Final Study Sheet PDF Morphology and Syntax

# Final Study Sheet ## Morphology - Study of words and their internal structure. - Where do words come from? - **Coinage:** Creating a word from scratch. - **Borrowing:** Taking a word from another language. - **Combinatorics:** Combining different meaningful elements into larger unit...

# Final Study Sheet ## Morphology - Study of words and their internal structure. - Where do words come from? - **Coinage:** Creating a word from scratch. - **Borrowing:** Taking a word from another language. - **Combinatorics:** Combining different meaningful elements into larger units. - **Lexicon:** Mental listing of words and their properties. - **Morpheme:** Smallest unit that carries meaning/grammatical function. ### Affix types - **Affix:** General term for morphemes attached to root/stem - **Prefix:** Beginning of stem. - **Suffix:** End of stem. - **Infix:** Inside stem. - **Circumfix:** Around stem. ### Affix vs. Stem vs. Root - **Base Form:** Affixes attach to. - **Stem:** Doesn't contain affixes, can't be analyzed into smaller parts. - **Root:** Contains affixes, can be analyzed into smaller parts. ### Derivational vs. Inflectional affixes - **Derivational:** Derive new basic/primary meaning. Often changes part of speech. - **Inflectional:** Indicate grammatical roles. Don't change basic meaning. Often further away from the root. ### Content vs. Function morphemes - **Content:** Referential or informational content. Has a 'core' meaning. - **Function:** Has a derivational affix. ### Bound vs. Free morphemes - **Bound:** Cannot occur in isolation. - **Free:** Can occur in isolation. ### Allomorphy - When the form of a morpheme can vary by phonological or morphological context. ### Productivity - Morphemes are productive if they freely combine with new stems. (Ex. *re-* is highly productive, *be-* is not.) ### Reduplication - *Ibu* "mother" *ibuibu* "mothers". There is also partial reduplication. *Bili* "buy" *bibili* "will buy." - Often involves intensification, repetition, plurality. - English has contrastive reduplication: "tuna salad" vs. "salad salad". ### Compounding - The formation of new words by the combination of two or more full stems. - **Root Compounds:** *psychopath, pyrophile*. - **Word Compounds:** *colorblind, easy-going, blackboard*. - Compounding can be recursive. *Failed password security question.* - There can be ambiguity: *[American history] teacher*, *[American [history teacher]]*. ### Blends - Like compounds, but parts of words are removed. - *motor + hotel = motel* - Also known as portmanteau words. ### Truncation (Clipping) - *gasoline = gas*, *delicatessen = deli*. - Assigning an already existing word to a new syntactic category. - Often called "zero derivation". - *to hit → a hit*, *empty → to empty*, *Adj → V*. ### Backformation - When a new stem is derived from a word that seems complex morphologically. - *cranberry → cran*, *editor → edit*. ## Syntax - Grammaticality and 'well-formed.' - Could you imagine a native speaker of your language uttering this sentence in regular conversation? - Notation for ungrammatical sentence: *[sentence]* - **Grammaticality**: Acceptability or felicity. - **The Principle of Compositionality:** The meaning of an expression depends on: - The meaning of the subparts (words/morphemes). - How the subparts are combined (morphology/syntax). ### Sentences have internal structure - **Constituent:** Collection of (adjacent) elements that behave like a single, unified object. ### Constituency Tests 1. **Substitution:** - 'do so' substitution for VPs. - 'one' substitution for NPs. - 'there' substitution for locations. 2. **Displacement:** (Only positive test) - Topicalization: *Tacos, I love. But burritos, I adore!* - Clefting: *I gave the book to them.* - *It was to them that I gave the book.