Long-Distance in Estonian

Scott Borgeson Stanford University

1 Abstract • Compensatory lengthening (CL) has traditionally been observed to be a purely local phenomenon, with the trigger and target segments being either adjacent to one another or separated by only one syllable boundary.

2 Abstract • Compensatory lengthening (CL) has traditionally been observed to be a purely local phenomenon, with the trigger and target segments being either adjacent to one another or separated by only one syllable boundary.

• In this talk, I present evidence from Estonian showing that CL can be long-distance (LD) as well, and provide an account that allows for LDCL while explaining its crosslinguistic rarity.

2 Abstract • Compensatory lengthening (CL) has traditionally been observed to be a purely local phenomenon, with the trigger and target segments being either adjacent to one another or separated by only one syllable boundary.

• In this talk, I present evidence from Estonian showing that CL can be long-distance (LD) as well, and provide an account that allows for LDCL while explaining its crosslinguistic rarity.

• If CL takes place in any given , it will be mediated by a constraint punishing the crossing of association lines (*CROSS). This enforces pure locality.

2 Abstract • Compensatory lengthening (CL) has traditionally been observed to be a purely local phenomenon, with the trigger and target segments being either adjacent to one another or separated by only one syllable boundary.

• In this talk, I present evidence from Estonian showing that CL can be long-distance (LD) as well, and provide an account that allows for LDCL while explaining its crosslinguistic rarity.

• If CL takes place in any given language, it will be mediated by a constraint punishing the crossing of association lines (*CROSS). This enforces pure locality.

• In Estonian, however, constraints forbidding unstressed long vowels (*VV) and geminates (*GEM) outrank *CROSS. As a result, morae are prohibited from landing in intermediate positions, and must travel longer distances to find a new home.

2 Abstract • Compensatory lengthening (CL) has traditionally been observed to be a purely local phenomenon, with the trigger and target segments being either adjacent to one another or separated by only one syllable boundary.

• In this talk, I present evidence from Estonian showing that CL can be long-distance (LD) as well, and provide an account that allows for LDCL while explaining its crosslinguistic rarity.

• If CL takes place in any given language, it will be mediated by a constraint punishing the crossing of association lines (*CROSS). This enforces pure locality.

• In Estonian, however, constraints forbidding unstressed long vowels (*VV) and geminates (*GEM) outrank *CROSS. As a result, morae are prohibited from landing in intermediate positions, and must travel longer distances to find a new home.

• LDCL is rare, then, because it only exists in that first enforce CL over mora deletion, and second, that possess constraints that force LDCL over local CL.

2 Roadmap

• Section I: Define compensatory lengthening (CL) and present a cross-linguistic typology.

3 Roadmap

• Section I: Define compensatory lengthening (CL) and present a cross-linguistic typology.

• Section II: Show that long-distance compensatory lengthening (LDCL) exists in Estonian.

3 Roadmap

• Section I: Define compensatory lengthening (CL) and present a cross-linguistic typology.

• Section II: Show that long-distance compensatory lengthening (LDCL) exists in Estonian.

• Overlength (“Q3”) will be the diagnostic for (LD)CL.

3 Roadmap

• Section I: Define compensatory lengthening (CL) and present a cross-linguistic typology.

• Section II: Show that long-distance compensatory lengthening (LDCL) exists in Estonian.

• Overlength (“Q3”) will be the diagnostic for (LD)CL. • Focus will be on the partitive case as a case study.

3 Roadmap

• Section I: Define compensatory lengthening (CL) and present a cross-linguistic typology.

• Section II: Show that long-distance compensatory lengthening (LDCL) exists in Estonian.

• Overlength (“Q3”) will be the diagnostic for (LD)CL. • Focus will be on the partitive case as a case study.

• Section III: Present an account of CL (both cross-linguistically and within Estonian) that...

3 Roadmap

• Section I: Define compensatory lengthening (CL) and present a cross-linguistic typology.

• Section II: Show that long-distance compensatory lengthening (LDCL) exists in Estonian.

• Overlength (“Q3”) will be the diagnostic for (LD)CL. • Focus will be on the partitive case as a case study.

• Section III: Present an account of CL (both cross-linguistically and within Estonian) that...

• Explains the behavior of CL cross-linguistically.

3 Roadmap

• Section I: Define compensatory lengthening (CL) and present a cross-linguistic typology.

• Section II: Show that long-distance compensatory lengthening (LDCL) exists in Estonian.

• Overlength (“Q3”) will be the diagnostic for (LD)CL. • Focus will be on the partitive case as a case study.

• Section III: Present an account of CL (both cross-linguistically and within Estonian) that...

• Explains the behavior of CL cross-linguistically. • Allows for it to be long-distance (in some situations).

3 Roadmap

• Section I: Define compensatory lengthening (CL) and present a cross-linguistic typology.

• Section II: Show that long-distance compensatory lengthening (LDCL) exists in Estonian.

• Overlength (“Q3”) will be the diagnostic for (LD)CL. • Focus will be on the partitive case as a case study.

• Section III: Present an account of CL (both cross-linguistically and within Estonian) that...

• Explains the behavior of CL cross-linguistically. • Allows for it to be long-distance (in some situations). • Captures the behavior of CL in Estonian.

3 Roadmap

• Section I: Define compensatory lengthening (CL) and present a cross-linguistic typology.

• Section II: Show that long-distance compensatory lengthening (LDCL) exists in Estonian.

• Overlength (“Q3”) will be the diagnostic for (LD)CL. • Focus will be on the partitive case as a case study.

• Section III: Present an account of CL (both cross-linguistically and within Estonian) that...

• Explains the behavior of CL cross-linguistically. • Allows for it to be long-distance (in some situations). • Captures the behavior of CL in Estonian.

• Section IV: Offer concluding remarks.

3 I. What is compensatory lengthening?

4 I. Compensatory Lengthening

• Compensatory lengthening (CL) is defined in Hayes (1989) as “the lengthening of a segment triggered by the deletion or shortening of a nearby segment.”

5 I. Compensatory Lengthening

• Compensatory lengthening (CL) is defined in Hayes (1989) as “the lengthening of a segment triggered by the deletion or shortening of a nearby segment.”

• In Mora Theory (ibid.), it amounts to the movement of a mora from one segment to another.

5 I. Compensatory Lengthening

• Compensatory lengthening (CL) is defined in Hayes (1989) as “the lengthening of a segment triggered by the deletion or shortening of a nearby segment.”

• In Mora Theory (ibid.), it amounts to the movement of a mora from one segment to another.

• It involves:

5 I. Compensatory Lengthening

• Compensatory lengthening (CL) is defined in Hayes (1989) as “the lengthening of a segment triggered by the deletion or shortening of a nearby segment.”

• In Mora Theory (ibid.), it amounts to the movement of a mora from one segment to another.

• It involves:

➢ a TRIGGER segment (the segment that is shortened / deleted), and

5 I. Compensatory Lengthening

• Compensatory lengthening (CL) is defined in Hayes (1989) as “the lengthening of a segment triggered by the deletion or shortening of a nearby segment.”

• In Mora Theory (ibid.), it amounts to the movement of a mora from one segment to another.

• It involves:

➢ a TRIGGER segment (the segment that is shortened / deleted), and ➢ a TARGET segment (the segment that is lengthened as a result).

5 I. Compensatory Lengthening

• Compensatory lengthening (CL) is defined in Hayes (1989) as “the lengthening of a segment triggered by the deletion or shortening of a nearby segment.”

• In Mora Theory (ibid.), it amounts to the movement of a mora from one segment to another.

• It involves:

➢ a TRIGGER segment (the segment that is shortened / deleted), and ➢ a TARGET segment (the segment that is lengthened as a result).

• The trigger and target can be either vowels or consonants, and either can precede the other.

5 I. Compensatory Lengthening

• Compensatory lengthening (CL) is defined in Hayes (1989) as “the lengthening of a segment triggered by the deletion or shortening of a nearby segment.”

• In Mora Theory (ibid.), it amounts to the movement of a mora from one segment to another.

• It involves:

➢ a TRIGGER segment (the segment that is shortened / deleted), and ➢ a TARGET segment (the segment that is lengthened as a result).

• The trigger and target can be either vowels or consonants, and either can precede the other.

➢ (Some combinations are much more frequent than others.) ➢ (Others are essentially identical to total assimilation.)

5 Trigger I. Compensatory Lengthening: Typology Target (from Gess 2011)

• V-trigger, V-target

› L ← R /nivem/ → [neːf] “snow” (Friulian; Hualde 1990)

› L → R /kasapan/ → [ksaːpan] “sand” (Macuxi; Carson 1981)

• V-trigger, C-target

› L ← R /luto-en/ → [lutːwen] “cook-goal focus” (Ilokano; Hayes 1989)

› L → R /li-kubo/ → [kːubo] “path” (LuGanda; Clements 1986)

6 Trigger I. Compensatory Lengthening: Typology Target (from Gess 2011)

• C-trigger, V-target

› L ← R /kɨlni/ → [kɨːni] “to hear” (Komi Ižma; Harms 1968)

› L → R /roɣos/ → [oːɣos] “spider” (Samothraki Greek; Topintzi 2006)

• C-trigger, C-target

› L ← R /al-ʃams/ → [aʃ-ʃams] “the sun” (Arabic; example mine)

› L → R /nisʕf/ → [nəsʕsʕ] “half” (Colloq. Arabic; Lipiński 2001)

7 I. Compensatory Lengthening (cont.)

• CL most commonly proceeds from right to left, and only occurs between segments (or syllables) that are immediately adjacent to one another, as in…

/kɨlni/ → [kɨːni] /nivem/ → [neːf]

8 I. Compensatory Lengthening (cont.)

• CL most commonly proceeds from right to left, and only occurs between segments (or syllables) that are immediately adjacent to one another, as in…

/kɨlni/ → [kɨːni] /nivem/ → [neːf]

• Directionality, however, is not absolute. What about locality? Traditionally, the answer is “yes.”

“[T]here are no long distance diachronic CL sound changes or synchronic CL alternations (for example, loss of a final vowel never triggers lengthening of a vowel in the first syllable in a trisyllabic word).” (Kavitskaya 2002 : 29-30)

8 I. Compensatory Lengthening (cont.)

• CL most commonly proceeds from right to left, and only occurs between segments (or syllables) that are immediately adjacent to one another, as in…

/kɨlni/ → [kɨːni] /nivem/ → [neːf]

• Directionality, however, is not absolute. What about locality? Traditionally, the answer is “yes.”

“[T]here are no long distance diachronic CL sound changes or synchronic CL alternations (for example, loss of a final vowel never triggers lengthening of a vowel in the first syllable in a trisyllabic word).” (Kavitskaya 2002 : 29-30)

• I argue, no: in Estonian, morae can travel from the third or fourth syllable of a word all the way to the first, across several syllable boundaries and even out of its original foot.

8 I. Compensatory Lengthening (cont.)

• CL most commonly proceeds from right to left, and only occurs between segments (or syllables) that are immediately adjacent to one another, as in…

/kɨlni/ → [kɨːni] /nivem/ → [neːf]

• Directionality, however, is not absolute. What about locality? Traditionally, the answer is “yes.”

“[T]here are no long distance diachronic CL sound changes or synchronic CL alternations (for example, loss of a final vowel never triggers lengthening of a vowel in the first syllable in a trisyllabic word).” (Kavitskaya 2002 : 29-30)

• I argue, no: in Estonian, morae can travel from the third or fourth syllable of a word all the way to the first, across several syllable boundaries and even out of its original foot.

• Proof of this is most easily obtained in the Estonian partitive case, the case study for this talk.

8 II. LDCL in the Estonian partitive case

9 Roadmap

2.1 Present the facts of the Estonian third quantity, “Q3” – the diagnostic for (LD)CL in Estonian

10 Roadmap

2.1 Present the facts of the Estonian third quantity, “Q3” – the diagnostic for (LD)CL in Estonian

› Q3 syllables are trimoraic.

10 Roadmap

2.1 Present the facts of the Estonian third quantity, “Q3” – the diagnostic for (LD)CL in Estonian

› Q3 syllables are trimoraic. › Q3 is derived from underlying “Q2” via CL.

10 Roadmap

2.1 Present the facts of the Estonian third quantity, “Q3” – the diagnostic for (LD)CL in Estonian

› Q3 syllables are trimoraic. › Q3 is derived from underlying “Q2” via CL.

2.2 Present the facts of the Estonian partitive case.

10 Roadmap

2.1 Present the facts of the Estonian third quantity, “Q3” – the diagnostic for (LD)CL in Estonian

› Q3 syllables are trimoraic. › Q3 is derived from underlying “Q2” via CL.

2.2 Present the facts of the Estonian partitive case.

› Can surface as [-tt, -t, -∅], or [-∅] + lengthening to Q3.

10 Roadmap

2.1 Present the facts of the Estonian third quantity, “Q3” – the diagnostic for (LD)CL in Estonian

› Q3 syllables are trimoraic. › Q3 is derived from underlying “Q2” via CL.

2.2 Present the facts of the Estonian partitive case.

› Can surface as [-tt, -t, -∅], or [-∅] + lengthening to Q3. › These realizations depend purely on phonological shape of stem.

10 Roadmap

2.1 Present the facts of the Estonian third quantity, “Q3” – the diagnostic for (LD)CL in Estonian

› Q3 syllables are trimoraic. › Q3 is derived from underlying “Q2” via CL.

2.2 Present the facts of the Estonian partitive case.

› Can surface as [-tt, -t, -∅], or [-∅] + lengthening to Q3. › These realizations depend purely on phonological shape of stem.

2.3 Present an analysis of the Estonian partitive case as pure .

10 Roadmap

2.1 Present the facts of the Estonian third quantity, “Q3” – the diagnostic for (LD)CL in Estonian

› Q3 syllables are trimoraic. › Q3 is derived from underlying “Q2” via CL.

