Questions & explanations
1. In the sentence 'The complex houses married and single soldiers,' what role does 'houses' initially suggest, and what is the correct role?
Initially, 'houses' looks like a verb (agent = complex), so the listener thinks 'the complex houses' means the complex provides housing. But then 'married and single soldiers' appears, and the correct parse is that 'houses' is a noun (patient of 'complex'), and the verb is missing? Actually, the sentence means 'The complex houses married and single soldiers' as a headline: 'houses' is a verb, and 'married and single soldiers' is the object. The garden path is that 'complex' is an adjective, not a noun. So the agent is 'complex' (as an adjective modifying houses? Wait, this is tricky). Let me clarify: The correct reading is 'The complex (adjective) houses (verb) married and single soldiers.' The initial misparse is that 'complex' is a noun and 'houses' is a verb, but then 'married' seems odd. Actually, the garden path is that 'houses' is first parsed as a verb, but it is actually a verb? Hmm. Better example: 'The old man the boat.' Here, 'old' is adjective, 'man' is verb, 'the boat' is object. The initial parse takes 'old man' as noun phrase, then 'the boat' as object? Actually, the g
2. How does the centering algorithm resolve the pronoun 'it' in 'John gave Mary a book. She read it.'?
The algorithm first identifies the backward-looking center of the second sentence. The forward-looking centers of the first sentence are John (subject), Mary (indirect object), book (direct object). The second sentence contains 'she' (Mary) and 'it' (book). The highest-ranked entity from the first sentence that appears in the second is Mary (rank 2), so the backward-looking center is Mary. Since 'it' is not the backward-looking center, the algorithm looks at the forward-looking centers of the second sentence: Mary (subject), book (object). 'It' is compatible with book (neuter), so it resolves to 'book'. The algorithm prefers the backward-looking center for pronouns, but 'it' cannot refer to Mary, so it picks the next highest.
3. Why does Neale reject the view that definite descriptions are referential, as argued by Donnellan?
Neale argues that Donnellan's referential use of descriptions (e.g., using 'the man drinking a martini' to refer to someone drinking water) is a pragmatic phenomenon, not semantic. For Neale, the semantics of 'the F is G' is always quantificational: it asserts existence and uniqueness. The referential use is a matter of speaker meaning, where the speaker uses the description to pick out an individual, but the literal meaning remains quantificational. Neale believes that a unified quantificational account is simpler and more systematic than a theory that posits two semantic types (referential and attributive). He maintains that Russell's theory can handle all uses when pragmatics is considered.
4. Compare the lexical deficits in semantic dementia to those in Alzheimer's disease.
In semantic dementia, the main problem is loss of word meanings, so patients often cannot name objects or understand what words refer to. In Alzheimer's disease, memory for recent events is lost first, but word meanings may be preserved longer. For example, an Alzheimer's patient may forget they had lunch but can still name a 'spoon'. In semantic dementia, the patient may know they had lunch but cannot say what they ate. Both diseases cause word-finding difficulty, but the reason differs: semantic dementia loses the concept, while Alzheimer's loses the memory of the event. Also, semantic dementia patients often have more trouble with low-frequency words like 'rhinoceros' than common ones.
5. Explain how lexical semantics is used in knowledge representation and reasoning.
In AI, knowledge representation stores facts about the world in a form computers can use. Lexical semantics helps by defining what words mean and how they relate. For example, a knowledge base might include that 'a cat is a mammal' and 'mammals have fur'. This allows a computer to reason: if something is a cat, it must have fur. Lexical semantics also handles ambiguity: 'bat' could be an animal or sports equipment, so the system needs context to choose. By linking words to concepts in an ontology (a structured set of terms), AI can answer questions like 'What animals fly?' by using the relation 'is a' and property 'can fly'. This is used in chatbots, search engines, and expert systems.
