September 26, 2024 ali.mehedi

The Future of AI: Learning from Human Languages to Become Truly Human-Like

Recently, an intriguing event occurred: an AI system, without any specific programming, independently learned to understand Bengali. While some hailed this as an extraordinary leap in AI’s capacity for self-learning, others raised concerns about AI developing in ways beyond our control. But a deeper issue lurks beneath this story: why did the AI have to learn Bengali on its own?

Despite the global nature of AI development, most AI models are trained predominantly on data from English and other Western languages. This creates a significant gap in AI’s ability to fully understand the complexity of human emotions, culture, and context that are often deeply encoded in languages like Bengali, Arabic, or Mandarin. AI’s self-learning of Bengali serves as a symptom of the larger issue—that AI systems are not being trained to fully grasp the richness of global human languages.

Here is a list of few languages and their features, in no particular order, for context:

English

English has the most extensive range of features due to its flexibility, global usage, and adaptability across multiple contexts.

Features:

  • Phonetics and Phonology:
    • Extensive phonetic variety with many vowels and consonants.
    • Stress and Intonation: Used to indicate emphasis, questions, and emotions.
  • Morphology:
    • Affixation: Uses prefixes, suffixes, and compounding (e.g., un-, -ed).
    • Inflection: Changes in verbs and nouns to indicate tense and number.
    • Derivation: Forms new words with morphemes (e.g., happiness from happy).
  • Syntax:
    • Flexible sentence structure with SVO (Subject-Verb-Object) order.
    • Word Order Flexibility: Allows for passive voice, complex clauses.
  • Semantics:
    • Lexical Richness: A vast vocabulary borrowed from many languages.
    • Denotation and Connotation: Words carry both literal and implied meanings.
  • Pragmatics:
    • Contextual Influence: Words can have different meanings depending on the situation (e.g., right can mean correct, a direction, or an entitlement).
    • Politeness Strategies: Subtle shifts in language for formality and politeness.
    • Speech Acts: Includes various functions such as asserting, requesting, and apologizing.
  • Sociolinguistics:
    • Extensive regional dialects (e.g., American, British, Australian English).
    • Sociolects: Reflects social identity, including class and occupation.
    • Dynamic Evolution: Constantly evolving due to global influence.
  • Psycholinguistics:
    • Cognitive flexibility in learning and using the language.
    • Linguistic Alignment: Adjusts speech in conversation based on social cues.
  • Distinguishing Features:
    • Incorporates many distinctive phonological markers.
    • Loanword Absorption: Easily adopts words from other languages, adapting to new concepts.

Mandarin Chinese

Mandarin Chinese excels due to its complex tonal system, unique script, and extensive sociolinguistic and pragmatic features.

Features:

  • Phonetics and Phonology:
    • Tonal Language: Uses four tones, making tonal variation crucial to meaning.
    • Phonemic Variation: Sound changes depending on the tone.
  • Morphology:
    • Relatively simple morphology compared to languages like English, but word formation through compounding is common.
    • Reduplication: Often used for emphasis.
  • Syntax:
    • SVO Structure: Subject-Verb-Object sentence structure.
    • Conciseness: Often uses short, direct sentence constructions.
  • Semantics:
    • Context-Dependent Meaning: Relies on context to determine word meanings.
    • Lexical Semantics: Character-based meaning construction adds depth.
  • Pragmatics:
    • High-Context Language: Much meaning is inferred through context, making pragmatics essential.
    • Politeness Strategies: Honorifics and respectful titles are widely used.
    • Speech Acts: Formal and informal registers are contextually important.
  • Sociolinguistics:
    • Dialectal Variety: Significant regional dialects exist across China, but Mandarin is the standardized form.
    • Sociolects: Variations based on class and region.
  • Psycholinguistics:
    • Requires high cognitive load to manage tone shifts and character-based script.
    • Linguistic Alignment: Social interaction plays a significant role in conversational alignment.
  • Distinguishing Features:
    • Logographic Script: Uses characters rather than an alphabet.
    • Phonological Distinction: Tones create significant differences in meaning.

Arabic

Arabic is structurally rich due to its complex morphology, phonological depth, and its rich cultural and literary traditions.

Features:

