Presented below is an English translation of the Japanese text that I prepared for my lecture at ELPA on June 19, 2021.
The translation process was as follows.
(1) I inputted the Japanese text with no particular pre-editing into DeepL.
(2) I post-edited the English output from DeepL by myself.
(3) I copied the revised English to Grammarly and edited it slightly more, following some of its suggestions.
The user experience was pleasant. Post-editing did not take much time because DeepL's output was satisfactory in most cases from my point of view. I saved a massive amount of time, which I would have needed without AI. Furthermore, I did not consult native English speakers. The experience may demonstrate that non-native English users benefit significantly from AI if they use it adequately, as I indicate below. AI may empower non-native English users to be more autonomous.
The PDF file is downloadable from here.
A Report on the Future
of University English Education
in Light of the
Development of AI
Analogy
Adults use calculators and
spreadsheet apps to perform complex calculations.
Few adults, however, allow children
to use them from the start.
Children should develop a sense of
number before using such convenient tools.
Hypothetical
Question
What if calculators and
spreadsheet apps were only right, say, 95% of the time?
0 Summary
Despite
its remarkable progress, the current mainstream AI (Artificial Intelligence) has
structural and functional limitations: it can assist and extend the
intelligence of the individual who uses it, but it can never replace it.
Therefore, for the use of English by non-native speakers, users need to judge
the output of AI and make necessary corrections because AI is only a tool as an
auxiliary and extended intelligence. For the area where AI is not of sufficient
help, in particular, English language users must acquire English language skills
in their flesh and blood. Therefore, future university English education, which
must assume that graduates will use AI later in their career, should transform
qualitatively and focus on teaching English skills that AI cannot aid, extend,
or substitute human abilities. Universities should enable learners to use
English after graduation by exploiting AI as a tool that can successfully
assist and extend their physically embodied English skills.
The
current paper presents the author’s proposal for the future of university
English education (teaching academic English at a research-oriented university)
based on a theoretical review of AI. The author believes that the proposal
should be examined for its validity and feasibility by all those involved in
English education (beneficiaries, instructors, policymakers) and that reform of
English education, both drastic and cautious, should be undertaken.
1 Theoretical Review of AI
1.1 Three principles for the
relationship between AI and humans
On
the basis of the structural and functional limitations of AI described below,
this paper argues that users should agree on the following three principles.
Principle
1: AI
can assist and extend human intelligence, but it will not be a complete
replacement.
Principle
2: Humans
must judge and correct AI output when necessary.
Principle
3: Humans
must take the initiative and responsibility for the use of AI.
1.2 Structural and functional limitations
of AI
Structural Limitation
(1) Lack of a biological body
filled with emotion:
AI behavior is categorically different from human behavior that prioritizes
life; AI possesses no biological body that produces "emotion" to
sustain life and thrive.
(2) Lack of sufficient world models: Since AI has not gone
through the process of evolution and selection, it has not internalized working
models of the world for better chances of survival. Therefore, AI can only use
the architecture and data provided by the programmer. Consequently, it cannot utilize
models of the world to successfully learn and make inferences from a small
amount of data, as humans do.
(3) Inability to understand
meanings and stories:
AI's perceptions are restricted to very limited distinctions. Thus, it cannot infer
from its perception other diverse aspects of the object it recognizes (the
"actuality" of meaning) or the numerous potential connections that
the object could have (the "potentiality" of meaning), as humans do.
Even when AI can recognize multiple objects simultaneously, it does not understand
the mutual relationships they can possess. In other words, AI cannot understand
in the form of a "story," which is a coherent constellation of various
meanings.
(4) Inability to create new values
and hypotheses:
AI can only learn and reason about predetermined issues. AI cannot invent new and
significant viewpoints (values) and conceptions (hypotheses) beyond the domain
of those issues.
(5) No social communication: AI only learns as an
individual entity and cannot maintain contingent correspondences (i.e., communication)
with equal but different entities. It, therefore, does not experience unexpected
qualitative transformations, as humans do.
Functional Limitation
(6) Weakness in long-tail phenomena: AI is weak in learning
atypical and exceptional phenomena (i.e., cases that are few in number but exist
in many kinds in the real world). AI often err on rare items in big data, including
specialized knowledge.
(7) Mistakes that are unthinkable
for humans:
Because AI’s "understanding" is categorically different from human’s
understanding, AI produces mistakes that humans do not usually predict.
(8) Only domain-specific learning
and reasoning with no analogical application: Even when AI shows superhuman
ability in a limited task, it does not possess reasonable ability in related domains
other than that. AI is not flexible or versatile.
2 The Future of University English
Education
2.1 General Discussion
From
the structural and functional limitations described above, we may formulate the
following guidelines for university English education in the future.
(1) Emphasizing emotional
experiences through English:
Instructors should not reduce English learning to a mere formal manipulation of
signs. They should emphasize the emotions that arise in students through
English and the reactions they initiate from those emotions, for emotions are
the source of human cognition and behavior.