2.2 Present the facts of the Estonian partitive case.

› Can surface as [-tt, -t, -∅], or [-∅] + lengthening to Q3. › These realizations depend purely on phonological shape of stem.

2.3 Present an analysis of the Estonian partitive case as pure phonology.

› Underlyingly a suffix, /-tɑ/, which is the same for all stems.

10 Roadmap

2.1 Present the facts of the Estonian third quantity, “Q3” – the diagnostic for (LD)CL in Estonian

› Q3 syllables are trimoraic. › Q3 is derived from underlying “Q2” via CL.

2.2 Present the facts of the Estonian partitive case.

› Can surface as [-tt, -t, -∅], or [-∅] + lengthening to Q3. › These realizations depend purely on phonological shape of stem.

2.3 Present an analysis of the Estonian partitive case as pure phonology.

› Underlyingly a suffix, /-tɑ/, which is the same for all stems. › T-deletion, syncope / apocope, and CL explain the various surface forms.

10 Roadmap

2.1 Present the facts of the Estonian third quantity, “Q3” – the diagnostic for (LD)CL in Estonian

› Q3 syllables are trimoraic. › Q3 is derived from underlying “Q2” via CL.

2.2 Present the facts of the Estonian partitive case.

› Can surface as [-tt, -t, -∅], or [-∅] + lengthening to Q3. › These realizations depend purely on phonological shape of stem.

2.3 Present an analysis of the Estonian partitive case as pure phonology.

› Underlyingly a suffix, /-tɑ/, which is the same for all stems. › T-deletion, syncope / apocope, and CL explain the various surface forms. › CL in the partitive case is long-distance in some cases.

10 Roadmap

2.1 Present the facts of the Estonian third quantity, “Q3” – the diagnostic for (LD)CL in Estonian

› Q3 syllables are trimoraic. › Q3 is derived from underlying “Q2” via CL.

2.2 Present the facts of the Estonian partitive case.

› Can surface as [-tt, -t, -∅], or [-∅] + lengthening to Q3. › These realizations depend purely on phonological shape of stem.

2.3 Present an analysis of the Estonian partitive case as pure phonology.

› Underlyingly a suffix, /-tɑ/, which is the same for all stems. › T-deletion, syncope / apocope, and CL explain the various surface forms. › CL in the partitive case is long-distance in some cases.

2.4 Briefly, extend this analysis to the partitive plural.

10 2.1

Q3 in Estonian

11 2.1 Q3 in Estonian

• Estonian is well-known for its three-way length distinction in vowels and consonants.

vinɑ “vodka.PART” (Q1) linɑ “sheets.NOM” (Q1) viːnɑ “vodka.GEN” (Q2) linːɑ “city.GEN” (Q2) viːːnɑ “smoke.NOM” (Q3) linːːɑ “city.PART” (Q3)

12 2.1 Q3 in Estonian

• Estonian is well-known for its three-way length distinction in vowels and consonants.

vinɑ “vodka.PART” (Q1) linɑ “sheets.NOM” (Q1) viːnɑ “vodka.GEN” (Q2) linːɑ “city.GEN” (Q2) viːːnɑ “smoke.NOM” (Q3) linːːɑ “city.PART” (Q3)

• Q3 exists only in primary-stressed syllables.

12 2.1 Q3 in Estonian

• Estonian is well-known for its three-way length distinction in vowels and consonants.

vinɑ “vodka.PART” (Q1) linɑ “sheets.NOM” (Q1) viːnɑ “vodka.GEN” (Q2) linːɑ “city.GEN” (Q2) viːːnɑ “smoke.NOM” (Q3) linːːɑ “city.PART” (Q3)

• Q3 exists only in primary-stressed syllables.

• For almost all native words, and many loanwords, this means the initial syllable. (Loanwords that do not have initial stress can often be treated as compounds of two words, the second of which does have initial stress.)

12 2.1 Q3 in Estonian

• Estonian is well-known for its three-way length distinction in vowels and consonants.

vinɑ “vodka.PART” (Q1) linɑ “sheets.NOM” (Q1) viːnɑ “vodka.GEN” (Q2) linːɑ “city.GEN” (Q2) viːːnɑ “smoke.NOM” (Q3) linːːɑ “city.PART” (Q3)

• Q3 exists only in primary-stressed syllables.

• For almost all native words, and many loanwords, this means the initial syllable. (Loanwords that do not have initial stress can often be treated as compounds of two words, the second of which does have initial stress.)

• The nature of Q3 syllables is somewhat controversial, but I argue that their defining property is trimoraicity (Hayes 1989, contra Ehala 2003, Bye 1997, and Prillop 2013).

12 Trigger 2.1 Q3 = trimoraic syllables Target (from Gess 2011)

• There are two reasons to think this.

47 Trigger 2.1 Q3 = trimoraic syllables Target (from Gess 2011)

• There are two reasons to think this.

1) If short vowels are monomoraic, and long vowels are bimoraic, then by analogy, overlong vowels must be trimoraic.

48 Trigger 2.1 Q3 = trimoraic syllables Target (from Gess 2011)

• There are two reasons to think this.

1) If short vowels are monomoraic, and long vowels are bimoraic, then by analogy, overlong vowels must be trimoraic.

2) Q3 syllables almost* always come about when an underlyingly bimoraic syllable is lengthened via CL, following the deletion of a short vowel elsewhere in the word (originally noted in Veske 1879, Collinder 1929).

*The only exception appears to be monosyllables, which are always Q3 (Prince 1980) , e.g. [mɑːː] “land.NOM.”

49 Trigger 2.1 Q3 = trimoraic syllables Target (from Gess 2011)

• There are two reasons to think this.

1) If short vowels are monomoraic, and long vowels are bimoraic, then by analogy, overlong vowels must be trimoraic.

2) Q3 syllables almost* always come about when an underlyingly bimoraic syllable is lengthened via CL, following the deletion of a short vowel elsewhere in the word (originally noted in Veske 1879, Collinder 1929).

/kuːmɑ/ → [kuːːm] “hot.NOM” cf. [kuːmɑ] (GEN) /kotːi/ → [kotːː] “bag.NOM” cf. [kotːi] (GEN) /sepːɑ/ → [sepːː] “smith.NOM” cf. [sep:ɑ] (GEN)

*The only exception appears to be monosyllables, which are always Q3 (Prince 1980) , e.g. [mɑːː] “land.NOM.”

50 Trigger 2.1 Q3 = trimoraic syllables Target (from Gess 2011)

• There are two reasons to think this.

1) If short vowels are monomoraic, and long vowels are bimoraic, then by analogy, overlong vowels must be trimoraic.

2) Q3 syllables almost* always come about when an underlyingly bimoraic syllable is lengthened via CL, following the deletion of a short vowel elsewhere in the word (originally noted in Veske 1879, Collinder 1929).

/kuːmɑ/ → [kuːːm] “hot.NOM” cf. [kuːmɑ] (GEN) /kotːi/ → [kotːː] “bag.NOM” cf. [kotːi] (GEN) /sepːɑ/ → [sepːː] “smith.NOM” cf. [sep:ɑ] (GEN)

• If the above vowels are bimoraic underlyingly, and gain a mora via CL, then their end state can only be trimoraic. In other words, Q3 syllables should be modelled as follows.

*The only exception appears to be monosyllables, which are always Q3 (Prince 1980) , e.g. [mɑːː] “land.NOM.”

51 2.1 Q3 = trimoraic syllables

Q1 Q2 Q3 (vowels) (vowels)

14 2.1 Q3 = trimoraic syllables

Q1 Q2 Q3 (consonants) (consonants)

15 2.1 Q3 = monosyllabic foot?

(Note: a secondary property of Q3 syllables might very well be that they are monosyllabic feet (Prince 1980). This explains why monosyllables are always Q3 in Estonian, e.g. …

16 2.1 Q3 = monosyllabic foot?

(Note: a secondary property of Q3 syllables might very well be that they are monosyllabic feet (Prince 1980). This explains why monosyllables are always Q3 in Estonian, e.g. …

[mɑːː] “land.NOM” *[mɑː] impossible [luːː] “bone.NOM” *[luː] impossible [soːː] “bog.NOM” *[soː] impossible

16 2.1 Q3 = monosyllabic foot?

(Note: a secondary property of Q3 syllables might very well be that they are monosyllabic feet (Prince 1980). This explains why monosyllables are always Q3 in Estonian, e.g. …

[mɑːː] “land.NOM” *[mɑː] impossible [luːː] “bone.NOM” *[luː] impossible [soːː] “bog.NOM” *[soː] impossible

Under this view, the contrast between Q2 and Q3 would be as follows.

[liːnɑ] “city.GEN” (Q2) = (lin.nɑ) (disyllabic foot) [liːːnɑ] “city.PART” (Q3) = (lin).nɑ (monosyllabic foot + extrametrical syllable)

16 2.1 Q3 = monosyllabic foot?

(Note: a secondary property of Q3 syllables might very well be that they are monosyllabic feet (Prince 1980). This explains why monosyllables are always Q3 in Estonian, e.g. …

[mɑːː] “land.NOM” *[mɑː] impossible [luːː] “bone.NOM” *[luː] impossible [soːː] “bog.NOM” *[soː] impossible

Under this view, the contrast between Q2 and Q3 would be as follows.

[liːnɑ] “city.GEN” (Q2) = (lin.nɑ) (disyllabic foot) [liːːnɑ] “city.PART” (Q3) = (lin).nɑ (monosyllabic foot + extrametrical syllable)

The two analyses are in no way incompatible. In fact, trimoraicity might be the ultimate cause of foot exhaustivity (Bye 1997, Prillop 2013). I leave this discussion aside for now—in either case, Q3 syllables must also be said to be trimoraic.)

16 2.2

The Estonian partitive case Surface Facts

17 2.2 The Estonian partitive case

18 2.2 The Estonian partitive case

• The partitive case of nouns will form the case study for this talk. It is used with…

18 2.2 The Estonian partitive case

• The partitive case of nouns will form the case study for this talk. It is used with…

• …the direct objects of negated, atelic, or imperfective verbs.

18 2.2 The Estonian partitive case

• The partitive case of nouns will form the case study for this talk. It is used with…

• …the direct objects of negated, atelic, or imperfective verbs. • …the nominal complements of numerals above one (kaks aastat “two years”).

18 2.2 The Estonian partitive case

• The partitive case of nouns will form the case study for this talk. It is used with…

• …the direct objects of negated, atelic, or imperfective verbs. • …the nominal complements of numerals above one (kaks aastat “two years”). • …other constructions involving measurement (tükk kooki “piece of cake”).

18 2.2 The Estonian partitive case

• The partitive case of nouns will form the case study for this talk. It is used with…

• …the direct objects of negated, atelic, or imperfective verbs. • …the nominal complements of numerals above one (kaks aastat “two years”). • …other constructions involving measurement (tükk kooki “piece of cake”).

• It has a number of different realizations (to follow) for different nouns, which…

18 2.2 The Estonian partitive case

• The partitive case of nouns will form the case study for this talk. It is used with…

• …the direct objects of negated, atelic, or imperfective verbs. • …the nominal complements of numerals above one (kaks aastat “two years”). • …other constructions involving measurement (tükk kooki “piece of cake”).

• It has a number of different realizations (to follow) for different nouns, which…

• Are entirely predictable based on underlying stem shape.

18 2.2 The Estonian partitive case

• The partitive case of nouns will form the case study for this talk. It is used with…

• …the direct objects of negated, atelic, or imperfective verbs. • …the nominal complements of numerals above one (kaks aastat “two years”). • …other constructions involving measurement (tükk kooki “piece of cake”).

• It has a number of different realizations (to follow) for different nouns, which…

• Are entirely predictable based on underlying stem shape. • Can easily be reduced to a single underlying form (/-tɑ/).

18 2.2 The Estonian partitive case

• The partitive case of nouns will form the case study for this talk. It is used with…

• …the direct objects of negated, atelic, or imperfective verbs. • …the nominal complements of numerals above one (kaks aastat “two years”). • …other constructions involving measurement (tükk kooki “piece of cake”).

• It has a number of different realizations (to follow) for different nouns, which…

• Are entirely predictable based on underlying stem shape. • Can easily be reduced to a single underlying form (/-tɑ/). • Display the full range of behavior this talk focuses on, i.e...

18 2.2 The Estonian partitive case

• The partitive case of nouns will form the case study for this talk. It is used with…

• …the direct objects of negated, atelic, or imperfective verbs. • …the nominal complements of numerals above one (kaks aastat “two years”). • …other constructions involving measurement (tükk kooki “piece of cake”).

• It has a number of different realizations (to follow) for different nouns, which…

• Are entirely predictable based on underlying stem shape. • Can easily be reduced to a single underlying form (/-tɑ/). • Display the full range of behavior this talk focuses on, i.e...

➢ Local compensatory lengthening.

18 2.2 The Estonian partitive case

• The partitive case of nouns will form the case study for this talk. It is used with…

• …the direct objects of negated, atelic, or imperfective verbs. • …the nominal complements of numerals above one (kaks aastat “two years”). • …other constructions involving measurement (tükk kooki “piece of cake”).

• It has a number of different realizations (to follow) for different nouns, which…

• Are entirely predictable based on underlying stem shape. • Can easily be reduced to a single underlying form (/-tɑ/). • Display the full range of behavior this talk focuses on, i.e...

➢ Local compensatory lengthening. ➢ Long-distance compensatory lengthening.

18 2.2 The Estonian partitive case

• The partitive case of nouns will form the case study for this talk. It is used with…

• …the direct objects of negated, atelic, or imperfective verbs. • …the nominal complements of numerals above one (kaks aastat “two years”). • …other constructions involving measurement (tükk kooki “piece of cake”).