6. Compare Recanati's mental files theory with Frege's sense and reference distinction. How are they similar?
Both Recanati and Frege distinguish between the way an object is presented (sense or mode of presentation) and the object itself (reference). Frege's sense is similar to Recanati's mental file: both capture the cognitive significance of a term. For Frege, the sense of 'Hesperus' differs from that of 'Phosphorus', explaining why 'Hesperus is Phosphorus' is informative. Recanati's mental files do the same work: separate files for the same object explain cognitive difference. However, Recanati emphasizes that files are mental constructs that can be created, merged, or split, whereas Frege's senses are abstract entities. Both theories aim to account for how we think about reference.
7. How can a knowledge graph be used to answer the question 'Which countries have a population larger than India?'
First, the knowledge graph must contain entities for countries and their population numbers, with a relationship like 'hasPopulation'. The system finds the node for India and retrieves its population value. Then it searches for all country nodes where the population value is greater than that number. The graph might also have additional facts, like continent membership, to filter results. The answer would be countries like China and the United States, if their populations are larger. This reasoning requires the graph to have accurate numerical data and the ability to compare values. Knowledge graphs can combine facts from multiple sources to answer such comparative questions.
8. Explain how lexical semantics is applied in medical information extraction.
Medical information extraction uses lexical semantics to pull structured data from clinical notes. For example, from the sentence 'Patient was prescribed 500 mg of metformin twice daily', the system must recognize 'metformin' as a drug, '500 mg' as dosage, and 'twice daily' as frequency. This requires understanding that 'prescribed' indicates a medication order. Lexical resources like RxNorm map drug names to standard identifiers. Semantic relations like 'treats' link drugs to diseases. By using ontologies, the system can also infer that 'metformin' is used for diabetes. This helps in building patient records, monitoring drug safety, and supporting clinical decisions.
9. Explain how semantic dementia can help us understand the organization of semantic memory.
By studying which categories are lost first in semantic dementia, researchers learn how the brain groups knowledge. For example, if a patient loses animal names but not tool names, it suggests animals and tools are stored in different brain areas. This supports the idea that semantic memory is not one big store but has separate systems for living things and man-made objects. Also, the pattern of loss can show that sensory features (like how something looks) and functional features (like what it does) are processed differently. Such findings help build models of how the brain organizes meaning. They also guide therapies that focus on the patient's remaining strengths.
10. What is a common challenge when reasoning over a knowledge graph that contains incomplete or incorrect facts?
A major challenge is that missing or wrong facts lead to incorrect answers. For example, if the graph lacks the population of Brazil, the system might wrongly conclude Brazil is smaller than India. Also, if a fact is outdated, like an old capital city, the answer will be wrong. To handle this, AI systems use techniques like probabilistic reasoning, where each fact has a confidence score. They can also infer missing facts by learning patterns from existing data, a process called knowledge graph completion. However, such inferences are not always accurate, so the system must be careful. Regular updates and validation from trusted sources help improve reliability.
11. What is the difference between a knowledge graph and a traditional database table?
A traditional database table stores data in rows and columns, like a list of employees and their salaries. It is good for simple lookups but not for understanding relationships between different types of data. A knowledge graph stores data as a network of nodes (entities) and edges (relationships), allowing flexible connections across many domains. For example, a knowledge graph can link a person to their job, the company they work for, and the city the company is in, all in one structure. This makes it easier to answer complex queries like 'Who works in tech companies in New York?' Knowledge graphs also handle ambiguity better by using unique identifiers.
12. In ergative-absolutive languages, how are the agent of a transitive verb and the subject of an intransitive verb marked? Give an example from a language like Basque.
In ergative-absolutive languages, the agent of a transitive verb (the doer) gets a special ergative case, while the subject of an intransitive verb (the single participant) gets the absolutive case, which is also used for the patient of a transitive verb. For example, in Basque, 'Gizonak ogia jan du' (man-ERG bread-ABS ate) means 'The man ate the bread.' Here, 'gizonak' has the ergative suffix '-ak', while 'ogia' (bread) is in the absolutive. In an intransitive sentence like 'Gizona etorri da' (man-ABS came), 'gizona' is in the absolutive. So the intransitive subject and the transitive patient share the same case, while the transitive agent is different.