* 3. **Coordination:** - *Conjunction: X and Y* - *Disjunction: X or Y* 4. **Fragments:** Answering questions. - Q: What was the cat doing? - A: Sleeping on the desk. - A: Sleeping on the desk yesterday. ## Lexical Category - 'Part of speech'. - Syntactic categories are similar, but at phrasal level (NP, VP, PP). ### Testing for: - **Nouns:** - Add 'the' before. - Modify with *Adj*. - Add '-s' and form possessive phrase. - Replace with pronoun/proper noun. - Subject of verb. - **NPs:** - Add '-s' and form possessive phrase. - Replace with pronoun/proper noun. - Subject of verb. - **Adj:** - Modify with degree words (very). - Comparative + superlative. - Adj phrases - place between determiner + noun. - **Verbs:** - Modify with *adv*. - Progressive -ing, post-ed. - **VPs:** - Combine w/ NPs to derive complete sentences. - 'do so' substitution. ## Lexical + syntactic categories are defined by: - Distributional properties. - Morphological properties. - Meaning (but weakly). ## Co-occurence - **Obligatory vs. Optional co-occurence:** Some morphological/syntactic items are allowed, but not required in particular environments. ### Kinds of co-occurence: - **Arguments:** 'required' constituents. - **Complement:** An element that is required or allowed to co-occur with a particular syntactic object. - Arguments can be entire sentences. - When some word *X* requires the presence of argument *A*: *X selects for A*. - **Adjuncts:** optional constituents. ### Agreement - Different expressions in the same sentence bearing morphology that expresses the same grammatical feature. - *"He faints"* vs. *"They faint"*. ### Arguments vs. Adjuncts - **Arguments:** Bare bones of sentence. - **Adjuncts:** Extra fluff. - Adjuncts ≠ optional arguments. - Adjuncts denote something whose existence is implied/entailed by the head that selects it. - *John ate (something)*. - Adjuncts are modifiers - extra, non-essential. - Do not change syntactic category of what they modify. ## Grammar consists of: 1. **Lexicon**. 2. **Phrase Structure Rules**. ### Extra: Major Syntactic Categories in English (from textbook) - **S:** Sentence; can occur in *"Sally thinks that"* - **NP:** Noun phrase; includes pronouns/proper nouns. - **N:** Noun; needs a *Det* to form *NP*. - **Det:** Determiner; ex. *the, every, this*. - **Adj:** Adjective. - **VP:** Verb phrase. - **TV:** Transitive verb; needs object. - **DTV:** Ditransitive verb; needs 2 NP complements. - **SV:** Sentential complement verb; ex. *believed, said*. - **Adv:** Adverb. - **P:** Preposition. - **PP:** Prepositional phrase; can be VP or NP adjunct. ## Semantics - **Lexical:** The meaning of words. - **Compositional:** The meaning of phrases. ### Sense vs. Reference - **Sense:** Mental representation of meaning. - **Reference:** Relationship between sense and the real world. - **The reference of an expression is the set of referents that it 'picks out.'** - Reference = extension = denotation. - **Sense:** Intension. ### Kinds of Adjectives - **Intersective vs. Relative adjectives:** - Unlike 'pure' intersectives, relatives don't obviously have a denotation independent of the nouns they modify. - Relative: vague/context-dependent/gradable. - **Intersective:** [[[AN]]] = [[A]] ∩ [[N]] - **Subsective:** [[[AN]]] = [[N]] - **Non-Subsective:** For at least one N: [[[AN]]] ⊆ [[N]] - **Privative:** For all N: [[[AN]]] ∩ [[N]] = ∅ ### Testing for adjective type - Many non-subsective and privative adjs don't occur as predicates. - **Ass test:** - *They are a...*: - ***drunk ass criminal*** (intersective) - ***big ass criminal*** (subsective) - ***alleged ass criminal*** (non-subsective) - ***former ass criminal*** (privative) ### Truth Values + Truth Conditions - **Reference of a proposition:** - **Sense of a proposition:** (the set of facts which would have to hold for a given proposition to be true) - **For predication:** Set membership. - [[[NPVP]]] = true iff [[NP]] ∈ [[VP]] ### Understanding a declarative sentence requires: - **Expression meaning:** Knowledge of a language's lexical + compositional sem. - **Utterance meaning:** Knowledge of context. ### Entailment - Relationship between propositions, such that truth of one guarantees truth of other. - **Notation:** P → Q (when P entails Q) - **Contradiction:** P → ¬ Q (¬ = 'not') equivalently: Q → ¬ P - **Logical Equivalence (Mutual Entailment):** - P → Q AND Q → P ## Pragmatics - Study of interactions between semantic meaning + context. ### Testing for entailment + contradiction - "In fact" - **For entailment:** *#I love all nectarines. In fact, it is not the case that I love yellow nectarines.