USC Graduate Phonology ✳︎ Fall 2019 ✳︎ Smith

The concept of the “phoneme” (a functionally significant unit in the rigidly defined pattern or configuration of sounds peculiar to a language), as distinct from that of the “sound” or “phonetic element” as such (an objectively definable entity in the articulated and perceived totality of speech), is becoming more and more familiar to linguists. The difficulty that many still seem to feel in distinguishing between the two must eventually disappear as the realization grows that no entity in human experience can be adequately defined as the mechanical sum or product of its physical properties."

– Sapir (1933)

Generative phonology

• The central premise of Generative Grammar is that linguistic knowledge consists of both a grammar and a lexicon.
• The lexicon contains all unpredictable information about words/morphemes, including their Underlying Representations (URs).
  • Hypothesis 1: every morpheme has exactly one UR.
  • Hypothesis 2: URs are built solely of sounds from a language’s phoneme inventory.
  • We’ll revise both hypotheses later, but they’re a good starting point.
• The grammar is a process or algorithm that takes URs and derives Surface Representations (SRs). The grammar captures all predictable information.
  • Hypothesis 3: if a morpheme has multiple SRs, differences between those SRs are derived via the grammar.
• The process of generating SRs from URs is a derivation.

• The grammar and lexicon together result in a set UR-SR mappings, a.k.a. input-output mappings:
  • /bræg+d/ → [brægd]
  • /hæk+d/ → [hækt]
  • /su+d/ → [sud]
  • etc.
• No matter whether you have rules or constraints:
  • A generative grammar conceives of language as a process or algorithm that generates all possible SRs from a set of URs, which can be concatenated in various ways.
  • A generative grammar has three major design goals:
    • Be parsimonious: make the algorithm as simple and general as possible.
    • Don’t undergenerate: ensure that all of the possible forms of a language are possible outputs of the grammar.
    • Don’t overgenerate: ensure that none of the impossible forms are generated by mistake.

Productivity

• An important feature: we can apply the grammar to novel words, just like kids and adults do.

• A grammatical process’s productivity in a particular language is a measure of…
  • The extent to which the process is applied to existing words.
  • The capacity of the process to apply in novel circumstances.
• The fact that speakers can apply processes to words they’ve never heard before, where memorization is impossible, supports the claim that speakers have internalized the grammatical processes of their language.
  • This argument is so important that Chomsky (1980) gave it a name: poverty of the stimulus.

Phonemes and allophones

• Every language has a limited set of phonemes.
  • A phoneme is a label for a set of sounds that are basically the same (more on this below).
  • We assume that every UR in a language consists solely of the phonemes of that language.
  • This is why URs are also called phonemic representations.
• The variants of a phoneme are called allophones.
• Allophonic variation is often structured and predictable based on phonological context.
  • English /l/
    • [l] in syllable onsets (leaf)
    • [ɫ] in syllable codas (feel)
  • American English /t/
    • [tʰ] before stressed vowels (time)
    • [ʔt] in syllable codas (cat)
    • [ɾ] between vowels (city)
    • [ʧ] before [r] (tree)
    • [t] elsewhere (sty)
• Labeled diagrams are often used to show the relationship between allophones and phonemes.

• When there are multiple allophones of a phoneme, the form of the phoneme usually matches one of the allophones.
  • One of the allophones is “basic”, matching the form of the phoneme (e.g. [t] above). This is the allophone that emerges when the grammar doesn’t change anything.
  • The other allophones are derived by the grammar.

• Notation and usage
  • Phonemes get slashes (e.g. /t/ above).
  • Allophones and surface forms get square brackets (e.g. [ɾ] above).
  • The term allophone is used only when are multiple variants of a phoneme. If there is just one surface form for a phoneme, that form isn’t an allophone.
  • If you just want to talk about something in brackets generally, use a phrase like surface form, surface realization, or phonetic form.
• Contrast and predictability
  • Phonemes are contrastive and unpredictable.
    • Two sounds are contrastive/distinctive if the difference between them is meaningful: substitution of one for the other can lead to a new word.
    • Unpredictable: “I have a new English word that rhymes with winter, does it begin with a t or l?”
  • Allophones are non-contrastive and (usually) predictable.
    • Predictable: “I have a new English word that rhymes with winter, does it begin with [l] or [ɫ]?”
    • If the allophones are unpredictable, that’s free variation.
  • Warning: the two statements above about the relationship between allophones, phonemes, contrast and predictability are simplified. They’re revisited below.
• Phonemes and allophones are psychologically real to some degree.
  • Language users often “feel” that two sounds count the same.
  • Language users often have trouble distinguishing between allophones of the same phoneme, especially when learning a new language where the difference is contrastive.
  • Writing systems and transcriptions in introductory linguistics courses are usually phonemic.