• It has a number of different realizations (to follow) for different nouns, which…

• Are entirely predictable based on underlying stem shape. • Can easily be reduced to a single underlying form (/-tɑ/). • Display the full range of behavior this talk focuses on, i.e...

➢ Local compensatory lengthening. ➢ Long-distance compensatory lengthening. ➢ Deletion without compensatory lengthening.

18 2.2 The Estonian partitive case

• The Estonian partitive singular can surface as any of four different forms, as follows.

Realization UR PART.SG Gloss

19 2.2 The Estonian partitive case

• The Estonian partitive singular can surface as any of four different forms, as follows.

Realization UR PART.SG Gloss [-tt] /kɑhuri/ → [kɑhuri-tt] “cannon” cf. GEN [kahur]

19 2.2 The Estonian partitive case

• The Estonian partitive singular can surface as any of four different forms, as follows.

Realization UR PART.SG Gloss [-tt] /kɑhuri/ → [kɑhuri-tt] “cannon” cf. GEN [kahur] [-t] /helmes/ → [helmes-t] “bead” cf. NOM [helmes]

19 2.2 The Estonian partitive case

• The Estonian partitive singular can surface as any of four different forms, as follows.

Realization UR PART.SG Gloss [-tt] /kɑhuri/ → [kɑhuri-tt] “cannon” cf. GEN [kahur] [-t] /helmes/ → [helmes-t] “bead” cf. NOM [helmes] [-∅] + Q3 /kɑːlu/ → [kɑːːlu-∅] “weight” cf. GEN [kɑːlu]

19 2.2 The Estonian partitive case

• The Estonian partitive singular can surface as any of four different forms, as follows.

Realization UR PART.SG Gloss [-tt] /kɑhuri/ → [kɑhuri-tt] “cannon” cf. GEN [kahur] [-t] /helmes/ → [helmes-t] “bead” cf. NOM [helmes] [-∅] + Q3 /kɑːlu/ → [kɑːːlu-∅] “weight” cf. GEN [kɑːlu] [-∅] /emɑ/ → [emɑ-∅] “mother” cf. GEN [ema]

19 2.2 The Estonian partitive case

• The Estonian partitive singular can surface as any of four different forms, as follows.

Realization UR PART.SG Gloss [-tt] /kɑhuri/ → [kɑhuri-tt] “cannon” cf. GEN [kahur] [-t] /helmes/ → [helmes-t] “bead” cf. NOM [helmes] [-∅] + Q3 /kɑːlu/ → [kɑːːlu-∅] “weight” cf. GEN [kɑːlu] [-∅] /emɑ/ → [emɑ-∅] “mother” cf. GEN [ema]

• These surface realizations are entirely predictable based on the shape of the underlying stem, as represented by the following decision tree.

19 2.2 The Estonian partitive case

20 78 2.2 The Estonian partitive case

[-t] : ...cvc# /helmes/ → [helmes-t] “bead” /rykːis/ → [rykːis-t] “rye” /kyːnɑl/ → [kyːnɑl-t] “candle” /ʋɤːrɑs/ → [ʋɤːrɑs-t] “stranger”

If the root is consonant-final, the partitive case will always surface as [-t].

(Some stem-final ghost consonants bring about this ending, but then disappear and assimilate to the [-t] suffix, producing [-tt]. We will not consider them here.)

21 79 2.2 The Estonian partitive case

[-tt] : (c)v.cv.cv /kɑ.hu.ri/ → [kɑhuri-tt] “cannon” /pɑ.pe.ri/ → [pɑperi-tt] “paper” /he.le.tɑ/ → [heletɑ-tt] “light” /rɑː.mɑt.tu/ → [rɑːmɑtːu-tt] “book”

If the root is vowel-final and has an odd number of syllables, then the partitive will always surface as [-tt].

22 80 2.2 The Estonian partitive case

[-t] : ...cvc# /kɑː.lu/ → [kɑːːlu-∅] “weight” /kot.ti/ → [kotːːi-∅] “bag” /loː.mɑ/ → [loːːmɑ-∅] “animal” /kyl.mɑ/ → [kylːmɑ-∅] “cold”

If the root is vowel-final, has an even number of syllables, and begins with a heavy syllable, then the partitive case will always surface as [-∅], with the first syllable becoming Q3.

23 81 2.2 The Estonian partitive case

[-t] : ...cvc# /e.mɑ/ → [emɑ-∅] “mother” /i.sɑ/ → [isɑ-∅] “father” /se.mi.nɑ.ri/ → [seminɑri-∅] “seminar” /mɑ.rɑ.to.ni/ → [mɑrɑtoni-∅] “marathon”

If the root is vowel-final, has an even number of syllables, and begins with a heavy syllable, then the partitive case will surface as [-∅] (which is identical to the genitive sg.).

24 2.2 The Estonian partitive case

• These regularities suggest a phonological account, not a morphological one.

25 2.2 The Estonian partitive case

• These regularities suggest a phonological account, not a morphological one.

➢ In other words, there are no declensions in the sense of Blevins (2008) – declensions cannot satisfyingly explain the correlations noted here, and must simply state (somewhat arbitrarily) that certain stem shapes are associated exclusively with certain declensions.

25 2.2 The Estonian partitive case

• These regularities suggest a phonological account, not a morphological one.

➢ In other words, there are no declensions in the sense of Blevins (2008) – declensions cannot satisfyingly explain the correlations noted here, and must simply state (somewhat arbitrarily) that certain stem shapes are associated exclusively with certain declensions.

• As I will now show, the partitive case can be reduced to a single suffix (/tɑ/), plus phonology.

25 2.3

The Estonian partitive case Analysis

26 2.3 The analysis

1) The partitive case is a single suffix, underlyingly /-tɑ/, that applies to all words equally.

27 2.3 The analysis

1) The partitive case is a single suffix, underlyingly /-tɑ/, that applies to all words equally.

2) The /t/ of this suffix is deleted if it is both extrametrical and intervocalic.

27 2.3 The analysis

1) The partitive case is a single suffix, underlyingly /-tɑ/, that applies to all words equally.

2) The /t/ of this suffix is deleted if it is both extrametrical and intervocalic.

➢ This occurs elsewhere, e.g. /hele-ta/ “light-ADJ.NOM” → (he.le).ta → [hele]

27 2.3 The analysis

1) The partitive case is a single suffix, underlyingly /-tɑ/, that applies to all words equally.

2) The /t/ of this suffix is deleted if it is both extrametrical and intervocalic.

➢ This occurs elsewhere, e.g. /hele-ta/ “light-ADJ.NOM” → (he.le).ta → [hele] ➢ This also occurs in Finnish (Keyser & Kiparsky 1984, Anttila 2012, and others).

27 2.3 The analysis

1) The partitive case is a single suffix, underlyingly /-tɑ/, that applies to all words equally.

2) The /t/ of this suffix is deleted if it is both extrametrical and intervocalic.

➢ This occurs elsewhere, e.g. /hele-ta/ “light-ADJ.NOM” → (he.le).ta → [hele] ➢ This also occurs in Finnish (Keyser & Kiparsky 1984, Anttila 2012, and others).

3) The same apocope / syncope processes we saw earlier will delete the vowel, /ɑ/.

27 2.3 The analysis

1) The partitive case is a single suffix, underlyingly /-tɑ/, that applies to all words equally.

2) The /t/ of this suffix is deleted if it is both extrametrical and intervocalic.

➢ This occurs elsewhere, e.g. /hele-ta/ “light-ADJ.NOM” → (he.le).ta → [hele] ➢ This also occurs in Finnish (Keyser & Kiparsky 1984, Anttila 2012, and others).

3) The same apocope / syncope processes we saw earlier will delete the vowel, /ɑ/.

4) This triggers compensatory lengthening, which lengthens [-t] to [-tt] after vowels, and Q2 syllables to Q3 – the latter will be shown to be long-distance. The behavior of CL will be summarized towards the end of this section.

27 2.3 The analysis

1) The partitive case is a single suffix, underlyingly /-tɑ/, that applies to all words equally.

2) The /t/ of this suffix is deleted if it is both extrametrical and intervocalic.

➢ This occurs elsewhere, e.g. /hele-ta/ “light-ADJ.NOM” → (he.le).ta → [hele] ➢ This also occurs in Finnish (Keyser & Kiparsky 1984, Anttila 2012, and others).

3) The same apocope / syncope processes we saw earlier will delete the vowel, /ɑ/.

4) This triggers compensatory lengthening, which lengthens [-t] to [-tt] after vowels, and Q2 syllables to Q3 – the latter will be shown to be long-distance. The behavior of CL will be summarized towards the end of this section.

We will now see how this explains all realizations of the partitive case, starting with [-tt].

27 2.3 Derivation of partitive [-tt]

Parse T-deletion V-del. + CL SR

/kɑhuri-tɑ/

UR: odd number of syllables, ends in –V#

28 2.3 Derivation of partitive [-tt]

Parse T-deletion V-del. + CL SR

/kɑhuri-tɑ/ (kɑ.hu).(ri.tɑ)

Parsing places partitive suffix /tɑ/ within a foot

29 2.3 Derivation of partitive [-tt]

Parse T-deletion V-del. + CL SR

/kɑhuri-tɑ/ (kɑ.hu).(ri.tɑ) “”

/t/ is not deleted, because it is not extrametrical.

30 2.3 Derivation of partitive [-tt]

Parse T-deletion V-del. + CL SR

/kɑhuri-tɑ/ (kɑ.hu).(ri.tɑ) “” kahuritt∅

Word-final /ɑ/ deletes. CL targets the preceding /t/, lengthening it to [tt].

31 2.3 Derivation of partitive [-tt]

Parse T-deletion V-del. + CL SR

/kɑhuri-tɑ/ (kɑ.hu).(ri.tɑ) “” kahuritt∅ [kahuritt]

Output: [-tt]

32 2.3 Derivation of partitive [-tt] (UR)

33 2.3 Derivation of partitive [-tt] (apocope, CL of /t/)

34 2.3 Derivation of partitive [-tt] (SR)

35 2.3 Derivation of partitive [-t]

Parse T-deletion V-del. + CL SR

/kɑhuri-tɑ/ (kɑ.hu).(ri.tɑ) “” kahuritt∅ [kahuritt]

/helmes-tɑ/

UR: ends in –C#

36 2.3 Derivation of partitive [-t]

Parse T-deletion V-del. + CL SR

/kɑhuri-tɑ/ (kɑ.hu).(ri.tɑ) “” kahuritt∅ [kahuritt]

/helmes-tɑ/ (hel.mes).tɑ

Parsing places partitive suffix outside of a foot.

37 2.3 Derivation of partitive [-t]

Parse T-deletion V-del. + CL SR

/kɑhuri-tɑ/ (kɑ.hu).(ri.tɑ) “” kahuritt∅ [kahuritt]

/helmes-tɑ/ (hel.mes).tɑ “”

The /t/ does not delete, because it is not intervocalic.

38 2.3 Derivation of partitive [-t]

Parse T-deletion V-del. + CL SR

/kɑhuri-tɑ/ (kɑ.hu).(ri.tɑ) “” kahuritt∅ [kahuritt]

/helmes-tɑ/ (hel.mes).tɑ “” helmest∅

The following /ɑ/ does delete, and the preceding coda /s/ absorbs its mora.

39 2.3 Derivation of partitive [-t]

Parse T-deletion V-del. + CL SR

/kɑhuri-tɑ/ (kɑ.hu).(ri.tɑ) “” kahuritt∅ [kahuritt]

/helmes-tɑ/ (hel.mes).tɑ “” helmest∅ [helmest]

Output: [-t]

40 2.3 Derivation of partitive [-t] (UR)

41 2.3 Derivation of partitive [-t] (apocope, CL)

42 2.3 Derivation of partitive [-t] (SR)

43 2.3 Derivation of partitive [-∅] + Q3

Parse T-deletion V-del. + CL SR

/kɑhuri-tɑ/ (kɑ.hu).(ri.tɑ) “” kahuritt∅ [kahuritt]

/helmes-tɑ/ (hel.mes).tɑ “” helmest∅ [helmest]

/kɑːlu-tɑ/

UR: ends in –V#, even number of syllables, first syllable heavy.

44 2.3 Derivation of partitive [-∅] + Q3

Parse T-deletion V-del. + CL SR

/kɑhuri-tɑ/ (kɑ.hu).(ri.tɑ) “” kahuritt∅ [kahuritt]

/helmes-tɑ/ (hel.mes).tɑ “” helmest∅ [helmest]

/kɑːlu-tɑ/ (kɑː.lu).tɑ

Partitive suffix is extrametrical

45 2.3 Derivation of partitive [-∅] + Q3

Parse T-deletion V-del. + CL SR

/kɑhuri-tɑ/ (kɑ.hu).(ri.tɑ) “” kahuritt∅ [kahuritt]

/helmes-tɑ/ (hel.mes).tɑ “” helmest∅ [helmest]

/kɑːlu-tɑ/ (kɑː.lu).tɑ kɑːlu∅ɑ

/t/ is both intervocalic and extrametrical, so it deletes.

46 2.3 Derivation of partitive [-∅] + Q3

Parse T-deletion V-del. + CL SR

/kɑhuri-tɑ/ (kɑ.hu).(ri.tɑ) “” kahuritt∅ [kahuritt]

/helmes-tɑ/ (hel.mes).tɑ “” helmest∅ [helmest]

/kɑːlu-tɑ/ (kɑː.lu).tɑ kɑːlu∅ɑ kɑːːlu∅

The following /ɑ/ then deletes. CL lengthens the first syllable, producing Q3.

