qualified team to analyze and synthesize clusters of words with unlimited computing resources for both structured and unstructured datasets.

Variations in performance: Performance of a given AI model may vary greatly depending on the type of language it is applied to, for example, customer reviews, contracts, medical records, scientific publications, patents, legal texts, long texts (such as books) very short texts (such as Tweets). The language used to train the algorithm should be the same as that of the text it will be applied to. For example, an NLP algorithm that was trained on Tweets would perform poorly if applied to medical records.

Labeling data: This is paramount. Determine if any part of the process requires humans to create labels for your data, and budget accordingly. Labels are necessary to perform supervised learning, whereby an algorithm learns from the training data set based on a predefined outcome. The collection, organization, and labeling of data are essential before embarking on an NLP path.

Scalability is another implementation factor. Algorithms and services will not work the same on every language and may work poorly or not at all for others.

NLP examples in use at mass scales now include:

• Conversational agents, such as virtual assistants and chatbots for customer support
• Machine translation (e.g., translating a website or a restaurant menu from Mandarin to English)
• Autocomplete and autocorrect on smartphones
• Information extraction from websites to optimize search engine results
• Classification of emails, such as distinguishing between spam and legitimate emails
• Improved customer service through the analysis of recorded customer calls
• Sentiment analysis of customer reviews used extensively in surveys.

For SMBs, these are some of the challenges of implementing NLP. In a follow-up post, we’ll discuss implementing Computer Vision for your business. Large enterprises with unlimited resources like Tesla have already had success with implementing CV, but it’s quite a challenge for small businesses.