(2) Learning about the world
through English:
Instructors should recognize again that learning English is about learning
about the world. They should choose English materials and learning tasks that
have strong connections to the real world. The relationship between language
and the world is one area in which AI is weak (AI processes language only as forms).
(3) Learning about the
possibilities of English expressions: Learners should sufficiently understand
the potentiality of meaning that an English expression can have in addition to its
literal meaning (the limited meaning that even objective tests can determine).
Learners should also be proficient in understanding and expressing meaning in a
narrative form that integrates various meanings coherently. AI cannot constellate
different meanings coherently.
(4) Emphasizing creative responses from
understanding English:
Even classes for receptive skills should not end with reading and listening; it
should begin with reading and listening. Even classes for reading and listening
should teach producing relevant responses that can result from that
comprehension. AI can only process what it is designed to process, unable to
respond creatively from that processing.
(5) Developing the ability to
collaborate with multiple people in English: Learners should use English
appropriately with numerous people with different backgrounds and understandings.
Instructors should stop basing lessons solely on the evaluation of
individual-based learning using uniform criteria. AI cannot perform social
communication, i.e., the coordination of relationships among multiple
individuals despite differences and contradictions.
(6) Emphasizing expressions about
non-typical and exceptional matters: Students should enhance the learning
of expressions that are “unusual” from the viewpoint of big data (e.g., technical
terms and Japanese idioms), where AI tends to err; AI is often wrong about such
long-tail phenomena.
(7) Learning about the mistakes
that AI can make:
Learners should learn to identify and correct unexpected mistakes that AI commits.
They should be free from a widespread myth that AI is accurate and fair because
it is a machine.
(8) Learning to respond flexibly: Learners should learn to
invent ad hoc expressions by which humans communicate without knowing precise
expressions. Humans should advance their flexible adaptability, which AI does
not possess.
2.2 Writing
In
Academic English writing, the following writing process will probably enhance non-native
English users’ writing skills through AI. Therefore, in university English
education, instructors should teach the writing strategies below. However,
instructors need advanced knowledge and skills in both Japanese and English.
(1) Writing in Japanese: Learners can write the
manuscript in Japanese, which allows them to organize and express their
thoughts most precisely without losing concentration over a long time. However,
additional guidance may be necessary for learners whose mother tongue is not
Japanese.
(2) Pre-Editing: After learning conspicuous
differences between Japanese and English, learners should revise the Japanese
manuscript into machine-friendly Japanese to prevent errors in machine
translation as much as possible in advance. However, critical consideration may
be necessary about this kind of modification of Japanese into an English style
from the perspective of linguistic and cultural diversity and the domination of
English that suppresses it.
(3) Post-editing: Students need to
revise the English output that machine translation produced. The revision
requires a high level of English language ability to objectively read the
output English from a third-party perspective and evaluate it stylistically.
2.3 Reading
As
mentioned above, advanced reading skills are essential for AI-based writing.
They are critical for learning the vocabulary necessary for listening and
speaking (i.e., acquiring detailed knowledge of the collocational
possibilities, which cannot be obtained by rote memorization of words). Reading
instruction in the future requires more than a rough translation for
approximate understanding, which machine translation achieves instantly.
Specifically, instructors can consider the following four policies.
(1) Stylistically analytic reading: Students should read carefully
selected English text, compare it with other possible expressions, and accurately
understand the English text’s meaning. Since AI cannot understand the subtle
possibilities of meaning, humans need to develop their skills in this stylistically
analytic reading. This intensive analysis is also necessary for post-editing in
writing.
(2) Psychosomatic expression in the
style of reading theatre:
Learners need to feel the sounds of the English text and the emotional
vibrations they generate in their bodies. They should also read the text aloud themselves,
resonating with their emotions in the manner of reading theatre. Since AI and
robots have no biological body like humans’ and cannot express themselves
emotionally, psychosomatic expression in the style of reading theatre may be crucial
for English learners.
(3) Translation writing: Instructors should
encourage students to write Japanese translations to know their
understanding of the high-quality English texts they selected. Translation
experience will help students gain a deeper and more accurate knowledge of both
Japanese and English. Translation writing is also critical for identifying and
correcting errors in AI translation in the post-editing process.
(4) Teaching with tasks and
projects:
The real value of English comprehension in the real world depends on what
actions the reader can take from that comprehension. To make reading
instruction "begin with reading" rather than "end with
reading," instructors should integrate English reading into other
meaningful tasks or projects. The successful use of language to complete tasks or
projects is necessary for the development of the human capacity to connect
symbolic understanding to real-world actions. AI cannot render symbolic
processing into real-world actions because it only processes symbols as formal
notations.
2.4 Speaking
Human
speaking is not just a production of linguistic signs; it is an expression that
involves the whole body's emotions. Humans express and understand each other
not only through linguistic signs. They also communicate through all paralinguistic
expressions (prosody, such as rhythm and intonation) and non-verbal expressions
(eye contact, facial expressions, gestures, movements, among others) that emerge
simultaneously. The AI/robot with no emotional body cannot perform this integrated
task. Therefore, with regard to speaking, humans cannot expect much assistance
or expansion from AI. Instructors must emphasize teaching speaking skills in
the future.