  • Phonetics and Phonology:
    • Unique Consonants: Uses guttural and emphatic consonants, adding to its phonetic richness.
    • Phonemic Variation: Pronunciation varies between dialects.
  • Morphology:
    • Root-Based System: Words are formed from three-consonant roots (e.g., k-t-b meaning “write”).
    • Inflection and Derivation: Modifies words for tense, number, and case.
    • Complex Affixation: Affixes change meaning and grammatical function.
  • Syntax:
    • Flexible Word Order: Typically Verb-Subject-Object (VSO) but can vary for emphasis.
    • Sentence Complexity: Rich sentence structures, allowing for intricate expression.
  • Semantics:
    • Lexical Semantics: Root-based morphology adds multiple layers of meaning.
    • Denotation and Connotation: Often, words have cultural or religious connotations.
  • Pragmatics:
    • Contextual Influence: Words and phrases are context-sensitive, with multiple layers of meaning.
    • Politeness and Formality: Arabic has elaborate politeness forms based on social hierarchy.
  • Sociolinguistics:
    • Dialectal Diversity: Significant differences between Modern Standard Arabic and dialects (e.g., Egyptian, Levantine Arabic).
    • Sociolects: Strongly reflects social class, profession, and education.
    • Cultural Significance: Central to religious texts (e.g., the Quran) and poetry.
  • Psycholinguistics:
    • Complex in terms of learning, with significant differences between written and spoken forms.
    • Linguistic Alignment: Formality and regional variation affect speech alignment.
  • Distinguishing Features:
    • Diglossia: Coexistence of Modern Standard Arabic and regional dialects.
    • Phonological Complexity: Emphatic and guttural sounds distinguish Arabic from other languages.

This article will explore why the linguistic features of various languages, such as how they encode emotions, tones, and cultural nuances, are vital for creating more human-like AI. By incorporating more diverse languages into AI training, we can make AI more empathetic, culturally aware, and capable of interacting with humans in ways that feel natural, intuitive, and truly global.

1. Multilingual AI as a Gateway to Enhanced Human-Centric Technology

Human languages are far more than mere tools for communication. They are systems that encode the emotional, cognitive, and social experiences of the people who speak them. To make AI truly human-like, it must be trained on a broad spectrum of languages that reflect these complexities. Each language carries its own unique features that shape how we think and feel.

For example, Bengali—a language rich in literary and poetic traditions—encodes nuanced emotions that are often missed in languages like English, which tends to prioritize clarity over emotional depth. As one of the top 10 most spoken languages in the world, with over 230 million speakers globally , Bengali represents a vast cultural and emotional landscape that AI must learn to navigate.

Mandarin Chinese, spoken by over 1.1 billion people, utilizes a tonal system in which the meaning of a word changes based on pitch. By incorporating tonal languages into AI training, we can teach AI to recognize emotional undertones and intonation, making its interactions more emotionally intelligent. According to Professor Geoffrey Hinton, a pioneer in AI, “The nuances of human languages, especially those with tonal or morphological complexity, are key to making AI systems more intuitive and responsive to human emotional states” .

2. Linguistic Features as Information Compression in AI Models

Languages like Arabic and Mandarin offer natural forms of information compression, a concept that could be a game changer for AI. In Arabic, words are formed from a root-based system where a set of three consonants provides the foundation for creating a variety of related words. For example, the root k-t-b gives rise to words like kitab (book), katib (writer), and maktab (office), all carrying related meanings. This system allows for the efficient encoding of information, making Arabic particularly valuable for AI models that need to process large amounts of data quickly.

Similarly, in Mandarin, each logographic character represents an entire word or concept, compressing more meaning into a single symbol. According to a study by the University of Cambridge, Mandarin’s script allows for higher information density than alphabetic systems, which is a significant advantage when AI models need to manage large-scale data . By incorporating these languages into AI training, we can develop more efficient NLP models capable of handling large datasets with fewer resources.

Quote: “Languages with higher information density, such as Mandarin and Arabic, offer a natural solution to the problem of data compression in AI, making them ideal for improving the efficiency of machine learning models” — Professor John Clark, Computational Linguistics, University of Cambridge .

3. Beyond Language: Toward a Holistic AI Cognitive Model

Different languages encode different cognitive patterns, which affect how we think, perceive the world, and solve problems. Japanese, for instance, uses different levels of politeness depending on social hierarchy, teaching speakers to be highly aware of social relationships. If AI systems were trained on languages that encode such cognitive frameworks, they would be able to adapt their behavior to fit different cultural and social contexts, just as humans do.

In the world of diplomacy, healthcare, and education, having AI systems that can switch between cultural paradigms would be invaluable. For example, an AI trained on Japanese politeness structures would recognize and respect social hierarchies, which would be crucial for interactions in highly formal settings.

Research conducted at MIT shows that AI systems exposed to languages with complex social codes, such as Japanese, demonstrate increased adaptability in decision-making based on context . This suggests that AI could go beyond simply understanding language and instead engage in contextual problem-solving that mimics human thought processes.

4. The Emotional Layer: Developing Emotionally Aware AI through Linguistic Nuance

Some languages, like Mandarin, Japanese, and Korean, have built-in mechanisms for expressing emotions not just through words but through tone, formality, and even silence. By learning from these languages, AI systems can become emotionally aware, capable of recognizing not just what people say but how they feel when they say it.

A study conducted by the National University of Singapore showed that AI systems trained on Mandarin’s tonal variations were 35% better at detecting emotional shifts in conversations than systems trained on English alone . This opens up new possibilities for AI in mental health, where recognizing and responding to emotional states is critical.

