Transcript durmieron
Usage-based phonology Why are lines in grocery store about equal? Usage-based phonology Why are lines in grocery store about equal? Generativists say stores do this to be efficient (e.g. languages are symmetrical, try to have CV) Usage-based phonology Why are lines in grocery store about equal? Generativists say stores do this to be efficient (e.g. languages are symmetrical, try to have CV) Usage-based say it result of customers trying to be efficient, not store trying to be efficient Usage-based phonology Why are lines in grocery store about equal? Generativists say stores do this to be efficient (e.g. languages are symmetrical, try to have CV) Usage-based say it result of customers trying to be efficient, not store trying to be efficient Languages don't have motivations, or goals People have motivations and goals Usage-based phonology Usage by speakers creates structure Structure isn't innate Usage-based phonology Usage by speakers creates structure Structure isn't innate Usage-based tries to be psychologically plausible Formal theories (generative, OT) don't care about plausibility, learnability, real-time processing. Usage-based phonology Usage by speakers creates structure Structure isn't innate Usage-based tries to be psychologically plausible Formal theories (generative, OT) don't care about plausibility, learnability, real-time processing. Formal theories see patterns and generalizations as primary data Usage-based say patterns emerge as people learn language. They aren't primary evidence German In German, [x] and [ç] appear to be in complementary distribution [ç] appears before front vowels and [n, r, l] [x] appears elsewhere German In German, [x] and [ç] appear to be in complementary distribution [ç] appears before front vowels and [n, r, l] [x] appears elsewhere But it's more complex because the suffix -chen always has [ç] REGARDLESS of what precedes it German In German, [x] and [ç] appear to be in complementary distribution [ç] appears before front vowels and [n, r, l] [x] appears elsewhere But it's more complex because the suffix -chen always has [ç] REGARDLESS of what precedes it So the alternation isn't purely phonological It is morphological This suggests words are stored with pronunciation and don't get it from a rule German Remember that [ç] appears before [r, l, n] as a kind of assimilation to front sounds. German Remember that [ç] appears before [r, l, n] as a kind of assimilation to front sounds. But [r] became uvular [R], and [ç] still appears before [R] even [R] can't be causing assimilation to front of mouth The alternation isn't phonetic any more German Remember that [ç] appears before [r, l, n] as a kind of assimilation to front sounds. But [r] became uvular [R], and [ç] still appears before [R] even [R] can't be causing assimilation to front of mouth The alternation isn't phonetic any more It looks like [ç] isn't just a phonetic variant of [x] anymore German How do you model this with rules or constraints? You can't without ad hoc mechanisms Granada Spanish Final /s/ is [h] or deleted (-s marks plural) When that happens, all vowels in word open Granada Spanish This looks like it's phonetically motivated What happens to singulars ending in -s? martes, wlunes If vowels open when s > h then it's phonetic If vowels open on plural but not on singular then it's morphological Usage-based Processes start out phonetic They are analyzed by people as morphological Word frequency What is role of frequency in generative? Word frequency What is role of frequency in generative? Non-existent Word frequency What is role of frequency in generative? Non-existent But frequency influences phonetics -t -d deletion 1.There is a phonetic tendency to delete -t -d -t -d deletion 1.There is a phonetic tendency to delete -t -d 2.Each instance of a word is stored 1.with -t -d 2.or without -t -d -t -d deletion 1.There is a phonetic tendency to delete -t -d 2.Each instance of a word is stored 1.with -t -d 2.or without -t -d 3.Hi freq words used more, deletion process has more chances of applying than to lo freq 4.All words are stored so there are more instances of deletion in hi freq words -t -d deletion In generative invariant UR has deletion rule applied rule application doesn't refer to frequency words aren't stored with deleted -t -d -t -d deletion In generative invariant UR has deletion rule applied rule application doesn't refer to frequency words aren't stored with deleted -t -d In usage-based all instances of experience are stored words are stored with phonetic detail not as phonemes Don't reduction Don't reduces drastically in “don't know” “I uh oh” It also reduces a lot before, want, care, like It doesn't reduce before other words “don't cows” “don't enunciate” Hi freq collocations reduce Rules/constraints can't explain this Spanish vowel raising In some verbal forms e > i, o > u menti, mentimos, mintieron dormi, dormimos, durmieron Spanish vowel raising In some verbal forms e > i, o > u menti, mentimos, mintieron dormi, dormimos, durmieron The same rule could “explain” both of them Spanish vowel raising In some verbal forms e > i, o > u menti, mentimos, mintieron dormi, dormimos, durmieron The same rule could “explain” both of them But in nonce word study people applied e > i, never o>u pertir > pirtio, but portir > *purtio Spanish vowel raising In some verbal forms e > i, o > u menti, mentimos, mintieron dormi, dormimos, durmieron The same rule could “explain” both of them But in nonce word study people applied e > i, never o>u pertir > pirtio, but portir > *purtio Spanish vowel raising Why? Only 2 verbs have o > u (dormir, morir) 60+ verbs have e > i (pedir) Spanish vowel raising Why? Only 2 verbs have o > u (dormir, morir) 60+ verbs have e > i (pedir) Generative approaches don't encode type frequency Usage-based approaches do Universals “All languages have CV so it is universal” Rule/Constraint Approaches Details of speech not stored Words stored in phonemic form Hearing entails stripping irrelevant detail then matching to phonemic form Rule/Constraint Approaches Details of speech not