Questions: Language and Artificial Intelligence

Question 1

Large language models trained on billions of words achieve high performance on many NLP tasks despite not having explicit linguistic rules programmed into them. This suggests:

• A. Linguistic rules are unnecessary and language is purely statistical
• B. Models can achieve human-level understanding through learning statistics alone
• C. Statistical patterns in text can approximate many linguistic phenomena, though understanding of structure and meaning remains limited
• D. Language is entirely computable without any formal structure

Answer: (C)

Models learn statistical patterns effectively, achieving strong performance on many tasks. However, they often fail on linguistic phenomena requiring structural understanding (e.g., complex syntax, long-range dependencies, abstract reasoning). Success on statistics doesn't prove rules are unnecessary or that models understand.

Question 2

When a language model generates text that appears meaningful and fluent but is factually false, this reveals:

• A. The model is intelligent and reasoning about the world
• B. The model lacks grounding in real-world knowledge and meaning; it's generating statistically plausible text, not reasoning
• C. Language is meaningless
• D. AI has achieved sentience

Answer: (B)

Models generate statistically plausible sequences but lack grounding in world meaning. They can produce fluent-sounding but factually false text because they're optimizing statistical likelihood, not truth. This shows a fundamental limitation: statistical learning alone doesn't produce understanding.

Question 3

Linguistic theory contributes to artificial intelligence primarily by:

• A. Providing rules to program into systems
• B. Identifying linguistic phenomena and structures that computational models should explain; understanding linguistic constraints informs model design
• C. Competing with AI for understanding language
• D. Linguistic theory is irrelevant to AI development

Answer: (B)

Linguistic theory identifies what needs to be explained (e.g., garden-path effects, pragmatic inference, abstract dependencies). This guides AI research: which phenomena should models handle? What structures are important? Linguistic insights inform model architecture and training approaches.

Question 4

Current language models demonstrate that human language understanding is based purely on statistical learning and no abstract linguistic structure or concepts are necessary.

• T. True
• F. False

Answer: False

Models achieve impressive statistics-based performance but show limitations on structural and semantic phenomena. Their capabilities don't prove human understanding is statistical. Human language involves abstraction, reasoning, embodied knowledge, and social understanding that statistical models haven't achieved.

Question 5

Explain the relationship between linguistic theory and AI development: how does understanding language structure inform better AI systems, and how do AI systems' capabilities and limitations inform linguistic theory?

Model answer: Linguistic theory identifies structures, principles, and phenomena that any complete language system must handle. This guides AI research: what to build systems to explain. When models struggle with phenomena linguistic theory emphasizes (recursion, long-range dependencies), this shows these are computationally challenging. When models succeed on unexpected tasks, this suggests additional structure/patterns in language. The relationship is bidirectional: theory guides AI research, and AI systems' performance informs theoretical understanding.

Theory and practice inform each other. Neither linguistic theory nor AI alone fully explains language. Together, they reveal what structure is computable through statistics, what requires explicit representation, and what aspects of language understanding remain mysterious.