Alternations and phonotactics

• We can talk about morphemes and allomorphs just as we talk about phonemes and allophones.
• When a morpheme has multiple surface variants (allomorphs), we call the pattern an alternation.
  • Terminology examples:
    • “In English, the plural morpheme alternates between [z] and [s]”
    • “In the English plural, [z] alternates with [s]”
    • “The English plural shows the alternation [z] ~ [s]”
• Alternations are often contrasted with phonotactics, static restrictions on the distribution of sounds.
  • Examples:
    • “English words cannot end in lax vowels, except [ə], [ɔ] and [ɑ]”
    • “Lax vowels in English cannot be immediately followed by a vowel”
• Some (but not all) phonotactics result in alternations.
• Most (but not all) alternations are the result of phonotactics.

Distributions and phonemic analysis

• Phonemic analysis is a method of phonological analysis that explores the contextual distribution of segments within a language.
  • Essentially, decide how to group the sounds into phonemes.
  • In the first half of the twentieth century, phonemic analysis was the goal of pre-generative Structuralists.
    • The Structuralist view of phonemes and allophones differs from the one here: there were no underlying forms or generative grammars.
    • According to Structuralist phonemic analysis, there is always a perfectly predictable mapping between phonemes and allophones.
    • Every phone belongs to exactly one phoneme. This isn’t always obtainable.

Minimal pairs and near-minimal pairs

• A minimal pair is a pair of words that differ in meaning and exactly one sound.
• Here are some English words with various vowels.
  • We can conclude some of these vowels belong to different phonemes.
  • Walk through the argument carefully. (What assumptions does it rely on?)

• Let’s look at consonants now.
• What if we can’t find minimal pairs for two sounds? (e.g. [ʃ] and [ʒ])
  • Answer: look for near-minimal pairs.
  • Near-minimal pairs show that prediction is impossible, and the two sounds are probably contrastive.
  • This breaks down when there’s free variation. What would we expect to see if there was free variation between [ʃ] and [ʒ]?

Complementary distribution

• If two sounds occur in complementary environments (are in complementary distribution) then they may be allophones of the same phoneme.

Practicing phonemic analysis

• In English, are [l] and [d] allophones of the same phoneme or realizations of different phonemes?
• Propose a phonemic analysis for [l] and [d] in Setswana, also known as Tswana, a Niger-Congo language from Botswana with about 4.5 million speakers.

• Important point to remember: whether two sounds are different phonemes or allophones of the same phoneme depends on the language.
• Another good example is how English and Spanish divide up the sounds [ɾ], [d], [ð]:
  • English [ɾ]~[d] vs. [ð]
  • Spanish [ɾ] vs. [d]~[ð]

Limitations of the flowchart

• There are three places where flowchart thinking needs tweaking.
  • If two sounds are in complementary distribution they may be allophones of the same phoneme, but they aren’t necessarily so.
  • Contextually-limited contrast is a case where different phonemes share an allophone. In that allophone’s environment, there’s no surface contrast between the phonemes.
  • If there is a minimal pair, we can only conclude that there’s some difference between the URs of the two words. It’s not necessarily the case that the differing sound is what’s contrastive.
• Based on these, together with free variation, it becomes clearer why a perfectly predictable mapping between phonemes and allophones is unobtainable.¹

Complementary distribution, revisited

• Two sounds in complementary distribution may be allophones of the same phoneme.
• In what environments do you find [l]? [ɫ̪]?

• Are [n] and [ɫ̪] in complementary distribution? [l] and [n̪]?

• Also consider English [h] and [ŋ].
• What does this mean for our algorithm of phonemic analysis?