stored Words stored in phonemic form Hearing entails stripping irrelevant detail then matching to phonemic form Production takes phonemic form and adds detail with rules Rule/Constraint Approaches Details of speech not stored Words stored in phonemic form Hearing entails stripping irrelevant detail then matching to phonemic form Production takes phonemic form and adds detail with rules Frequency is irrelevant Newly formed words don't affect memory Exemplar Theory Words stored with phonetic detail Same words has many stored instances Frequency is important Hi freq words recognized faster than lo freq words Exemplar Theory Can frequency of phonetic patterns influence speech processing? Abstract/formal/generative says no Exemplar models say yes -t -d releasing In speech 59% of -t -d are released, and 41% are not released -t -d releasing In speech 59% of -t -d are released, and 41% are not released Task: Press button when you hear -t or -d in a word A words pronounced with a released stop recognized faster than when it is heard with an unreleased stop Word medial flaps 96% of words like pretty, city have flap [t] is very uncommon Word medial flaps Experiment VOT manipulated to create words that varied between VOT manipulated to create words that varied between People asked to determine if word was -p or -b Word medial flaps Abstract theories say that existence of flap or -tshouldn't influence perception of initial p- or b- Word medial flaps Abstract theories say that existence of flap or -tshouldn't influence perception of initial p- or b- Exemplar theories say forms with flap should be biased toward p- because flap is more commonly experienced in this context than [t] Exemplar theories say forms with [t] weren't perceived to begin with p- as often pre[t]y / bri[t]y versus pre[ɾ]y / bri[ɾ]y Word medial flaps Results – Words that varied between were heard as p- more often – Word that varied between were not heard as p- more often Schwa perception Words like history have schwa Words manipulated to give forms with different lengths of schwa (hist[Ə]ry) to complete deletion of schwa (histry) Schwa perception Words like history has schwa Words manipulated to give forms with different lengths of schwa (history) to complete deletion of schwa (histry) People listened to forms and judged if schwa was there or not Schwa perception Words like history has schwa Words manipulated to give forms with different lengths of schwa (history) to complete deletion of schwa (histry) People listened to forms and judged if schwa was there or not Some words (history) have high deletion Some words (mammary) have low deletion Schwa perception Words like history has schwa Words manipulated to give forms with different lengths of schwa (history) to complete deletion of schwa (histry) People listened to forms and judged if schwa was there or not Some words (history) have high deletion Some words (mammary) have low deletion People heard more schwa in mammary and less in history REGARDLESS of actual length Formal vs. Usage-based Formal includes generative, OT, lexical phonology Usage-based includes connectionism, exemplar models, construction grammar Formal vs. Usage-based What must be stored and what is computed? Formal theories Regularities are computed Morphology: walked, starved Phonology: aspiration, flapping Idiosyncrasies are stored Morphology: went, worse Phonology: Samari[t]an is exceptional Formal vs. Usage-based How is language change viewed? Formal Rules/constraints are added, deleted, reranked Language changes happens in rule system Final -t deletion -t is deleted more often in irregulars (swept, kept) than in regulars (walked) Lexical Phonology explains this Irregulars have two chances for deletion to happen sweep sweep talk talk Stratum 1 Irregular derivation swept -t deletion (optional) swept --- --- swep --- --- --- talked talked talk --- Stratum 2 Regular derivation -t deletion (optional) swep Final -t deletion In OT variation is due to constraint ranking Dialect 1: FAITHC >> *C[t]# Dialect 2: *C[t]# >> FAITHC Input: swept FAITHC > swept swep * *! *C[t]# swept > swep *C[t]# FAITHC *! * Formal vs. Usage-based What must be stored and what is computed? Usage-based theories Everything is stored Regularities are stored Morphology: walked, starved Phonology: aspiration, flapping Idiosyncrasies are stored Morphology: went, worse Phonology: Samari[t]an is exceptional Formal vs. Usage-based What must be stored and what is computed? Usage-based theories Language structure arises from use Storage explains frequency effects Words and sentences are stored and connected to each other in networks Formal vs. Usage-based What must be stored and what is computed? Usage-based theories Language structure arises from use Storage explains frequency effects Words and sentences are stored and connected to each other in networks There are no innate cognitive language entities Formal vs. Usage-based What must be stored and what is computed? Usage-based theories Language structure arises from use Storage explains frequency effects Words and sentences are stored and connected to each other in networks There are no innate cognitive language entities Language is bridge between semantics and phonetics Formal vs. Usage-based Mental lexicon in usage-based theories Words stored with connections to other words based on Sound Meaning Spelling Context of usage All experienced instances of a word stored with phonetic detail Formal vs. Usage-based Mental lexicon in usage-based theories Words stored with connections to other words based on Sound Meaning Spelling Context of usage All experienced instances of a word stored with phonetic detail Words with many instances (hi freq) are recognized faster Patterns among many words are productive Formal vs. Usage-based Rule vs. schema Rule tells you what to do Add -ed to a verb Schema describes a pattern Some past tenses end in -ew Formal vs. Usage-based Rule vs. schema Rule tells you what to do Add -ed to a verb Schema describes a pattern Some past tenses end in -ew Rules apply whenever their environment exists Schema are gradient