CL cannot lengthen any other segments; this would either lengthen a short vowel to long (*kɑːluː) or a short consonant to a geminate, (*kɑːlːu), neither of which is permitted. It must therefore be long-distance. (This will be explored in greated detail below.) 47 2.3 Derivation of partitive [-∅] + Q3

Parse T-deletion V-del. + CL SR

/kɑhuri-tɑ/ (kɑ.hu).(ri.tɑ) “” kahuritt∅ [kahuritt]

/helmes-tɑ/ (hel.mes).tɑ “” helmest∅ [helmest]

/kɑːlu-tɑ/ (kɑː.lu).tɑ kɑːlu∅ɑ kɑːːlu∅ [kɑːːlu]

Output: [-∅] + Q3

48 2.3 Derivation of partitive [-∅] + Q3 (UR)

49 2.3 Derivation of partitive [-∅] + Q3 (t-deletion, apocope, CL)

50 2.3 Derivation of partitive [-∅] + Q3 (t-deletion, apocope, CL)

↑ NOTICE: the mora travels across two syllable boundaries. This is our first instance of long-distance CL.

51 2.3 Derivation of partitive [-∅] + Q3 (SR)

52 2.3 Derivation of partitive [-∅] + Q3 (SR)

↑ Q3 syllable

53 2.3 Derivation of partitive [-∅]

Parse T-deletion V-del. + CL SR

/kɑhuri-tɑ/ (kɑ.hu).(ri.tɑ) “” kahuritt∅ [kahuritt]

/helmes-tɑ/ (hel.mes).tɑ “” helmest∅ [helmest]

/kɑːlu-tɑ/ (kɑː.lu).tɑ kɑːlu∅ɑ kɑːːlu∅ [kɑːːlu]

/emɑ-tɑ/

UR: ends in –V#, even number of syllables, first syllable light.

54 2.3 Derivation of partitive [-∅]

Parse T-deletion V-del. + CL SR

/kɑhuri-tɑ/ (kɑ.hu).(ri.tɑ) “” kahuritt∅ [kahuritt]

/helmes-tɑ/ (hel.mes).tɑ “” helmest∅ [helmest]

/kɑːlu-tɑ/ (kɑː.lu).tɑ kɑːlu∅ɑ kɑːːlu∅ [kɑːːlu]

/emɑ-tɑ/ (e.mɑ).tɑ

Parsing places partitive suffix outside of a foot

55 2.3 Derivation of partitive [-∅]

Parse T-deletion V-del. + CL SR

/kɑhuri-tɑ/ (kɑ.hu).(ri.tɑ) “” kahuritt∅ [kahuritt]

/helmes-tɑ/ (hel.mes).tɑ “” helmest∅ [helmest]

/kɑːlu-tɑ/ (kɑː.lu).tɑ kɑːlu∅ɑ kɑːːlu∅ [kɑːːlu]

/emɑ-tɑ/ (e.mɑ).tɑ emɑ∅ɑ

/t/ is both intervocalic and extrametrical, and therefore deletes.

56 2.3 Derivation of partitive [-∅]

Parse T-deletion V-del. + CL SR

/kɑhuri-tɑ/ (kɑ.hu).(ri.tɑ) “” kahuritt∅ [kahuritt]

/helmes-tɑ/ (hel.mes).tɑ “” helmest∅ [helmest]

/kɑːlu-tɑ/ (kɑː.lu).tɑ kɑːlu∅ɑ kɑːːlu∅ [kɑːːlu]

/emɑ-tɑ/ (e.mɑ).tɑ emɑ∅ɑ emɑ∅

The following /ɑ/ then deletes.

CL does not take place—it cannot lenghten vowels from short to long (*emɑː, *eːmɑ), nor lengthen consonants to gemiantes (*emːɑ).

57 2.3 Derivation of partitive [-∅]

Parse T-deletion V-del. + CL SR

/kɑhuri-tɑ/ (kɑ.hu).(ri.tɑ) “” kahuritt∅ [kahuritt]

/helmes-tɑ/ (hel.mes).tɑ “” helmest∅ [helmest]

/kɑːlu-tɑ/ (kɑː.lu).tɑ kɑːlu∅ɑ kɑːːlu∅ [kɑːːlu]

/emɑ-tɑ/ (e.mɑ).tɑ emɑ∅ɑ emɑ∅ [emɑ]

Output: [-∅]

58 2.3 Derivation of partitive [-∅] (UR)

59 2.3 Derivation of partitive [-∅] (T-deletion, apocope)

60 2.3 Derivation of partitive [-∅] (SR)

61 2.3 Partitive-case derivation in toto

Parse T-deletion V-del. + CL SR

/kɑhuri-tɑ/ (kɑ.hu).(ri.tɑ) “” kahuritt∅ [kahuritt]

/helmes-tɑ/ (hel.mes).tɑ “” helmest∅ [helmest]

/kɑːlu-tɑ/ (kɑː.lu).tɑ kɑːlu∅ɑ kɑːːlu∅ [kɑːːlu]

/emɑ-tɑ/ (e.mɑ).tɑ emɑ∅ɑ emɑ∅ [emɑ]

62 2.3 Review of the analysis

• This analysis of the partitive case accomplishes the following.

63 2.3 Review of the analysis

• This analysis of the partitive case accomplishes the following.

1. It explains the correlation between stem shape and affix realization.

63 2.3 Review of the analysis

• This analysis of the partitive case accomplishes the following.

1. It explains the correlation between stem shape and affix realization.

2. It explains the [∅ : t : tt] continuum.

63 2.3 Review of the analysis

• This analysis of the partitive case accomplishes the following.

1. It explains the correlation between stem shape and affix realization.

2. It explains the [∅ : t : tt] continuum.

The UR contains /t/.

63 2.3 Review of the analysis

• This analysis of the partitive case accomplishes the following.

1. It explains the correlation between stem shape and affix realization.

2. It explains the [∅ : t : tt] continuum.

The UR contains /t/.

→ If no other processes apply, it remains [t].

63 2.3 Review of the analysis

• This analysis of the partitive case accomplishes the following.

1. It explains the correlation between stem shape and affix realization.

2. It explains the [∅ : t : tt] continuum.

The UR contains /t/.

→ If no other processes apply, it remains [t]. → In some words, CL lengthens it to [tt].

63 2.3 Review of the analysis

• This analysis of the partitive case accomplishes the following.

1. It explains the correlation between stem shape and affix realization.

2. It explains the [∅ : t : tt] continuum.

The UR contains /t/.

→ If no other processes apply, it remains [t]. → In some words, CL lengthens it to [tt]. → In others, T-deletion produces [∅].

63 2.3 Review of the analysis

• This analysis of the partitive case accomplishes the following.

1. It explains the correlation between stem shape and affix realization.

2. It explains the [∅ : t : tt] continuum.

The UR contains /t/.

→ If no other processes apply, it remains [t]. → In some words, CL lengthens it to [tt]. → In others, T-deletion produces [∅].

1. It only makes use of phonological processes that must be posited for Estonian anyway, i.e. syncope, apocope, and T-deletion.

63 2.3 Review of the analysis (cont.)

• It also reveals that CL in Estonian behaves as follows.

64 2.3 Review of the analysis (cont.)

• It also reveals that CL in Estonian behaves as follows.

• Most importantly, it CAN lengthen segments an unbounded distance away.

64 2.3 Review of the analysis (cont.)

• It also reveals that CL in Estonian behaves as follows.

• Most importantly, it CAN lengthen segments an unbounded distance away.

• Second, it CANNOT make new long vowels. /kɑːlu-tɑ/ ↛ *[kɑːluː], /emɑ-tɑ/ ↛ [eːmɑ]

64 2.3 Review of the analysis (cont.)

• It also reveals that CL in Estonian behaves as follows.

• Most importantly, it CAN lengthen segments an unbounded distance away.

• Second, it CANNOT make new long vowels. /kɑːlu-tɑ/ ↛ *[kɑːluː], /emɑ-tɑ/ ↛ [eːmɑ]

• Third, it CANNOT make new geminates. /emɑ-tɑ/ ↛ *[emːɑ]

64 2.3 Review of the analysis (cont.)

• It also reveals that CL in Estonian behaves as follows.

• Most importantly, it CAN lengthen segments an unbounded distance away.

• Second, it CANNOT make new long vowels. /kɑːlu-tɑ/ ↛ *[kɑːluː], /emɑ-tɑ/ ↛ [eːmɑ]

• Third, it CANNOT make new geminates. /emɑ-tɑ/ ↛ *[emːɑ]

• Nevertheless, it CAN make coda consonants moraic. /kɑhuri-tɑ/ → [kɑhuritt]

64 2.3 Review of the analysis (cont.)

• It also reveals that CL in Estonian behaves as follows.

• Most importantly, it CAN lengthen segments an unbounded distance away.

• Second, it CANNOT make new long vowels. /kɑːlu-tɑ/ ↛ *[kɑːluː], /emɑ-tɑ/ ↛ [eːmɑ]

• Third, it CANNOT make new geminates. /emɑ-tɑ/ ↛ *[emːɑ]

• Nevertheless, it CAN make coda consonants moraic. /kɑhuri-tɑ/ → [kɑhuritt]

• Finally, it CAN create trimoraic syllables. /kɑːlu-tɑ/ → [kɑːːlu]

64 2.3 Review of the analysis (cont.)

• It also reveals that CL in Estonian behaves as follows.

• Most importantly, it CAN lengthen segments an unbounded distance away.

• Second, it CANNOT make new long vowels. /kɑːlu-tɑ/ ↛ *[kɑːluː], /emɑ-tɑ/ ↛ [eːmɑ]

• Third, it CANNOT make new geminates. /emɑ-tɑ/ ↛ *[emːɑ]

• Nevertheless, it CAN make coda consonants moraic. /kɑhuri-tɑ/ → [kɑhuritt]

• Finally, it CAN create trimoraic syllables. /kɑːlu-tɑ/ → [kɑːːlu]

• We will return to these behaviors in§3. For now, let us briefly consider the partitive plural, which displays CL across even longer distances.

64 2.4

The Estonian partitive plural

65 2.4 The Estonian partitive plural

• The partitive plural consists of the plural suffix (-i) plus the partitive (-tɑ).

66 2.4 The Estonian partitive plural

• The partitive plural consists of the plural suffix (-i) plus the partitive (-tɑ).

• The plural suffix can also be seen with the locative cases:

• [rɑːmɑtːu-st] “book-ELATIVE” • [rɑːmɑtːu-i-st] “book-PL-ELATIVE.”

66 2.4 The Estonian partitive plural

• The partitive plural consists of the plural suffix (-i) plus the partitive (-tɑ).

• The plural suffix can also be seen with the locative cases:

• [rɑːmɑtːu-st] “book-ELATIVE” • [rɑːmɑtːu-i-st] “book-PL-ELATIVE.”

• The partitive plural has different surface realizatons from the partitive singular. Phonologically, however, it behaves exactly the same way, but with the addition of…

66 2.4 The Estonian partitive plural

• The partitive plural consists of the plural suffix (-i) plus the partitive (-tɑ).

• The plural suffix can also be seen with the locative cases:

• [rɑːmɑtːu-st] “book-ELATIVE” • [rɑːmɑtːu-i-st] “book-PL-ELATIVE.”

• The partitive plural has different surface realizatons from the partitive singular. Phonologically, however, it behaves exactly the same way, but with the addition of…

• Some vowel changes in vowel-final roots. (not considered here)

66 2.4 The Estonian partitive plural

• The partitive plural consists of the plural suffix (-i) plus the partitive (-tɑ).

• The plural suffix can also be seen with the locative cases:

• [rɑːmɑtːu-st] “book-ELATIVE” • [rɑːmɑtːu-i-st] “book-PL-ELATIVE.”

• The partitive plural has different surface realizatons from the partitive singular. Phonologically, however, it behaves exactly the same way, but with the addition of…

• Some vowel changes in vowel-final roots. (not considered here) • A suffix, /-sit/, that applies irregularly to /(C)V.CV/ roots. (not considered here)

66 2.4 The Estonian partitive plural

• The partitive plural consists of the plural suffix (-i) plus the partitive (-tɑ).

• The plural suffix can also be seen with the locative cases:

• [rɑːmɑtːu-st] “book-ELATIVE” • [rɑːmɑtːu-i-st] “book-PL-ELATIVE.”

• The partitive plural has different surface realizatons from the partitive singular. Phonologically, however, it behaves exactly the same way, but with the addition of…

• Some vowel changes in vowel-final roots. (not considered here) • A suffix, /-sit/, that applies irregularly to /(C)V.CV/ roots. (not considered here) • Finally, a rule that deletes /s/ between vowels if outside the first foot.

66 2.4 The Estonian partitive plural

• The partitive plural consists of the plural suffix (-i) plus the partitive (-tɑ).

• The plural suffix can also be seen with the locative cases:

• [rɑːmɑtːu-st] “book-ELATIVE” • [rɑːmɑtːu-i-st] “book-PL-ELATIVE.”

• The partitive plural has different surface realizatons from the partitive singular. Phonologically, however, it behaves exactly the same way, but with the addition of…

• Some vowel changes in vowel-final roots. (not considered here) • A suffix, /-sit/, that applies irregularly to /(C)V.CV/ roots. (not considered here) • Finally, a rule that deletes /s/ between vowels if outside the first foot.

• Here, we will only consider the behavior of consonant-final roots, as these will be shown to display LDCL that crosses three syllable boundaries. Derivations are as follows.

66 2.4 Partitive plural derivation: C-final roots

Parse S-deletion T-deletion V-del. + CL SR

/ʋɤːrɑs-i-tɑ/

/helmes-i-tɑ/

/rykːis-i-tɑ/

/kyːnɑl-i-tɑ/

URs: consonant-final root + plural /i/ + partitive /tɑ/

67 2.4 Partitive plural derivation: C-final roots

Parse S-deletion T-deletion V-del. + CL SR

/ʋɤːrɑs-i-tɑ/ (ʋɤː.rɑ).(si.tɑ)

/helmes-i-tɑ/ (hel.me).(si.tɑ)

/rykːis-i-tɑ/ (ryk.ki).(si.tɑ)

/kyːnɑl-i-tɑ/ (kyː.nɑ).(li.tɑ)

Parsing, as before.

