Language Intelligent Times Call for Linguistics Theory Innovation
March 21, 2023 09:46 Source: "China Social Sciences" March 21, 2023 Issue 2614 Author: Li Bin Zhang Songsong

ChatGPT and other language intelligent technologies appear and apply,Linguistics、Language Teaching、Writing、Publishing and other fields brought a certain impact,also caused a lot of controversy。With the continuous influx of technology companies and the continuous increase in global users,Various support and opposition sounds continue to appear。Modern linguistics that is born in the early 20th century structural linguistics has developed for 100 years。Facing the challenges of language intelligent technology such as ChatGPT,Can linguistics answer the questions brought by the intelligent language intelligence? How to face the sound of support and opposition?、New Challenge,Need to actively respond to linguistics through unremitting exploration and theoretical innovation。

  bet365 best casino games  

In the relevant discussion on ChatGPT,Supported sounds can be summarized at least two types。First, ChatGPT effectively promotes the development of big data and machine learning models,In funds、Talent、Technology、Application and other aspects will attract more resources,may be able to achieve greater technical progress and breakthroughs。It can be said to a certain extent,ChatGPT has passed the Turing test at the text level (the machine can talk to people like people),This will make more and more resources help this technology development based on big data and machine learning。The second is that ChatGPT is very easy to use,Should make good use of。ChatGPT is a very convenient tool,A simple command can generate the results of the required requirements。Although it is not perfect,But it can save a lot of time and energy。

At the same time,Related criticism and opposition sound can be summarized into three types。First, ChatGPT doesn’t know what you are doing。It Bet365 app download is just a model trained based on ultra -large -scale language data,I don't have much learning and cognitive ability,Not to mention thinking。Sometimes,The content it generates is not accurate,Frequent Zhang Guan Li Dai,It just looks very smooth。The second is that ChatGPT will interfere with the normal order of school teaching and thesis writing。Students can use ChatGPT to generate text to complete homework or write papers、Reading Report、program code, etc.,This may make it difficult for many students to receive normal knowledge and skills training,to interfere with the normal teaching order。Third is that ChatGPT challenges traditional linguistics、Psychology、Literature and other humanities disciplines,Many intellectual property rights have also produced、Internet fraud and other related legal issues。ChatGPT basically does not use traditional linguistics、Research results of psychology,and mainly depend on big data and neural network models。For linguists,This is a very big challenge。Linguistics has a glorious history,There are a lot of phenomenon description and law summary of various languages,but failed to provide the theoretical foundation for products such as ChatGPT。

 Reasonable view 

How should linguists treat the impact of language intelligent technology such as ChatGPT rationally? Want to answer this question,Still going back to the papers published in the 1950s "The Three Models Describing Language"。In this paper,Jameski pointed out the problem of the Markov model,It is believed that the mathematical model of Markov is not enough to generate a legal natural language sentence。When comparing human children with this empiricist mathematical model,Jamesky thinks,Human children can learn to speak around 3 years old,But they do not need large -scale language data as the basis of learning,You only need less samples to learn language。and,Jameski distinguishes human congenital language Bet365 lotto review acquisition devices (brain hardware) and the acquired language acquisition process,Study on sentences that can be generated to generate legal sentences。In recent interviews,Jamesky thinks,Chatgpt is based on high -tech "plagiarism" on massive data,Chatgpt is a waste of resources。

The Malkov model pays attention to the continuing probability problem between the internal and afterward words of the sentence。As a pioneer of language data -based statistical learning model,This model was proposed as early as the 20th century。But until the 1980s and 1990s,With the continuous development of the large -scale storage capacity of language data and the continuous development of computer computing power,This model is only voice recognition、Input method、Word marking and other tasks are shining,and occupy a dominant position in the field of computing linguistics for about 20 years。After that,This model is gradually replaced by other better statistical learning models (such as maximum entropy model、Support vector machine、Conditions Raise Pyms, etc.)。Since 2006,Neural network model based on deep learning continues to make progress,In voice、Image、Text and other fields have achieved extraordinary achievements。and the neural network model has been proposed in the 1940s。After more than 60 years,This model continues to evolve with the continuous development of computer software and hardware,The effect is getting better and better,Not only can it generate more and more legal sentences,can also "understand" human language better。Natural language processing technology based on big data and machine learning,It has become the mainstream of computing linguistics and industry。

Chatgpt also experienced this evolution,Previous GPT 1-3 generation,The performance is getting stronger,Keep refreshing the cognition of linguists。2018,GPT-1 is trained on data about 4.5GB,The parameters of the model are about 120 million。2020,GPT-3 is learning training on the corpus of about 570GB,The parameters Bet365 app download of the model are as many as 175 billion。ChatGPT uses deep learning technology to train a large model on large -scale language data,Generate a answer according to the user's question。In this process,It completes the two major tasks of natural language understanding and generating。Different computers and human brains,It is difficult to characterize like humans、Meaning of perception and understanding。The so -called understanding and generation,In the dialogue task, it has become a big model to generate answers。From the actual effect,,The role of machine learning is similar to the acquisition mechanism of human language,Massive data is similar to the language acquisition data of human beings,while the big model is like human language ability。In the foreseeable future,ChatGPT will continue to develop,or an integrated voice、Image、Video and even more modular machine perception data,Use the indication and operation of the meaning of multi -mode approximation,Forms a multi -mode dialogue system that is constantly updated and even more natural、A perfect human -computer interaction system。This strong development momentum,It should attract enough attention,Reason analysis of its principle、Advantages and Deadies,Discussion strategy。