* - **For contradiction:** *#I don't own any mammals. In fact, I own a chipmunk*. ### Sentence vs. Utterance - **Well-formed:** String of words. - **Utterance:** Particular context/conversation. ### An expression is deictic if its interpretation is crucially context-dependent. - Also called **indexical expression**: - Pronoun (*you*), - Location (*here*), - Time (*now*) - (In)felicitous - An utterance is (in)felicitous in the given context - NOT about politeness. ### Presupposition - Information that is assumed as part of the conversational background. - Assumptions taken for granted. ### Testing for presupposition - **'Hey, wait a minute' test (rough test):** When a speaker utters *P*, felicitous to respond: *if it is 'Hey, wait a minute, I didn't know X', then P presupposes X*. ### Family of Sentences test (better test) - If A entails B, we can test if 'A but not B' is felicitous: - *#I have a sheltie but I don't have a dog. (I don't have a sheltie but I don't have a dog)* ### Presuppositions are sticky entailments: - *I fed my rat (*# but I don't have a rat)'. - *I didn't feed my rat (*# but I don't have a rat)'. ### If P presupposes X, then P and ¬ X ## The Cooperative Principle (Grice) - Make contributions appropriate to conversation. - Conversations have goals. - Conversational maxims (Gricean maxims): - **Maxim of Quality:** Do your best to be truthful. - **Maxim of Relevance:** - **Maxim of Quantity:** No more/no less info. - **Maxim of Manner:** Be clear, avoid ambiguity, be brief, be orderly. ### Implicature - A message that an utterance indirectly communicates, but does not entail. ## Historical Linguistics (diachronic) - How languages change over time. ### Why do languages change? - **Transmission to younger generations is not exact.** - **Variation developing due to ambiguity.** - **Change is facilitated by variation within speech community and for each individual.** - **Change is neither good nor bad.** ### Types of sound change: - **Change in realization:** *u > y* (where */y/ didn't exist) - **Merger of phonemes:** *a > ø* (where */ø/ already existed) - **Development of new allophones** - **Sound change is regular if change occurs in one word, it occurs in all other words with the same environment.** (There may be caveats.) - **Morphology can change as a side effect of phonological change.** ### Semantic/Lexical Change - Changes in specific words' meaning and how that word combines with other words. ### Syntactic Change - Change in word order/grammatical functions. ### How do we reconstruct? - **Comparative w/ related languages.** - **Data from old borrowings.** - **Data from spelling in old texts and patterns in poetry.** - **Data from modern variation within languages.** - **As languages change, subgroups often end up diverging.** ## Computational Linguistics ### Chomsky Hierarchy - **Regular languages:** Languages that allow some finite and predetermined amount of memory, map onto finite-state automata. - **Context-free languages:** Can generate *anb*, equivalent to pushdown automata. ### Word token - Instance of some word. ### Word type - Unique instance of some word. ### Some notation: - **Probabilities:** *Pr("give") := Count(give)/Sum(count(w))* - **Conditional Probabilities:** *Pr("abode" | "humble") := Count(humble abode)/Count(humble x)* ### Smoothing - Assigning probability mass to things we've never seen. ### Aim of language models - To assign probabilities to language input+output, or to individual tokens. ### Bigram Model - Gets probability of each word condition based on previous word. ### Trigram Model - Gets probability of each word condition based on previous 2 words. - Etc, etc, but at some point we run out of data. ### FLOPS (floating point operations) - E.g. multiplying 2 probabilities. - Used to calculate how much computing power something needs. ### Mechanical Turk / Turkers - Tasks easy for humans, hard for computers. - *"I'm not a robot"* tasks. ### Softmax - Operation that turns preferences (numeric weights) into probabilities. ### Attention - The fact that some words matter more than others when predicting a masked item. ### Hidden State - Model's 'memory' of what it has already seen. ### Token that's intentionally masked. - Model is tasked with predicting for that token.