68 2.4 Partitive plural derivation: C-final roots

Parse S-deletion T-deletion V-del. + CL SR

/ʋɤːrɑs-i-tɑ/ (ʋɤː.rɑ).(si.tɑ) (ʋɤː.rɑ).(∅i.tɑ)

/helmes-i-tɑ/ (hel.me).(si.tɑ) (hel.me).(∅i.tɑ)

/rykːis-i-tɑ/ (ryk.ki).(si.tɑ) (ryk.ke).(∅i.tɑ)

/kyːnɑl-i-tɑ/ (kyː.nɑ).(li.tɑ) “”

Intervocalic /s/ is deleted.

(the vowel change in /rykːiitɑ/ → [rykːeitɑ] is a separate fact.)

69 2.4 Partitive plural derivation: C-final roots

Parse S-deletion T-deletion V-del. + CL SR

/ʋɤːrɑs-i-tɑ/ (ʋɤː.rɑ).(si.tɑ) (ʋɤː.rɑ).(∅i.tɑ) “”

/helmes-i-tɑ/ (hel.me).(si.tɑ) (hel.me).(∅i.tɑ) “”

/rykːis-i-tɑ/ (ryk.ki).(si.tɑ) (ryk.ke).(∅i.tɑ) “”

/kyːnɑl-i-tɑ/ (kyː.nɑ).(li.tɑ) “” “”

/t/ is parsed, and is therefore protected from deletion.

70 2.4 Partitive plural derivation: C-final roots

Parse S-deletion T-deletion V-del. + CL SR

/ʋɤːrɑs-i-tɑ/ (ʋɤː.rɑ).(si.tɑ) (ʋɤː.rɑ).(∅i.tɑ) “” ʋɤːːrɑit∅

/helmes-i-tɑ/ (hel.me).(si.tɑ) (hel.me).(∅i.tɑ) “” helːmeit∅

/rykːis-i-tɑ/ (ryk.ki).(si.tɑ) (ryk.ke).(∅i.tɑ) “” rykːːeit∅

/kyːnɑl-i-tɑ/ (kyː.nɑ).(li.tɑ) “” “” kyːːn∅lɑit∅

Final /ɑ/ is deleted. CL takes place and the mora moves to the first syllable.

(It cannot land on the second syllable, as this would have produced an unstressed Q3 syllable—either [-ɑiːt], [-ɑːit], or [-ɑitː]. Recall that Estonian only allows Q3 in primary-stressed syllables.)

71 2.4 Partitive plural derivation: C-final roots

Parse S-deletion T-deletion V-del. + CL SR

/ʋɤːrɑs-i-tɑ/ (ʋɤː.rɑ).(si.tɑ) (ʋɤː.rɑ).(∅i.tɑ) “” ʋɤːːrɑit∅ ʋɤːːrɑit

/helmes-i-tɑ/ (hel.me).(si.tɑ) (hel.me).(∅i.tɑ) “” helːmeit∅ helːmeit

/rykːis/ (ryk.ki).(si.tɑ) (ryk.ke).(∅i.tɑ) “” rykːːeit∅ rykːːeit

/kyːnɑl-i-tɑ/ (kyː.nɑ).(li.tɑ) “” “” kyːːn∅lait∅ kyːːnlɑit

SR: Q3 on first syllable, partitive suffix surfaces as [-t].

72 2.4 Partitive plural derivation: UR

73 2.4 Partitive plural derivation: S-deletion

74 2.4 Partitive plural derivation: apocope + CL

75 2.4 Partitive plural derivation: apocope + CL

↑ CL takes place across two intervening syllables.

76 2.4 Partitive plural derivation: SR

77 III. An account of (LD)CL C ROSS - LINGUISTICALLY, AND IN E STONIAN

78 Roadmap

3.1 Present an account of CL (not limited to Estonian) that allows for the possibility of LDCL.

79 Roadmap

3.1 Present an account of CL (not limited to Estonian) that allows for the possibility of LDCL.

• Present the mora preservation constraints (MAX-μ) and show their inadequacy.

79 Roadmap

3.1 Present an account of CL (not limited to Estonian) that allows for the possibility of LDCL.

• Present the mora preservation constraints (MAX-μ) and show their inadequacy. • Present the locality constraint *CROSS and show its inadequacy.

79 Roadmap

3.1 Present an account of CL (not limited to Estonian) that allows for the possibility of LDCL.

• Present the mora preservation constraints (MAX-μ) and show their inadequacy. • Present the locality constraint *CROSS and show its inadequacy. • Present a constraint-based typology of CL that combines the two.

79 Roadmap

3.1 Present an account of CL (not limited to Estonian) that allows for the possibility of LDCL.

• Present the mora preservation constraints (MAX-μ) and show their inadequacy. • Present the locality constraint *CROSS and show its inadequacy. • Present a constraint-based typology of CL that combines the two.

3.2 Revisit the behavior of (LD)CL in Estonian and build an account to explain it.

79 Roadmap

3.1 Present an account of CL (not limited to Estonian) that allows for the possibility of LDCL.

• Present the mora preservation constraints (MAX-μ) and show their inadequacy. • Present the locality constraint *CROSS and show its inadequacy. • Present a constraint-based typology of CL that combines the two.

3.2 Revisit the behavior of (LD)CL in Estonian and build an account to explain it.

• Build a set of constraints that explain the behavior of CL in Estonian.

79 Roadmap

3.1 Present an account of CL (not limited to Estonian) that allows for the possibility of LDCL.

• Present the mora preservation constraints (MAX-μ) and show their inadequacy. • Present the locality constraint *CROSS and show its inadequacy. • Present a constraint-based typology of CL that combines the two.

3.2 Revisit the behavior of (LD)CL in Estonian and build an account to explain it.

• Build a set of constraints that explain the behavior of CL in Estonian. • Sort out their relative rankings, showing that…

79 Roadmap

3.1 Present an account of CL (not limited to Estonian) that allows for the possibility of LDCL.

• Present the mora preservation constraints (MAX-μ) and show their inadequacy. • Present the locality constraint *CROSS and show its inadequacy. • Present a constraint-based typology of CL that combines the two.

3.2 Revisit the behavior of (LD)CL in Estonian and build an account to explain it.

• Build a set of constraints that explain the behavior of CL in Estonian. • Sort out their relative rankings, showing that…

• *VV and *GEM prohibit CL in words displaying partitive [-∅].

79 Roadmap

3.1 Present an account of CL (not limited to Estonian) that allows for the possibility of LDCL.

• Present the mora preservation constraints (MAX-μ) and show their inadequacy. • Present the locality constraint *CROSS and show its inadequacy. • Present a constraint-based typology of CL that combines the two.

3.2 Revisit the behavior of (LD)CL in Estonian and build an account to explain it.

• Build a set of constraints that explain the behavior of CL in Estonian. • Sort out their relative rankings, showing that…

• *VV and *GEM prohibit CL in words displaying partitive [-∅]. • *VV and *GEM also prohibit local CL in words displaying partitive [-∅] + Q3.

79 Roadmap

3.1 Present an account of CL (not limited to Estonian) that allows for the possibility of LDCL.

• Present the mora preservation constraints (MAX-μ) and show their inadequacy. • Present the locality constraint *CROSS and show its inadequacy. • Present a constraint-based typology of CL that combines the two.

3.2 Revisit the behavior of (LD)CL in Estonian and build an account to explain it.

• Build a set of constraints that explain the behavior of CL in Estonian. • Sort out their relative rankings, showing that…

• *VV and *GEM prohibit CL in words displaying partitive [-∅]. • *VV and *GEM also prohibit local CL in words displaying partitive [-∅] + Q3. • *CROSS demands purely local CL if possible, explaining partitive [-t, -tt].

79 3.1

Compensatory Lengthening cross-linguistically

80 3.1 CL in Mora theory

• I will be assuming a Mora-Theoretic (Hayes 1989) account of CL. Under this theory, CL amounts to the movement of a mora (an abstract unit of phonological length) from one segment to another.

81 3.1 CL in Mora theory

• I will be assuming a Mora-Theoretic (Hayes 1989) account of CL. Under this theory, CL amounts to the movement of a mora (an abstract unit of phonological length) from one segment to another.

• Although originally couched in rules, it can easily be adapted to OT with the following constraints.

81 3.1 CL in Mora theory

• I will be assuming a Mora-Theoretic (Hayes 1989) account of CL. Under this theory, CL amounts to the movement of a mora (an abstract unit of phonological length) from one segment to another.

• Although originally couched in rules, it can easily be adapted to OT with the following constraints.

• First, some constraint(s) must delete or shorten a trigger segment. This must be language-specific, and thus will not be considered here.

81 3.1 CL in Mora theory

• I will be assuming a Mora-Theoretic (Hayes 1989) account of CL. Under this theory, CL amounts to the movement of a mora (an abstract unit of phonological length) from one segment to another.

• Although originally couched in rules, it can easily be adapted to OT with the following constraints.

• First, some constraint(s) must delete or shorten a trigger segment. This must be language-specific, and thus will not be considered here.

• Second, the mora of the trigger segment is not deleted, but merely moved. Therefore, some other constraint must be preserving it from deletion (MAX-μ).

81 3.1 CL in Mora theory

• I will be assuming a Mora-Theoretic (Hayes 1989) account of CL. Under this theory, CL amounts to the movement of a mora (an abstract unit of phonological length) from one segment to another.

• Although originally couched in rules, it can easily be adapted to OT with the following constraints.

• First, some constraint(s) must delete or shorten a trigger segment. This must be language-specific, and thus will not be considered here.

• Second, the mora of the trigger segment is not deleted, but merely moved. Therefore, some other constraint must be preserving it from deletion (MAX-μ).

• Finally, in the usual case, morae must move to immediately adjacent segments or syllables. Therefore, some constraint must guarantee this as well, punishing candidates in which the mora moves any farther than necessary (*CROSS).

81 3.1 CL in Mora theory

• I will be assuming a Mora-Theoretic (Hayes 1989) account of CL. Under this theory, CL amounts to the movement of a mora (an abstract unit of phonological length) from one segment to another.

• Although originally couched in rules, it can easily be adapted to OT with the following constraints.

• First, some constraint(s) must delete or shorten a trigger segment. This must be language-specific, and thus will not be considered here.

• Second, the mora of the trigger segment is not deleted, but merely moved. Therefore, some other constraint must be preserving it from deletion (MAX-μ).

• Finally, in the usual case, morae must move to immediately adjacent segments or syllables. Therefore, some constraint must guarantee this as well, punishing candidates in which the mora moves any farther than necessary (*CROSS).

• We will now consider these last two, beginning with MAX-μ.

81 3.1 Mora preservation constraints

• Kavitskaya (2002) suggests (but ultimately rejects) a hierarchy of constraints that punish the deletion of morae within certain phonological domains.

82 3.1 Mora preservation constraints

• Kavitskaya (2002) suggests (but ultimately rejects) a hierarchy of constraints that punish the deletion of morae within certain phonological domains.

• MAX-μ(σ) : “Every mora in the input (or an intermediate representation) has a correspondent within the same syllable in the output.” (pg. 181)

82 3.1 Mora preservation constraints

• Kavitskaya (2002) suggests (but ultimately rejects) a hierarchy of constraints that punish the deletion of morae within certain phonological domains.

• MAX-μ(σ) : “Every mora in the input (or an intermediate representation) has a correspondent within the same syllable in the output.” (pg. 181)

Rephrased slightly: “Do not delete morae from syllables.”

82 3.1 Mora preservation constraints

• Kavitskaya (2002) suggests (but ultimately rejects) a hierarchy of constraints that punish the deletion of morae within certain phonological domains.

• MAX-μ(σ) : “Every mora in the input (or an intermediate representation) has a correspondent within the same syllable in the output.” (pg. 181)

Rephrased slightly: “Do not delete morae from syllables.”

• MAX-μ(φ) : Do not delete morae from feet.

82 3.1 Mora preservation constraints

• Kavitskaya (2002) suggests (but ultimately rejects) a hierarchy of constraints that punish the deletion of morae within certain phonological domains.

• MAX-μ(σ) : “Every mora in the input (or an intermediate representation) has a correspondent within the same syllable in the output.” (pg. 181)

Rephrased slightly: “Do not delete morae from syllables.”

• MAX-μ(φ) : Do not delete morae from feet.

• MAX-μ(ω) : Do not delete morae from the word.

82 3.1 Mora preservation constraints

• Kavitskaya (2002) suggests (but ultimately rejects) a hierarchy of constraints that punish the deletion of morae within certain phonological domains.

• MAX-μ(σ) : “Every mora in the input (or an intermediate representation) has a correspondent within the same syllable in the output.” (pg. 181)

Rephrased slightly: “Do not delete morae from syllables.”

• MAX-μ(φ) : Do not delete morae from feet.

• MAX-μ(ω) : Do not delete morae from the word.

• This will capture some data, but does not capture locality per se: if a mora is moved out of its foot (violating MAX-μ(φ)), then it will not matter what syllable within its new foot the mora moves to—the two candidates will be identical. This can be seen in the following tableau.

82 “lake MAX- MAX- MAX- From Kavitskaya (2002); (dimin.)” μ(ω) μ(φ) μ(σ) tableau and prosodic hierarchies mine.

• Jers (ŭ) delete for independent reasons

☞ a. * • /eː/ raises to [iː] for independent reasons

 b. *

83 “lake MAX- MAX- MAX- From Kavitskaya (2002); (dimin.)” μ(ω) μ(φ) μ(σ) tableau and prosodic hierarchies mine.