Positional neutralization / contextually-limited contrast

• Positional neutralization occurs when an underlying contrast between two phonemes disappears in a certain environment. In that position, the contrast is neutralized.
• This is also called contextually-limited contrast, since the contrast is limited to a particular (non-neutralizing) context.

• Another example: English nasals
  • /n/ optionally assimilates to the place of articulation of the following segment.
  • The prefix in-
    • [n] inaccessible, intolerant
    • [m] impossible
    • [ŋ] inconsolable
  • The word phone
    • [n] phone, phone number
    • [m] phone book
    • [ŋ] phone cord
  • Contrast: foam ≠ phone
  • Positional neutralization: foam book = phone book
  • But: /ŋ/ is always [ŋ].
    • [ŋ] not [n]: ringtone (≠ ri[n]tone)
    • [ŋ] not [m]: songbook (≠ so[m]book)

Minimal pairs, revisited (displaced contrast)

• Based on the data below, do [aɪ] and [ʌɪ] contrast in this dialect of English, or are they allophones of the same phoneme (if so, describe their distribution)?

• This process interacts with tapping.

• Now looking at the second set of data, would you conclude that [aɪ] and [ʌɪ] are contrastive in this dialect?
• [ɹʌɪɾɚ] and [ɹaɪɾɚ] are pronounced differently, because their underlying forms are different.
  • What is the contrast between their underlying forms?
  • In one sense, the underlying contrast has been neutralized, but in another sense, the underlying contrast can still be seen, because of the effects of another phonological process (raising).
  • So, we can say that the underlying contrast between the phonemes /t/ and /d/ has been displaced onto the allophones [aɪ] and [ʌɪ] of the phoneme /aɪ/.
• What does this mean for our algorithm of phonemic analysis?
  • If two different words' surface forms are exactly the same except for one sound, we previously concluded that means those two sounds come from different phonemes.
  • But really, all we can conclude is that there is some difference between the underlying forms.
  • Moral: it’s safest to start by looking at monomorphemic words, since any minimal contrasts there probably come from exactly the sounds that differ.

Recap

• Things to know
  • Lexicon and grammar
  • Underlying forms, surface forms, mappings
  • Phonemes and allophones
  • Morphemes and allomorphs
  • Alternations and phonotactics
  • Phonemic analysis and distributions
    • Contrastive distribution
      • Minimal pairs and near-minimal pairs
      • Displaced contrast
    • Complementary distribution
    • Contextually-limited contrast / positional neutralization
    • Free variation

• A final note: terminology can be confusing, making tables and flowcharts imperfect.

  • Remember that contrast and allophony can co-occur (e.g. positional neutralization).
  • Also, predictabiliy and allophony don’t always go together (e.g. free variation).
  • In common usage, the term allophony usually implies that the pattern isn’t positional neutralization or free variation, but allophony is technically present in both.

• In the end, our generative model (rules or constraints) will guide our phonemic analysis. The guiding principles will be “account for predictable things using rules or constraints” and “come up with an analysis that works.”

What are the goals and data?

• Repeated from above:
  • A generative grammar conceives of language as a process or algorithm that generates all possible SRs from a set of URs, which can be concatenated in various ways.
  • A generative grammar has three major design goals:
    • Be parsimonious: make the algorithm as simple and general as possible.
    • Don’t undergenerate: ensure that all of the possible forms of a language are possible outputs of the grammar.
    • Don’t overgenerate: ensure that none of the impossible forms are generated by mistake.
• What are other possible goals?
• Phonologists rarely disagree about the goals, but often disagree on the specifics:
  • How do we decide what’s possible/impossible in a language? This is especially relevant for exceptions and gaps.
  • What counts as data? What about judgments? Corpus data? Language games?
  • What does it mean for an algorithm to be simple or general? Is it fewer symbols? Shorter running time? Better performance on new data?
  • Which design goal(s) should be prioritized?
  • Should the algorithm be abstract or grounded in human cognition?
  • Relatedly: do we want the grammar to model knowledge of what’s grammatical (linguistic competence) or mimic actual human performance?
  • A clear example of the last question: do we want our model to generate humanlike speech errors?