Even
when humans use AI to convey information that does not require much emotional
expression or understanding, AI cannot accurately represent long-tail items
such as genre-specific technical terms that academic English contains. Speaking
instruction in university English education should aim at developing learner’s
embodied speaking skills without the help of AI.
2.5 Listening
AI
correctly recognizes suprasegmental sound changes (linking, reduction,
assimilation, among others), which are features that many Japanese learners
have not mastered. On the other hand, AI does not always successfully recognize
specialized expressions, which are well-known to specialists but belong to the
long tail in Big Data. Thus, while AI can help with teaching suprasegmental features,
it cannot entirely replace human listening skills.
Listening
is not limited to the reception of linguistic signs, either. It also includes
the emotional responses that emerge in the listener’s body. The speaker observes
those emotional expressions to judge whether the listener understands
appropriately. Listening instructors must also attend to the listener's
emotional response.
In
addition, the understanding and embodiment of the sound features of English is
critical are speaking. Speakers who acquired sound feature patterns can
reproduce them appropriately when they express themselves in speaking. Speech
with appropriate sound features definitely promotes better comprehension in
listeners.
Those
considerations above lead to the following two guiding principles.
(1) Listening experience until the sound
features of English are embodied: The goal of listening instruction should not be
limited to accurate recognitions of linguistic signs. It should include appropriate
emotional response and, ultimately, the embodiment of sound features of
English. The embodiment, in particular, requires analytical and conscious
training because they are distinct from Japanese features. With sound features
of English embodied, learners can express themselves far more effectively.
Also, in silent reading, learners understand more comfortably if they can appropriately
vocalize the printed text in their mind; silent reading is, after all, listening
comprehension of sounds rendered from letters on the text. Writing, too, is producing
texts that readers can comfortably vocalize in their minds. Therefore,
mastering the sound features of English through listening helps learners to
write better. Listening instruction should no longer be about receiving
information and determining its correctness through multiple-choice tests.
(2) Selecting listening materials on
the basis of individual learners' interests: The English subtitling function
of the Chrome browser automatically recognizes and transcribes English audio on
the Web. It converts English videos that most stimulate learners’ particular intellectual
interests into learning materials with subtitles. (Imperfections in the
subtitles need to be addressed, though.) Learners can turn the transcribing function
on and off and change the difficulty of listening so that they can train
themselves to improve their listening skills. Since learners are motivated by
the material they choose, they more accurately perceive the subtle nuances of
meaning expressed in its sound features. Students can repeat this meaningful
listening experience abundantly until learners embody the sound features of
English they hear.
3 In Closing
The
direction of university English education described above obviously assumes
that learners have acquired a certain level of English ability before they
enter university. As an English instructor at university, it is difficult for
me to predict how English education in elementary, junior high, and high schools
will change with the development of AI. However, the review and analysis above
suggest that English teaching before university should maintain the principles
of "teaching and evaluating abilities students embody in their flesh and
blood" and "fostering human-specific abilities.” The challenge for
education in the future lies in developing human intelligence that coexists with
AI. Because English education at and before university is closely related, those
involved in university English education must continue to pay attention to the
state of English education in elementary, junior high, and high schools.
The
current report theoretically reviewed AI to propose some possibilities of the
future English education at university (and below that level, briefly). As
mentioned at the beginning, this report merely reflects the author’s
consideration. The author hopes that this document will serve as a starting
point for a deeper discussion on the future of English education.
References
瀧田寧・西島佑（編著）
(2019) 『機械翻訳と未来社会』 社会評論社
藤本浩司・柴原一友
(2019a) 『AIにできること、できないこと』 日本評論社
藤本浩司・柴原一友
(2019b) 『続 AIにできること、できないこと』 日本評論社
松尾豊
(2015) 『人工知能は人間を超えるか ディープラーニングの先にあるもの』角川書店
松尾豊・塩野誠
(2016) 『人工知能はなぜ未来を変えるのか』角川書店
松尾豊
(2019) 「深層学習と人工物工学」 https://www.jstage.jst.go.jp/article/oukan/2019/0/2019_F-5-2/_pdf
松尾豊
(2020) 「人工知能 ディープラーニングの新展開」、西山圭太・松尾豊・小林慶一郎 (2020) 『相対化する知性』日本評論社 (pp. 1-103)
丸山宏
(2019) 「高次元科学への誘い」https://japan.cnet.com/blog/maruyama/2019/05/01/entry_30022958/
丸山宏
(2019) 「人工知能研究者として私たちがすべきこと」https://japan.cnet.com/blog/maruyama/2019/12/31/entry_30022985/
ミッチェル,
M. 著、尼丁千津子訳 (2021) 『教養としてのAI講義』日経BP
(Mitchell, M. (2020) Artificial Intelligence: A Guide for Thinking Humans.
Pelican.)
Posts