• Jers (ŭ) delete for independent reasons

☞ a. * • /eː/ raises to [iː] for independent reasons

Unviolated because the input and output both have 4 morae.  b. *

84 “lake MAX- MAX- MAX- From Kavitskaya (2002); (dimin.)” μ(ω) μ(φ) μ(σ) tableau and prosodic hierarchies mine.

• Jers (ŭ) delete for independent reasons

☞ a. * • /eː/ raises to [iː] for independent reasons

Unviolated because the trigger syllable is completely deleted, and thus cannot be  b. * evaluated by MAX-μ(σ).

85 “lake MAX- MAX- MAX- From Kavitskaya (2002); (dimin.)” μ(ω) μ(φ) μ(σ) tableau and prosodic hierarchies mine.

• Jers (ŭ) delete for independent reasons

☞ a. * • /eː/ raises to [iː] for independent reasons

Syllable moves out of its foot, to the closest possible landing site.  b. *

86 “lake MAX- MAX- MAX- From Kavitskaya (2002); (dimin.)” μ(ω) μ(φ) μ(σ) tableau and prosodic hierarchies mine.

• Jers (ŭ) delete for independent reasons

☞ a. * • /eː/ raises to [iː] for independent reasons

Syllable moves out of its foot, but not to the closest landing site.  b. * EQUIVALENT VIOLATIONS!

87 3.1 CL in OT: locality constraints

• An additional constraint is therefore needed that enforces locality without relying only on phonological constituents. We will assume:

88 3.1 CL in OT: locality constraints

• An additional constraint is therefore needed that enforces locality without relying only on phonological constituents. We will assume:

➢ *CROSS: do not cross association lines

88 3.1 CL in OT: locality constraints

• An additional constraint is therefore needed that enforces locality without relying only on phonological constituents. We will assume: 1 violation ➢ *CROSS: do not cross association lines

88 3.1 CL in OT: locality constraints

• An additional constraint is therefore needed that enforces locality without relying only on phonological constituents. We will assume: 1 violation ➢ *CROSS: do not cross association lines

3 violations

88 3.1 CL in OT: locality constraints

• An additional constraint is therefore needed that enforces locality without relying only on phonological constituents. We will assume: 1 violation ➢ *CROSS: do not cross association lines

• If *CROSS is undominated, then CL will never take place, except between two adjacent syllables or segments (i.e. if no association lines are crossed).

3 violations

88 3.1 CL in OT: locality constraints

• An additional constraint is therefore needed that enforces locality without relying only on phonological constituents. We will assume: 1 violation ➢ *CROSS: do not cross association lines

• If *CROSS is undominated, then CL will never take place, except between two adjacent syllables or segments (i.e. if no association lines are crossed).

(For the sake of simplicity, we’ll only consider the line between a mora and its segment [see right], and not that 3 violations between a mora and its syllable—adding the latter would appear to do nothing except increase the violation count without changing the optimality of the candidates relative to one another. For now, I will leave this discussion, along with further refinement of this constraint, to future work.)

88 3.1 CL in OT: locality constraints (cont.)

• Are *CROSS and the MAX-μ constraints both necessary? Yes.

89 3.1 CL in OT: locality constraints (cont.)

• Are *CROSS and the MAX-μ constraints both necessary? Yes.

• We’ve already seen that MAX-μ constraints are not sufficient by themselves.

89 3.1 CL in OT: locality constraints (cont.)

• Are *CROSS and the MAX-μ constraints both necessary? Yes.

• We’ve already seen that MAX-μ constraints are not sufficient by themselves.

• But *CROSS isn’t, either. Consider Friulian (Hualde 1990), in which...

89 3.1 CL in OT: locality constraints (cont.)

• Are *CROSS and the MAX-μ constraints both necessary? Yes.

• We’ve already seen that MAX-μ constraints are not sufficient by themselves.

• But *CROSS isn’t, either. Consider Friulian (Hualde 1990), in which...

…morae are preserved within the foot (CL) /(ru.do)/ → [ruːt] (with final devoicing)

89 3.1 CL in OT: locality constraints (cont.)

• Are *CROSS and the MAX-μ constraints both necessary? Yes.

• We’ve already seen that MAX-μ constraints are not sufficient by themselves.

• But *CROSS isn’t, either. Consider Friulian (Hualde 1990), in which...

…morae are preserved within the foot (CL) /(ru.do)/ → [ruːt] (with final devoicing) …but not within the word, outside of feet. /(u.mi).do/ ↛ *[umiːt], *[u:mit]

89 3.1 CL in OT: locality constraints (cont.)

…morae are preserved within the foot (CL) /(ru.do)/ → [ruːt] (with final devoicing) …but not within the word, outside of feet. /(u.mi).do/ ↛ *[umiːt], *[u:mit]

• Something must rule out *[uːmit]. If the only constraint involved is *CROSS, then it must simultaneously rule out CL in *[uːmit], and fail to rule out CL in [ruːt]. This is not possible.

90 3.1 CL in OT: locality constraints (cont.)

…morae are preserved within the foot (CL) /(ru.do)/ → [ruːt] (with final devoicing) …but not within the word, outside of feet. /(u.mi).do/ ↛ *[umiːt], *[u:mit]

• Something must rule out *[uːmit]. If the only constraint involved is *CROSS, then it must simultaneously rule out CL in *[uːmit], and fail to rule out CL in [ruːt]. This is not possible.

• Fortunately, we already have the solution:

90 3.1 CL in OT: locality constraints (cont.)

…morae are preserved within the foot (CL) /(ru.do)/ → [ruːt] (with final devoicing) …but not within the word, outside of feet. /(u.mi).do/ ↛ *[umiːt], *[u:mit]

• Something must rule out *[uːmit]. If the only constraint involved is *CROSS, then it must simultaneously rule out CL in *[uːmit], and fail to rule out CL in [ruːt]. This is not possible.

• Fortunately, we already have the solution:

➢ in [ruːt], *CROSS is violated in order to preserve morae within a single foot (i.e. in order to obey the demands of MAX-μ(φ)).

90 3.1 CL in OT: locality constraints (cont.)

…morae are preserved within the foot (CL) /(ru.do)/ → [ruːt] (with final devoicing) …but not within the word, outside of feet. /(u.mi).do/ ↛ *[umiːt], *[u:mit]

• Something must rule out *[uːmit]. If the only constraint involved is *CROSS, then it must simultaneously rule out CL in *[uːmit], and fail to rule out CL in [ruːt]. This is not possible.

• Fortunately, we already have the solution:

➢ in [ruːt], *CROSS is violated in order to preserve morae within a single foot (i.e. in order to obey the demands of MAX-μ(φ)).

➢ In [umit], *CROSS cannot be violated to preserve morae across feet, i.e. within the word as a whole (in order to obey the demands of MAX-μ(ω)).

90 3.1 CL in OT: locality constraints (cont.)

…morae are preserved within the foot (CL) /(ru.do)/ → [ruːt] (with final devoicing) …but not within the word, outside of feet. /(u.mi).do/ ↛ *[umiːt], *[u:mit]

• Something must rule out *[uːmit]. If the only constraint involved is *CROSS, then it must simultaneously rule out CL in *[uːmit], and fail to rule out CL in [ruːt]. This is not possible.

• Fortunately, we already have the solution:

➢ in [ruːt], *CROSS is violated in order to preserve morae within a single foot (i.e. in order to obey the demands of MAX-μ(φ)).

➢ In [umit], *CROSS cannot be violated to preserve morae across feet, i.e. within the word as a whole (in order to obey the demands of MAX-μ(ω)).

➢ In other words, MAX-μ(φ) >> *CROSS >> MAX-μ(ω)

90 /umido/ “Humid” (MASC) MAX-μ(φ) *CROSS MAX-μ(ω)

a. [umiːt] *!

b. [uːmit] ***!

☞ c. [umit] *

*CROSS >> MAX-μ(ω)

Do not cross association lines, even if it means deleting morae from the word.

91 /rudo/ “pure” (MASC) MAX-μ(φ) *CROSS MAX-μ(ω)

a. [ruːt] *

☞ b. [rut] *!

MAX-μ(φ) >> *CROSS

...but do preserve all morae within a foot—even at the cost of crossing association lines.

92 3.1 *CROSS and MAX-μ in Czech

• Moreover, combining *CROSS with the MAX-μ constraints appears to solve the Czech problem: *CROSS now differentiates between the candidate with local CL (a) and the one with LDCL (b).

/jezerŭko/ “lake (dimin.)” MAX-μ(ω) *CROSS MAX-μ(φ)

☞ a. [jeziːrko] * *

b. [ji:zerko] ***! * [jezerko] c. *! * (no CL)

93 3.1 A typology of compensatory lengthening

• Let us now consider what sorts of behaviors these constraints can explain.

94 3.1 A typology of compensatory lengthening

• Let us now consider what sorts of behaviors these constraints can explain.

• First, let us assume some constraint that punishes the addition of morae to segments (call it DEP-μ). If DEP-μ is undominated, then no compensatory lengthening will take place.

94 3.1 A typology of compensatory lengthening

• Let us now consider what sorts of behaviors these constraints can explain.

• First, let us assume some constraint that punishes the addition of morae to segments (call it DEP-μ). If DEP-μ is undominated, then no compensatory lengthening will take place.

• Next, If *CROSS is undominated, and DEP-μ is dominated by any MAX-μ constraint, then CL will take place, but will never involve the crossing of association lines—that is, it will be tautosyllabic, or will only take place between two adjacent, onsetless syllables.

/CVN/ → [CVː] Latin /*kasnus/ → [kaːnus] “gray” (Hayes 1989) /CVV/ → [CVː] Colloq. Finnish /mɑkeɑ/ → [mɑkeː] “sweet”

94 3.1 A typology of compensatory lengthening

• Let us now consider what sorts of behaviors these constraints can explain.

• First, let us assume some constraint that punishes the addition of morae to segments (call it DEP-μ). If DEP-μ is undominated, then no compensatory lengthening will take place.

• Next, If *CROSS is undominated, and DEP-μ is dominated by any MAX-μ constraint, then CL will take place, but will never involve the crossing of association lines—that is, it will be tautosyllabic, or will only take place between two adjacent, onsetless syllables.

/CVN/ → [CVː] Latin /*kasnus/ → [kaːnus] “gray” (Hayes 1989) /CVV/ → [CVː] Colloq. Finnish /mɑkeɑ/ → [mɑkeː] “sweet”

(This is the most common type of CL, because it is the least restrictive in terms of rankings.)

94 3.1 A typology of compensatory lengthening (cont.)

• If MAX-μ(φ) >> *CROSS >> MAX-μ(ω), then morae may cross association lines in order to be preserved within the foot, but not within the word. CL will be tautopedal.

(CV.CV) → [CVːC] Friulian /rudo/ → [ruːt] but (CV.CV).CV → [CV.CVC] Friulian /umido/ → [umit] (no CL)

95 3.1 A typology of compensatory lengthening (cont.)

• If MAX-μ(φ) >> *CROSS >> MAX-μ(ω), then morae may cross association lines in order to be preserved within the foot, but not within the word. CL will be tautopedal.

(CV.CV) → [CVːC] Friulian /rudo/ → [ruːt] but (CV.CV).CV → [CV.CVC] Friulian /umido/ → [umit] (no CL)

• Finally, if MAX-μ(ω) >> *CROSS, then all morae must be preserved within a word—compen- satory lengthening will move morae out of one foot and into another, making it transpedal.

(CV.CV).(CV.CV) → (CV.CVː).(CV) Czech /jezerŭko/ → [jeziːrko]

95 3.1 A typology of compensatory lengthening (cont.)

• If MAX-μ(φ) >> *CROSS >> MAX-μ(ω), then morae may cross association lines in order to be preserved within the foot, but not within the word. CL will be tautopedal.

(CV.CV) → [CVːC] Friulian /rudo/ → [ruːt] but (CV.CV).CV → [CV.CVC] Friulian /umido/ → [umit] (no CL)

• Finally, if MAX-μ(ω) >> *CROSS, then all morae must be preserved within a word—compen- satory lengthening will move morae out of one foot and into another, making it transpedal.

(CV.CV).(CV.CV) → (CV.CVː).(CV) Czech /jezerŭko/ → [jeziːrko]

• At the same time, even transpedal CL must still be as local as possible—the mere presence of *CROSS in the system guarantees that any candidates with non-local CL will be harmonically bounded by candidates with local CL. What, then, allows for LDCL?

95 3.1 A typology of compensatory lengthening (cont.)

• What allows for LDCL? Something must dominate *CROSS, forcing additional violations of it.

96 3.1 A typology of compensatory lengthening (cont.)

• What allows for LDCL? Something must dominate *CROSS, forcing additional violations of it.

• In Estonian, this is *VV and *GEM: local CL would always result in forming new geminates or long vowels, which are prohibited. If LDCL can preserve the mora without doing violating these constraints, it must be employed (/kɑːlu-tɑ/ “weight-PART” → [kɑːːlu]).

96 3.1 A typology of compensatory lengthening (cont.)

• What allows for LDCL? Something must dominate *CROSS, forcing additional violations of it.

• In Estonian, this is *VV and *GEM: local CL would always result in forming new geminates or long vowels, which are prohibited. If LDCL can preserve the mora without doing violating these constraints, it must be employed (/kɑːlu-tɑ/ “weight-PART” → [kɑːːlu]).

• (If even LDCL cannot save the mora, then no CL takes place: /emɑ-tɑ/ “mother-PART” → [emɑ]).

96 3.1 A typology of compensatory lengthening (cont.)

• What allows for LDCL? Something must dominate *CROSS, forcing additional violations of it.

• In Estonian, this is *VV and *GEM: local CL would always result in forming new geminates or long vowels, which are prohibited. If LDCL can preserve the mora without doing violating these constraints, it must be employed (/kɑːlu-tɑ/ “weight-PART” → [kɑːːlu]).

