What is Natural Language Processing (NLP)?

The study of how computers interact with human language, particularly how to design computers to process and analyze massive volumes of natural language data, is known as natural language processing (NLP), a subject of linguistics, computer science, and artificial intelligence. The ultimate objective is to create a machine that can “understand” the contents of papers, including the linguistic subtleties that arise from their context. After that, the system can accurately extract the knowledge and insights from the papers as well as classify and arrange the documents themselves.

History of NLP

The field of natural language processing dates back to the 1950s. Although it was not recognized at the time as a different issue from artificial intelligence, Alan Turing presented the now-famous Turing test as a measure of intelligence in a 1950 paper titled “Computing Machinery and Intelligence.” The suggested test consists of a task that calls for the automatic synthesis and interpretation of natural language.

Symbolic NLP (1950s – early 1990s)

John Searle’s Chinese Room Experiment is a good summary of symbolic NLP’s foundational idea: The computer simulates natural language comprehension (or other NLP tasks) by applying a set of rules (for example, a Chinese phrasebook with queries and corresponding answers) to the data it encounters.

• 1950s: Fully automated translation of more than sixty Russian phrases into English was used in the Georgetown experiment in 1954. The authors predicted that machine translation will be a solved problem in three to five years. Real advancement was, however, far slower, and funding for machine translation was drastically cut following the ALPAC report in 1966, which concluded that ten years of study had fallen short of expectations. Up to the creation of the first statistical machine translation systems in the late 1980s, little more study into machine translation was done.
• 1960s: ELIZA, a simulation of a Rogerian psychotherapist created by Joseph Weizenbaum between 1964 and 1966, and SHRDLU, a natural language system operating in constrained “blocks worlds” with constrained vocabulary, were two of the more successful NLP systems created in the 1960s. ELIZA occasionally offered a surprisingly human-like encounter while having essentially little knowledge of human cognition or emotion. When the “patient” went beyond the extremely tiny knowledge base, ELIZA may give a general answer, such as asking “Why do you claim your head aches?” in response to a statement like “My head hurts.”
• 1970s: Many programmers started creating “conceptual ontologies” in the 1970s, which organized real-world facts into computer-understandable data. MARGIE (Schank, 1975), SAM (Cullingford, 1978), PAM (Wilensky, 1978), TaleSpin (Meehan, 1976), QUALM (Lehnert, 1977), Politics (Carbonell, 1979), and Plot Units are a few examples (Lehnert 1981). The earliest chatterbots, including PARRY, were created during this period.
• 1980s: The golden age of NLP’s symbolic techniques was in the 1980s and early 1990s. Research on rule-based parsing, morphology, semantics, reference (e.g., within Centering Theory), semantic operationalization of generative grammar (e.g., the development of HPSG as a computational operationalization of generative grammar), and other areas of natural language understanding (e.g., in the Rhetorical Structure Theory) were focus areas at the time. Additional research was carried out, such as the creation of chatterbots using Racter and Jabberwacky. The increasing significance of quantitative evaluation at this time was a significant development (which ultimately contributed to the statistical shift in the 1990s).

Statistical NLP (1990s–2010s)

The majority of natural language processing systems were built on intricate, manually developed rules up until the 1980s. But with the advent of machine learning algorithms for language processing in the late 1980s, there was a revolution in natural language processing. This was caused by both the gradual decline of Chomskyan linguistic theories (such as transformational grammar), whose theoretical underpinnings discouraged the kind of corpus linguistics that is the basis of the machine-learning approach to language processing, and the steady increase in computational power (see Moore’s law).

• 1990s: The majority of natural language processing systems were built on intricate, manually developed rules up until the 1980s. But with the advent of machine learning algorithms for language processing in the late 1980s, there was a revolution in natural language processing. This was caused by both the gradual decline of Chomskyan linguistic theories (such as transformational grammar), whose theoretical underpinnings discouraged the kind of corpus linguistics that is the basis of the machine-learning approach to language processing, and the steady increase in computational power (see Moore’s law). Due mostly to work done at IBM Research, many of the early major accomplishments on statistical approaches in NLP happened in the area of machine translation. Due to regulations requiring the translation of all governmental proceedings into all official languages of the relevant systems of government, these systems were able to make use of multilingual textual corpora already created by the European Union and the Canadian Parliament. However, the majority of other systems relied on corpora created especially for the tasks that these systems accomplished, which was (and frequently still is) a fundamental constraint in the effectiveness of these systems. As a result, a lot of study has been done on how to learn from limited bits of information more successfully.
• 2000s: Since the mid-1990s, as the web has developed, a growing amount of raw (unannotated) linguistic data has become accessible. As a result, unsupervised and semi-supervised learning algorithms have received more attention in research. These algorithms can learn from data that hasn’t been manually annotated with the answers they’re looking for or from a mix of annotated and unannotated data. Compared to supervised learning, this activity is often significantly more challenging and typically yields less accurate outcomes for a given quantity of input data. However, if the algorithm employed has a low enough temporal complexity to be useful, the vast amount of non-annotated data accessible (containing, among other things, the whole corpus of the World Wide Web) may frequently make up for the subpar results.

Neural NLP (present)

Representation learning and machine learning techniques inspired by deep neural networks gained popularity in natural language processing in the 2010s. This popularity was partially attributed to a rush of studies demonstrating the state-of-the-art performance these approaches can attain in a variety of natural language tasks, such as language modeling and parsing. This is becoming more crucial in the fields of medicine and healthcare, where NLP is used to evaluate notes and text included in electronic health records that would otherwise be inaccessible for research in the quest to enhance treatment.

Common NLP tasks

The following is a list of some of the most commonly researched tasks in natural language processing. Some of these tasks have direct real-world applications, while others more commonly serve as subtasks that are used to aid in solving larger tasks.

Though natural language processing tasks are closely intertwined, they can be subdivided into categories for convenience. A coarse division is given below.

Text and speech processing

• Optical character recognition (OCR): Determine the text that corresponds to a picture of printed text.
• Speech recognition: Determine the textual representation of the speech from a sound sample of the speaker(s). This is the inverse of text to speech and is one of the exceedingly tough challenges informally labeled “AI-complete” . Since there are seldom any gaps between words in actual speech, speech segmentation is a crucial component of speech recognition. The process of converting an analog signal to discrete characters may be exceedingly challenging since, in the majority of spoken languages, the sounds that correspond to successive letters coarticulate, or mix into one another. The voice recognition software must also be able to detect the wide range of input as being equal to each other in terms of its written equivalent since words in the same language are pronounced by persons with various accents.
• Speech segmentation: Identify the words in a sound sample of a speaker or speakers. Usually grouped with voice recognition as a subtask.
• Text-to-speech: After the text has been transformed, provide a spoken representation of those components. Text-to-speech technology can be helpful for those who are visually impaired.
• Word segmentation (Tokenization): Divide a long string of words into individual ones. This is not too difficult in a language like English because spaces are typically used to separate words. However, other written languages, such as Chinese, Japanese, and Thai, do not distinguish between words in this way, and text segmentation in those languages is a difficult operation that necessitates familiarity with the lexicon and morphology of the language. This procedure is occasionally applied for creating bags of words (BOW) for data mining.