So far, this series has addressed two key elements of Generative AI: Large Language Models and Natural Language Processing, as well as the surprising long history of NLP. Now the author wraps up the series with some practical insights about the advantages and disadvantages of using NLP.
The Advantages of Natural Language Processing
Just like Generative AI, Natural Language Processing offers a distinct set of benefits, which are as follows:
- Big Data: NLP models can take an enormous amount of data such as images, video, audio, documents, and completely ingest them for training purposes. Further, the NLP model can also make these datasets scalable so that they can be used for any kind of subsequent applications.
- An objective analysis: NLP is quite beneficial in terms of offering an unbiased view that you may get from a research instrument, such as that of a survey. Humans can be biased when interpreting the results, and NLP can be a great help here.
- Improvement in end user satisfaction: Another great benefit of an NLP model is that it can run on a 24 X 7 X 365 basis. Thus, it would be perfect for the Digital Person, but also for soliciting feedback from end users. In fact, it can also integrate with some of the leading Customer Relation Management (CRM) systems, such as Salesforce and Microsoft Dynamics.
- Foreign language automation: The business world today is expansive and global, so if your business is large enough, there is a good chance that you can find offices in other countries as well. For the employees that work in those offices, English could very well be a second language. The advantage of NLP here is that your foreign based employees can speak in their native tongue, and this can be translated back into English, by making use of an NLP model. By giving them this kind of flexibility, overall productivity could also be increased as well.
The Disadvantages of Natural Language Processing
Along with the benefits of Natural Language Processing come the disadvantages as well. Here is a sampling of them:
- The use of contextual words: Although NLP is getting better at Word Sense Disambiguation, it can still get “confused” if a word is used multiple times in a long block of text. For example:
-
- “I went to the store because we ran out of baby formula”
-
- In the above sentence, the word “ran” has two different contextual meanings to it. An NLP model may not be able to discriminate this, but it can if the above sentence were broken down into two, separate sentences:
-
- “I ran to the store.”
-
- “I ran out of baby formula.”
-
- Also, the use of synonyms (in which two different words have the same meaning, such as “shut” and “close”) and homonyms (in which two words sound the same, but have a totally different meaning, such as “there” and “their”) can also potentially “confuse” an NLP model.
- The use of irony and sarcasm: These are words that are spoken in everyday language and conversations, and as humans, we can do a reasonably good job in discerning them. But this is not the case with an NLP model. For instance, while it can ascertain if a sentence is positive or negative, it cannot determine the opposite of the other. Here is an example:
-
- “United Airlines just told me that I will get a refund for my ticket”
-
- “Yea, right!!!”
-
- While the NLP model can infer that the first sentence is positive sounding, it will not be able to conclude that the second sentence is negative, or sarcastic, in nature.
- Understanding the different kinds of ambiguity: Ambiguity refers to sentences or even phrases that can have different interpretations associated with them. There are two specific types of ambiguity, which are as follows:
-
- Lexical ambiguity: This is where a word can be used as a verb, noun, or adjective.
-
- Semantic ambiguity: This is where an entire sentence can be interpreted in two different ways.
- Errors in text and speech: These include errors that the NLP model does not pick up on, including:
-
- Misspelled words
-
- Misused words
-
- Mispronunciations
-
- Different (foreign) accents
-
- Stuttering.
- The use of industry specific language: At this time, an NLP system cannot be used as a “one size fits all” solution, because one NLP model cannot be used for different market applications or industries, since each one has its own lingo and terminology. Therefore, you would need a dedicated NLP model for each one.
- The use of Lower-Level languages: A Lower-Level Language, aka low resource language, can be technically defined as follows:
“Low resource languages are those that have relatively less data available for training conversational AI systems.”1
This refers to less commonly spoken languages (such as Vietnamese, Swahili, Urdu, etc.). There is not enough data about such languages to feed into an NLP model so that it can learn them. In contrast, the Higher-Level Resource Languages (such as English, Chinese, French, Spanish, German, etc.) are far more widely used and spoken, thus there is much more data to ingest into an NLP model as compared to Lower-Level Languages.
- Ethical/social considerations: It is important to note that while the primary intention of an NLP model is to make it unbiased, neutral, and objective, NLP is still inherently none of those things. At this time, NLP only reflects what has been fed into it along with other various resources that it has found in order to compute the output. As a result, NLP can potentially greatly magnify the biases and flaws in thinking that exist in the datasets, or even worse, in society as a whole.
This concludes our series on Generative AI. Given how rapidly this technology is evolving, for better and for worse, we will undoubtedly continue to address this topic in future articles.
Sources/References:
1.Keyreply.com
Ravi Das is a Cybersecurity Consultant and Business Development Specialist. He also does Cybersecurity Consulting through his private practice, RaviDas Tech, Inc. He also possesses the Certified in Cybersecurity (CC) cert from the ISC2.
Visit his website at mltechnologies.io