• (If even LDCL cannot save the mora, then no CL takes place: /emɑ-tɑ/ “mother-PART” → [emɑ]).

• But even in a system like Estonian, which tolerates LDCL, the existence of *CROSS will always favor local CL over LDCL if the two are otherwise equal.

96 3.1 A typology of compensatory lengthening (cont.)

• What allows for LDCL? Something must dominate *CROSS, forcing additional violations of it.

• In Estonian, this is *VV and *GEM: local CL would always result in forming new geminates or long vowels, which are prohibited. If LDCL can preserve the mora without doing violating these constraints, it must be employed (/kɑːlu-tɑ/ “weight-PART” → [kɑːːlu]).

• (If even LDCL cannot save the mora, then no CL takes place: /emɑ-tɑ/ “mother-PART” → [emɑ]).

• But even in a system like Estonian, which tolerates LDCL, the existence of *CROSS will always favor local CL over LDCL if the two are otherwise equal.

• We will now consider this in more detail.

96 3.2

LDCL in Estonian

97 3.2 Behavior of LDCL in Estonian, revisited

• Recall that compensatory lengthening in Estonian...

a) CANNOT make new long vowels.

b) CANNOT make new geminates, but

c) CAN make coda consonants moraic.

d) CAN create trimoraic syllables.

e) CAN lengthen segments an unbounded distance away, and additionally,

f) MUST lengthen the nearest possible segment.

98 3.2 Behavior of LDCL in Estonian, revisited

• Recall that compensatory lengthening in Estonian...

a) CANNOT make new long vowels.

b) CANNOT make new geminates, but

c) CAN make coda consonants moraic.

d) CAN create trimoraic syllables.

e) CAN lengthen segments an unbounded distance away, and additionally,

f) MUST lengthen the nearest possible segment.

• To illustrate, consider the following:

98 3.2 CL (a) CANNOT make new long vowels

99 3.2 CL (a) CANNOT make new long vowels

100 3.2 CL (b) CANNOT make new geminates

101 228 3.2 CL (c) CAN make codas moraic

102 3.2 CL (d,e) CAN create 3μ syllables, and be LD

103 230 3.2 CL (f) MUST target the nearest possible segment

*

104 3.2 (LD)CL in Estonian: constraints

• Let’s consider what constraints must be responsible for these behaviors.

105 3.2 (LD)CL in Estonian: constraints

• Let’s consider what constraints must be responsible for these behaviors.

• First, the constraints from our typology of CL.

105 3.2 (LD)CL in Estonian: constraints

• Let’s consider what constraints must be responsible for these behaviors.

• First, the constraints from our typology of CL.

• The first of these is the locality constraint *CROSS.

105 3.2 (LD)CL in Estonian: constraints

• Let’s consider what constraints must be responsible for these behaviors.

• First, the constraints from our typology of CL.

• The first of these is the locality constraint *CROSS.

• This will explain (f) “CL must target the closest possible segment.”

105 3.2 (LD)CL in Estonian: constraints

• Let’s consider what constraints must be responsible for these behaviors.

• First, the constraints from our typology of CL.

• The first of these is the locality constraint *CROSS.

• This will explain (f) “CL must target the closest possible segment.”

• Next, a mora preservation constraint.

105 3.2 (LD)CL in Estonian: constraints

• Let’s consider what constraints must be responsible for these behaviors.

• First, the constraints from our typology of CL.

• The first of these is the locality constraint *CROSS.

• This will explain (f) “CL must target the closest possible segment.”

• Next, a mora preservation constraint.

• Since morae are preserved at the word-level, and not merely the foot-level, we will need to posit MAX-μ(ω), and rank it above *CROSS.

105 3.2 (LD)CL in Estonian: constraints

• Next, constraints based on the behavior of CL in Estonian.

106 3.2 (LD)CL in Estonian: constraints

• Next, constraints based on the behavior of CL in Estonian.

• First, (a) “CL cannot produce new long vowels” shows that some constraint must punish long vowels. We’ll call it *VV “vowels must be monomoraic.”

106 3.2 (LD)CL in Estonian: constraints

• Next, constraints based on the behavior of CL in Estonian.

• First, (a) “CL cannot produce new long vowels” shows that some constraint must punish long vowels. We’ll call it *VV “vowels must be monomoraic.”

• Second, (b) “CL cannot produce new geminates” shows that some constraint must punish geminates. Nevertheless, (c) “CL can make coda consonants moraic” reveals that it does NOT punish the creation of heavy syllables or moraic consonants. Therefore, we’ll posit *GEM: “consonants must belong to at most one syllable.”

106 3.2 (LD)CL in Estonian: constraints

• Next, constraints based on the behavior of CL in Estonian.

• First, (a) “CL cannot produce new long vowels” shows that some constraint must punish long vowels. We’ll call it *VV “vowels must be monomoraic.”

• Second, (b) “CL cannot produce new geminates” shows that some constraint must punish geminates. Nevertheless, (c) “CL can make coda consonants moraic” reveals that it does NOT punish the creation of heavy syllables or moraic consonants. Therefore, we’ll posit *GEM: “consonants must belong to at most one syllable.”

• Both *GEM and *VV must dominate MAX-μ(ω): the phonology may not create new long vowels or geminates, even if it means deleting a mora from the word.

106 3.2 (LD)CL in Estonian: constraints

• Next, constraints based on the behavior of CL in Estonian.

• First, (a) “CL cannot produce new long vowels” shows that some constraint must punish long vowels. We’ll call it *VV “vowels must be monomoraic.”

• Second, (b) “CL cannot produce new geminates” shows that some constraint must punish geminates. Nevertheless, (c) “CL can make coda consonants moraic” reveals that it does NOT punish the creation of heavy syllables or moraic consonants. Therefore, we’ll posit *GEM: “consonants must belong to at most one syllable.”

• Both *GEM and *VV must dominate MAX-μ(ω): the phonology may not create new long vowels or geminates, even if it means deleting a mora from the word.

• This gives us the following constraints so far, in order of ranking.

106 3.2 (LD)CL in Estonian: constraints

CONSTRAINT DESCRIPTION EFFECT

*VV Vowels must be short / monomoraic. No new long vowels. *GEM Consonants must belong to a single syllable. No new geminates. MAX-μ(ω) Do not delete morae from the word. Demands that CL take place. *CROSS Do not cross association lines. Keep CL as local as possible.

107 3.2 (LD)CL in Estonian: constraints

CONSTRAINT DESCRIPTION EFFECT

*VV Vowels must be short / monomoraic. No new long vowels. *GEM Consonants must belong to a single syllable. No new geminates. MAX-μ(ω) Do not delete morae from the word. Demands that CL take place. *CROSS Do not cross association lines. Keep CL as local as possible.

What other constraints must be present?

108 3.2 (LD)CL in Estonian: constraints

CONSTRAINT DESCRIPTION EFFECT

*VV Vowels must be short / monomoraic. No new long vowels. *GEM Consonants must belong to a single syllable. No new geminates. MAX-μ(ω) Do not delete morae from the word. Demands that CL take place. *CROSS Do not cross association lines. Keep CL as local as possible.

First, something must dominate *VV and *GEM: long vowels and geminates do exist in Estonian—but only in the primary-stressed syllable*. Therefore, some other, more specific MAX-μ constraint must be protecting them there and only there.

*Except for the genitive plural, /-tte/, and the “long” illative, /-sse/. 109 3.2 (LD)CL in Estonian: constraints

CONSTRAINT DESCRIPTION EFFECT

MAX-μ(σ́) Do not delete morae from primary-stressed syllables. Protects the first syllable. *VV Vowels must be short / monomoraic. No new long vowels. *GEM Consonants must belong to a single syllable. No new geminates. MAX-μ(ω) Do not delete morae from the word. Demands that CL take place. *CROSS Do not cross association lines. Keep CL as local as possible.

110 3.2 (LD)CL in Estonian: constraints

CONSTRAINT DESCRIPTION EFFECT

MAX-μ(σ́) Do not delete morae from primary-stressed syllables. Protects the first syllable. *VV Vowels must be short / monomoraic. No new long vowels. *GEM Consonants must belong to a single syllable. No new geminates. MAX-μ(ω) Do not delete morae from the word. Demands that CL take place. *CROSS Do not cross association lines. Keep CL as local as possible.

Next, recall that CL creates trimoraic syllables in Estonian. Any constraint that would normally prohibit this must be very low-ranking in Estonian.

111 3.2 (LD)CL in Estonian: constraints

CONSTRAINT DESCRIPTION EFFECT

MAX-μ(σ́) Do not delete morae from primary-stressed syllables. Protects the first syllable. *VV Vowels must be short / monomoraic. No new long vowels. *GEM Consonants must belong to a single syllable. No new geminates. MAX-μ(ω) Do not delete morae from the word. Demands that CL take place. *CROSS Do not cross association lines. Keep CL as local as possible.

*TERN(ARITY) Do not have ternary-branching structures (trimoraic syllables) Would have prevented Q3.

112 3.2 (LD)CL in Estonian: constraints

CONSTRAINT DESCRIPTION EFFECT

MAX-μ(σ́) Do not delete morae from primary-stressed syllables. Protects the first syllable. *VV Vowels must be short / monomoraic. No new long vowels. *GEM Consonants must belong to a single syllable. No new geminates. MAX-μ(ω) Do not delete morae from the word. Demands that CL take place. *CROSS Do not cross association lines. Keep CL as local as possible. *TERN(ARITY) Do not have ternary-branching structures (trimoraic syllables) Would have prevented Q3.

Finally, recall that only primary-stressed syllables can be trimoraic. We can capture this by positing a WTS constraint specific to trimoraic syllables, and a constraint demanding that stress be on the leftmost syllable. Both will be undominated.

(…and as such, will not be included in the tableaux below.)

113 3.2 (LD)CL in Estonian: constraints

CONSTRAINT DESCRIPTION EFFECT 3μ → S Trimoraic syllables must be stressed. Jointly: no non-initial Q3*. ALIGN-L(S,ω) Put primary stress as far left as possible. MAX-μ(σ́) Do not delete morae from primary-stressed syllables. Protects the first syllable. *VV Vowels must be short / monomoraic. No new long vowels. *GEM Consonants must belong to a single syllable. No new geminates. MAX-μ(ω) Do not delete morae from the word. Demands that CL take place. *CROSS Do not cross association lines. Keep CL as local as possible. *TERN(ARITY) Do not have ternary-branching structures (trimoraic syllables) Would have prevented Q3.

114 3.2 (LD)CL in Estonian: constraints

CONSTRAINT DESCRIPTION EFFECT 3μ → S Trimoraic syllables must be stressed. Jointly: no non-initial Q3*. ALIGN-L(S,ω) Put primary stress as far left as possible. MAX-μ(σ́) Do not delete morae from primary-stressed syllables. Protects the first syllable. *VV Vowels must be short / monomoraic. No new long vowels. *GEM Consonants must belong to a single syllable. No new geminates. MAX-μ(ω) Do not delete morae from the word. Demands that CL take place. *CROSS Do not cross association lines. Keep CL as local as possible. *TERN(ARITY) Do not have ternary-branching structures (trimoraic syllables) Would have prevented Q3.

This is our complete constraint list. Specific rankings are as follows.

115 3.2 (LD)CL in Estonian: constraints

CONSTRAINT DESCRIPTION EFFECT 3μ → S Trimoraic syllables must be stressed. Jointly: no non-initial Q3*. ALIGN-L(S,ω) Put primary stress as far left as possible. MAX-μ(σ́) Do not delete morae from primary-stressed syllables. Protects the first syllable. *VV Vowels must be short / monomoraic. No new long vowels. *GEM Consonants must belong to a single syllable. No new geminates. MAX-μ(ω) Do not delete morae from the word. Demands that CL take place. *CROSS Do not cross association lines. Keep CL as local as possible. *TERN(ARITY) Do not have ternary-branching structures (trimoraic syllables) Would have prevented Q3.

116 3.2 (LD)CL in Estonian: constraints

CONSTRAINT DESCRIPTION EFFECT 3μ → S Trimoraic syllables must be stressed. Jointly: no non-initial Q3*. ALIGN-L(S,ω) Put primary stress as far left as possible. MAX-μ(σ́) Do not delete morae from primary-stressed syllables. Protects the first syllable. *VV Vowels must be short / monomoraic. No new long vowels. *GEM Consonants must belong to a single syllable. No new geminates. MAX-μ(ω) Do not delete morae from the word. Demands that CL take place. *CROSS Do not cross association lines. Keep CL as local as possible. *TERN(ARITY) Do not have ternary-branching structures (trimoraic syllables) Would have prevented Q3.

Let’s consider the tableaux.

117 3.2 (LD)CL in Estonian: rankings

• From this point, we will only consider what happens after T-deletion and syncope / apocope have already applied, i.e. just after the partitive /ɑ/ has deleted, severing its mora.

118 3.2 (LD)CL in Estonian: rankings

• From this point, we will only consider what happens after T-deletion and syncope / apocope have already applied, i.e. just after the partitive /ɑ/ has deleted, severing its mora.

➢ (In reality, the timing is more complicated. Syncope / apocope and CL must take place simultaneously, while T- deletion must take place before both. We will not consider these facts here.)

118 3.2 (LD)CL in Estonian: rankings

• From this point, we will only consider what happens after T-deletion and syncope / apocope have already applied, i.e. just after the partitive /ɑ/ has deleted, severing its mora.

➢ (In reality, the timing is more complicated. Syncope / apocope and CL must take place simultaneously, while T- deletion must take place before both. We will not consider these facts here.)

• In the tableaux to follow, then, inputs will be presented as e.g. /emɑ - μ/, meaning “emɑ, plus the mora that was stranded after its original host vowel was deleted.”