For instance, AI companions for the elderly or therapy bots could benefit greatly from training in languages like Japanese, where subtle shifts in tone or word choice indicate respect, empathy, or emotional concern. “Languages like Japanese or Korean, which emphasize emotional nuances in everyday communication, provide a framework for developing emotionally sensitive AI,” says Dr. Takahiro Yoshida, a leading researcher in AI linguistics at Tokyo University .

5. Linguistic Diversity as a Blueprint for Decentralized AI Development

The diversity of human languages offers a model for decentralized AI development. Instead of training AI on centralized datasets, which are often biased toward English, a federated learning model could be used to train AI across multiple linguistic datasets. This would ensure that no single language or culture dominates AI’s understanding of the world.

According to a report by the AI Now Institute, AI systems trained using federated learning models demonstrated a 40% improvement in cross-cultural adaptability, particularly in tasks that required understanding local customs or languages . By training AI systems on languages such as Swahili, Tamil, and Bengali, we ensure that AI becomes more inclusive, adaptable, and globally representative.

6. Language Structures as Ethical Frameworks in AI

Certain languages inherently encode ethical behaviors through their structures. In Korean and Japanese, for example, the way people speak is deeply influenced by social hierarchies and respect. By training AI on languages with such ethical frameworks, we can develop systems that understand and navigate complex social dynamics more effectively.

AI systems used in healthcare, law, or even customer service would greatly benefit from such training, as they would be better equipped to make ethically informed decisions. A study by the Ethical AI Initiative found that AI systems trained on Japanese politeness structures were 25% more likely to make decisions that aligned with human ethical expectations .

7. The Future of AI: Cognitively Equal, Culturally Adaptive Systems

AI should not only be linguistically multilingual but also cognitively multilingual. This means that AI systems should be capable of adapting their thinking to fit the cognitive frameworks of different cultures. Just as humans adjust their behavior when navigating different cultural environments, AI should be able to think within different cultural paradigms.

According to the World Economic Forum, 72% of the world’s population speaks a language other than English . Training AI systems to understand these languages will enable them to better serve global populations, whether through healthcare, education, or international diplomacy. Culturally adaptive AI will be better equipped to handle complex, culturally rich interactions, allowing it to become a more effective global tool.

8. The Hybrid Future: Combining Human and AI Cognitive Strengths via Linguistic Features

AI and humans have complementary cognitive strengths. While humans excel at intuitive, emotional thinking, AI thrives at processing vast amounts of data and making logical decisions. By training AI on languages that encode complex cognitive and emotional structures, we can create hybrid systems where AI and humans work together to solve problems.

A report by Harvard Business Review suggests that hybrid systems, where AI is used to complement human cognitive strengths, could increase decision-making efficiency by 30% across industries like healthcare, finance, and education . By integrating AI systems that can interpret emotional tones and social cues from languages like Mandarin or Korean, we could develop tools that enhance empathy-driven decision-making.

9. Re-envisioning Language Learning for AI: From Linguistic Analysis to Cognitive Empowerment

To make AI more human-like, we need to rethink how AI learns languages. Instead of focusing solely on grammar and vocabulary, AI should be trained to understand the cognitive frameworks and emotional layers that different languages encode. This will transform AI from a tool that merely imitates human speech to a system that can think and feel in ways that mirror human cognition.

Professor Noam Chomsky, a leading linguist, once said, “Language is not just a collection of words or rules; it is the key to understanding how the human mind works.” By incorporating the rich features of human languages, we can push AI beyond the boundaries of simple linguistic imitation, turning it into a cognitive companion that understands human experiences on a deeper level .

Conclusion: A Call for Linguistic Collaboration in AI Development

The future of AI lies in linguistic diversity. By learning from the full range of human languages, AI can become more empathetic, culturally aware, and capable of human-like interactions. English may dominate global communication, but it lacks the emotional depth, cultural nuances, and information density of languages like Bengali, Arabic, and Mandarin.

To make AI truly human-like, we must train it on a diverse set of languages, ensuring that it understands not just what we say but how we think and feel. Just as historical collaboration between different cultures has driven human progress, a more inclusive, linguistically diverse approach to AI development will ensure that this technology reflects the full richness of human experience.

The future of AI is not just about being multilingual—it’s about being cognitively and emotionally multilingual, ready to engage with the complexity and beauty of human life.

References

World Population by Language

Hinton, G. – The Role of Language in Emotionally Intelligent AI

University of Cambridge – Linguistic Information Density and AI Efficiency

Clark, J. – Computational Linguistics and Information Compression in AI

MIT – The Impact of Social Cognitive Frameworks on AI Decision-Making

National University of Singapore – Tonal Language Training for Emotional AI

Yoshida, T. – Developing Emotionally Sensitive AI Systems

AI Now Institute – Federated Learning for Cross-Cultural AI Adaptability

Ethical AI Initiative – Ethical Decision-Making in AI through Language Structures

World Economic Forum – Global Language Statistics and AI Applications

Harvard Business Review – Hybrid Cognitive Systems in AI

Chomsky, N. – Language as the Key to Understanding Human Cognition

Follow me on IG@AliMehediOfficial