  Active response 

Now it seems,Big data input+neural network model,or can be regarded as another language acquisition and generation mechanism that can be regarded as a human brain。Just like an airplane invented by humans,Flying does not necessarily need to have two wings that can be fanned like birds。Air Powerment、material science and various engines, etc.,Open the new world of aerospace。The main problem here is,Some new technologies are not born from traditional disciplines。Chatgpt's development route,It is a technical path independent of linguistics。It itself except the mathematical foundation and software and hardware technology,It has not established the theory of perfect Bet365 lotto review language,Getting major progress。For this,We need to develop new theories based on these technical practice、New method,The theoretical innovation of linguistics becomes a top priority。Specifically,New linguistics theory needs to explain three new problems。

First,ChatGPT why does not need the human brain,Can you get better human -machine dialogue effects under the conditions of big data and large computing power? in other words,Computers based on the von Nokaman structure and a neural network -based mathematical model,What kind of problems have been solved,Make ChatGPT can imitate human language ability to a certain extent。At present,This mainly relies on experts in the field of machine learning and computing linguists in the industry.。But in the existing discussion,They are also very surprised by the performance of ChatGPT,It can achieve better performance in the general field (rather than just specific fields such as weather forecast)。Current,They do not have a very clear theoretical system and theoretical interpretation。This may require the common participation of linguists,Clarifying the basic reason why ChatGPT is relatively successful。

Next,Can you use the technology of ChatGPT,The mystery of exploring human language ability based on big data? at present,Chatgpt's English ability is better than Chinese。Whether it is simply a data quantity,Or is it difficult to deal with Chinese than English? at the same time,We also need to further consider such a problem: Can we use big data and artificial intelligence methods to study language? The amount of data in human language is huge,But most of the ancients could not be recorded。In the 21st century information age,Human language,Especially the language written on electronic equipment,Milling hundreds of millions of land every day。Past,Linguist mainly uses the method of investigation,Research Language Phenomenon、Summary language rules。Today,Massive data on the Internet,Provided bet365 live casino games a lot of research materials for linguists。The scale of this original material is huge,It is difficult to read and grasp only by personal strength。For more than 400 years,The continuous development of astronomy uses the equipment such as telescope and other devices observes a large amount of astronomical data,Then use the calculation modeling method to continuously reconstruct the cosmic model,I got many important breakthroughs。So,In the 21st century,Can we use artificial intelligence and big data analysis technology,Mathematical models to help linguists analyze and build human language?

Last,Can I study computer -based language acquisition theory and methods? Super computers can conduct various parameter training based on massive language data in a short period of time。With the development of language intelligent technology,We may need to distinguish between two different language theories based on people and computers。On the one hand,The combination of two compositions can better study the fundamental attributes and laws of language。On the other hand,Explore machine -based language theory,It can help artificial intelligence technology move towards more mature language intelligence stages,So as to produce more and more useful language intelligent products for human society。More importantly,Language intelligent technology is constantly making computers a new test field outside the human brain。Experiments on the human brain have ethics、Limited to many factors such as the law; and on the new test field of the computer,Researchers can learn linguistics、Psychology、All discovery of disciplines such as neuroscience、Various laws、A variety of mathematical models for operation and experiments,so as to make it an important basis for verification and improvement theory,Help further development in these fields。

Chatgpt and other language intelligent technologies methods and applications,It brings a certain challenge to traditional linguistic theory,At the same time, it also brought an opportunity for linguistic theoretical bet365 live casino games innovation。Massive real language data、Ultra -large -scale data analysis and machine learning technology,All bring new resources and methods to linguistics,It provides an important foundation for linguistic theoretical innovation。

 (This article is the "Fourteenth Five -Year" planning project "New Ecological Construction and Practice for Teaching Resources for Artificial Intelligence" (D/2021/01/120) in Jiangsu Province. 

(Author Unit: School of Literature, Nanjing Normal University; School of Foreign Languages ​​at the School of Foreign Languages ​​of Jinling University of Science and Technology) 

Editor in charge: Zhang Jing
QR code icon 2.jpg
Key recommendation
The latest article
Graphics
bet365 live casino games
Video

Friendship link:

Website filing number: Jinggong.com Anmi 11010502030146 Ministry of Industry and Information Technology:

All rights reserved by China Social Sciences Magazine shall not be reprinted and used without permission

General Editor Email: zzszbj@126.com This website Contact information: 010-85886809 Address: 11-12, Building 1, Building 1, No. 15, Guanghua Road, Chaoyang District, Beijing: 100026