118 3.2 (LD)CL in Estonian: rankings

• From this point, we will only consider what happens after T-deletion and syncope / apocope have already applied, i.e. just after the partitive /ɑ/ has deleted, severing its mora.

➢ (In reality, the timing is more complicated. Syncope / apocope and CL must take place simultaneously, while T- deletion must take place before both. We will not consider these facts here.)

• In the tableaux to follow, then, inputs will be presented as e.g. /emɑ - μ/, meaning “emɑ, plus the mora that was stranded after its original host vowel was deleted.”

➢ (Winners will be highlighted in green.)

118 257 3.2 (LD)CL in Estonian, tableaux: partitive [-∅]

/emɑ - μ/ )

ω (

“mother- μ

PART” -

*VV *GEM MAX a. [emɑː] *! b. [emːɑ] *! c. [eːmɑ] *! ☞ d. [emɑ] *

119 258 3.2 (LD)CL in Estonian, tableaux: partitive [-∅]

/emɑ - μ/ ) ω

• First, *VV and *GEM must dominate (

“mother- μ

MAX-μ(ω): you cannot create new -

long vowels or new geminates in PART”

*GEM MAX order to save a mora from deleting. *VV a. [emɑː] *! b. [emːɑ] *! c. [eːmɑ] *! ☞ d. [emɑ] *

119 259 3.2 (LD)CL in Estonian, tableaux: partitive [-∅]

/emɑ - μ/ ) ω

• First, *VV and *GEM must dominate (

“mother- μ

MAX-μ(ω): you cannot create new -

long vowels or new geminates in PART”

*GEM MAX order to save a mora from deleting. *VV a. [emɑː] *! • This explains the [∅]-partitive: any form of CL would have resulted in an b. [emːɑ] *! extra long vowel or geminate. c. [eːmɑ] *! ☞ d. [emɑ] *

119 260 3.2 (LD)CL in Estonian, tableaux: partitive [-∅] + Q3

/kɑːlu - μ/ )

ω (

“weight- μ

• Second, MAX-μ(ω) must -

dominate *CROSS, or CL will PART”

*GEM MAX *CROSS not take place to begin with. *VV ☞ a. [kɑːːlu] * **** b. [kɑːlu] * *!

120 261 3.2 (LD)CL in Estonian, tableaux: partitive [-∅] + Q3

/kɑːlu - μ/ ) ω

• MAX-μ(ω) must also (

“weight- μ

dominate *TERN, or CL will -

not be able to produce PART”

*GEM MAX *CROSS

trimoraic (Q3) syllables. *VV *TERN ☞ a. [kɑːːlu] * **** * b. [kɑːlu] * *!

121 262 3.2 (LD)CL in Estonian, tableaux: partitive [-∅] + Q3

• We’ve already seen that *VV and *GEM must dominate

MAX-μ(ω). Ergo, they must /kɑːlu - μ/ ) ω also dominate *CROSS. ( “weight- μ

PART” -

*VV *GEM MAX *CROSS *TERN ☞ a. [kɑːːlu] * **** * b. [kɑːlu] * *! c. [kɑːluː] **! d. [kɑːlːu] * *! ***

122 263 3.2 (LD)CL in Estonian, tableaux: partitive [-∅] + Q3

• We’ve already seen that *VV and *GEM must dominate

MAX-μ(ω). Ergo, they must /kɑːlu - μ/ ) ω also dominate *CROSS. ( “weight- μ

PART” -

*GEM MAX *CROSS

• This ranking is what gives *VV *TERN rise to LDCL: additional violations of *CROSS must be ☞ a. [kɑːːlu] * **** * incurred in order to avoid b. [kɑːlu] * *! violating *VV (creating new long vowels) or *GEM c. [kɑːluː] **! (creating new geminates). d. [kɑːlːu] * *! ***

122 264 3.2 (LD)CL in Estonian, tableaux: partitive [-∅] + Q3

• We’ve already seen that *VV and *GEM must dominate

MAX-μ(ω). Ergo, they must /kɑːlu - μ/ ) ω also dominate *CROSS. ( “weight- μ

PART” -

*GEM MAX *CROSS

• This ranking is what gives *VV *TERN rise to LDCL: additional violations of *CROSS must be ☞ a. [kɑːːlu] * **** * incurred in order to avoid b. [kɑːlu] * *! violating *VV (creating new long vowels) or *GEM c. [kɑːluː] **! (creating new geminates). d. [kɑːlːu] * *! *** • This explains the partitive [-∅] + lengthening to Q3.

122 265

3.2 (LD)CL in Estonian, tableaux: partitive [-∅] + Q3

́) )

/kɑːlu - μ/ σ

(

ω

(

μ -

“weight- μ -

• Finally, MAX-μ(σ́) must PART”

MAX

*GEM MAX *CROSS dominate *VV in order to *VV protect long vowels in the *TERN primary-stressed syllable ☞ a. [kɑːːlu] * **** * (which do not shorten). b. [kɑːlu] * *! c. [kɑːluː] **! d. [kɑːlːu] * *! *** e. [kɑlu] *! **

123 266

3.2 (LD)CL in Estonian, tableaux: partitive [-∅] + Q3

́) )

/kɑːlu - μ/ σ

(

ω

(

μ -

“weight- μ -

• Finally, MAX-μ(σ́) must PART”

MAX

*GEM MAX *CROSS dominate *VV in order to *VV protect long vowels in the *TERN primary-stressed syllable ☞ a. [kɑːːlu] * **** * (which do not shorten). b. [kɑːlu] * *! • The same is true for MAX-μ(σ́) c. [kɑːluː] **! and *GEM (not shown here). d. [kɑːlːu] * *! *** e. [kɑlu] *! **

123 267

3.2 (LD)CL in Estonian, tableaux: partitive [-∅] + Q3

́) )

/kɑːlu - μ/ σ

(

ω

(

μ -

• Notice the winning candidate “weight- μ - does not incur any additional PART”

violations of *VV, even though it

MAX

*VV *GEM MAX *CROSS does lengthen its vowel. *TERN ☞ a. [kɑːːlu] * **** * b. [kɑːlu] * *! c. [kɑːluː] **! d. [kɑːlːu] * *! *** e. [kɑlu] *! **

124 268

3.2 (LD)CL in Estonian, tableaux: partitive [-∅] + Q3

́) )

/kɑːlu - μ/ σ

(

ω

(

μ -

• Notice the winning candidate “weight- μ - does not incur any additional PART”

violations of *VV, even though it

MAX

*VV *GEM MAX *CROSS does lengthen its vowel. *TERN ☞ a. [kɑːːlu] * **** * • This is because long vowels and b. [kɑːlu] * *! overlong vowels both violate *VV to an equal degree—changing c. [kɑːluː] **! from one to the other does not make that violation any “worse.” d. [kɑːlːu] * *! *** e. [kɑlu] *! **

124 3.2 (LD)CL in Estonian, tableaux: partitive [-t] and [-tt]

• Finally, let’s consider what happens in those words where the partitive surfaces as [-t] and [-tt]

125 3.2 (LD)CL in Estonian, tableaux: partitive [-t] and [-tt]

• Finally, let’s consider what happens in those words where the partitive surfaces as [-t] and [-tt]

• In these forms, *CROSS ensures that the mora will not travel to the first syllable and (wrongly) produce Q3—candidates with LDCL will be harmonically bounded. (Tableaux to follow.)

125 3.2 (LD)CL in Estonian, tableaux: partitive [-t] and [-tt]

• Finally, let’s consider what happens in those words where the partitive surfaces as [-t] and [-tt]

• In these forms, *CROSS ensures that the mora will not travel to the first syllable and (wrongly) produce Q3—candidates with LDCL will be harmonically bounded. (Tableaux to follow.)

/helmes-tɑ/ “bead-PART” → /helmest – μ/ → [helmest] not *[helːmest] /ɤpːikːu-tɑ/ “textbook-PART” → /ɤpːikːut – μ/ → [ɤpːikːutt] not *[ɤpːːikːut] nor *[ɤpːikːːut]

125 3.2 (LD)CL in Estonian, tableaux: partitive [-t] and [-tt]

• Finally, let’s consider what happens in those words where the partitive surfaces as [-t] and [-tt]

• In these forms, *CROSS ensures that the mora will not travel to the first syllable and (wrongly) produce Q3—candidates with LDCL will be harmonically bounded. (Tableaux to follow.)

/helmes-tɑ/ “bead-PART” → /helmest – μ/ → [helmest] not *[helːmest] /ɤpːikːu-tɑ/ “textbook-PART” → /ɤpːikːut – μ/ → [ɤpːikːutt] not *[ɤpːːikːut] nor *[ɤpːikːːut]

(*TERN also rules these losing candidates out: they create trimoraic syllables when there is another candidate that does not, but which still preserves all morae within the word.)

125 3.2 (LD)CL in Estonian, tableaux: partitive [-t] and [-tt]

• Finally, let’s consider what happens in those words where the partitive surfaces as [-t] and [-tt]

• In these forms, *CROSS ensures that the mora will not travel to the first syllable and (wrongly) produce Q3—candidates with LDCL will be harmonically bounded. (Tableaux to follow.)

/helmes-tɑ/ “bead-PART” → /helmest – μ/ → [helmest] not *[helːmest] /ɤpːikːu-tɑ/ “textbook-PART” → /ɤpːikːut – μ/ → [ɤpːikːutt] not *[ɤpːːikːut] nor *[ɤpːikːːut]

(*TERN also rules these losing candidates out: they create trimoraic syllables when there is another candidate that does not, but which still preserves all morae within the word.)

• In other words: even in Estonian, a language that tolerates LDCL, local CL is the preferred means of preserving morae, all else being equal. This is entirely expected given the apparent cross-linguistic rarity of LDCL.

125

274 -

/helmest - μ/ -

)

́)

σ

ω (

“bead-PART” (

MAX

*VV *GEM MAX *CROSS

μ *TERN μ

☞ a.

b. ***! *!

c. *!

126

275 -

/ɤpːikːut - μ/ -

)

́)

σ

ω (

“textbook-PART” (

MAX

*VV *GEM MAX *CROSS

μ *TERN μ

☞ a. **

b. ** *****! *!

c. ** *!

127 276 3.2 (LD)CL in Estonian: complete ranking

In short, the complete ranking of constraints for Estonian is as shown to the left.

128 277 3.2 (LD)CL in Estonian: complete ranking

In short, the complete ranking of constraints for Estonian is as shown to the left.

Pulling back a bit—let’s consider some broader questions. If LDCL is possible, then why is it so rare? And if LDCL is so rare, then why does Estonian still display it?

128 3.2 Preconditions to LDCL

According to the account advanced here, LDCL is so rare because it can only happen under the following very specific set of circumstances:

129 3.2 Preconditions to LDCL

According to the account advanced here, LDCL is so rare because it can only happen under the following very specific set of circumstances:

• First, a moraic segment must be deleted or shortened, dislodging its mora.

129 3.2 Preconditions to LDCL

According to the account advanced here, LDCL is so rare because it can only happen under the following very specific set of circumstances:

• First, a moraic segment must be deleted or shortened, dislodging its mora.

• Second, MAX-μ(ω) must rank above *CROSS.

129 3.2 Preconditions to LDCL

According to the account advanced here, LDCL is so rare because it can only happen under the following very specific set of circumstances:

• First, a moraic segment must be deleted or shortened, dislodging its mora.

• Second, MAX-μ(ω) must rank above *CROSS.

› This ensures that no morae may be deleted from the word, even if preserving them means finding them a new host that lies an unbounded distance away.

129 3.2 Preconditions to LDCL

According to the account advanced here, LDCL is so rare because it can only happen under the following very specific set of circumstances:

• First, a moraic segment must be deleted or shortened, dislodging its mora.

• Second, MAX-μ(ω) must rank above *CROSS.

› This ensures that no morae may be deleted from the word, even if preserving them means finding them a new host that lies an unbounded distance away.

› But this isn’t enough by itself. In Estonian words that display the [-t] and [-tt] partitive, *CROSS guarantees that CL will always be local, even though the language tolerates LDCL elsewhere.

129 3.2 Preconditions to LDCL

According to the account advanced here, LDCL is so rare because it can only happen under the following very specific set of circumstances:

• First, a moraic segment must be deleted or shortened, dislodging its mora.

• Second, MAX-μ(ω) must rank above *CROSS.

› This ensures that no morae may be deleted from the word, even if preserving them means finding them a new host that lies an unbounded distance away.

› But this isn’t enough by itself. In Estonian words that display the [-t] and [-tt] partitive, *CROSS guarantees that CL will always be local, even though the language tolerates LDCL elsewhere.

• Some other constraint must force the phonology’s hand—it must dominate *CROSS, preventing morae from settling in any intermediate positions, forcing LDCL over local CL.

129 3.2 Preconditions to LDCL (cont.)

• In Estonian, this role was played by *VV and *GEM.

130 3.2 Preconditions to LDCL (cont.)

• In Estonian, this role was played by *VV and *GEM.

• In fact, these two constraints pull triple-duty:

130 3.2 Preconditions to LDCL (cont.)

• In Estonian, this role was played by *VV and *GEM.

• In fact, these two constraints pull triple-duty:

1) They explain the very robust prohibition in Estonian against long vowels and geminates outside the primary-stressed syllable*.

*Again, except in the genitive plural suffix, /-tte/, and the “long” illative, /-sse/.

130 3.2 Preconditions to LDCL (cont.)

• In Estonian, this role was played by *VV and *GEM.

• In fact, these two constraints pull triple-duty:

1) They explain the very robust prohibition in Estonian against long vowels and geminates outside the primary-stressed syllable*.

2) They force LDCL by barring morae from landing in intermediate positions.

*Again, except in the genitive plural suffix, /-tte/, and the “long” illative, /-sse/.

130 3.2 Preconditions to LDCL (cont.)

• In Estonian, this role was played by *VV and *GEM.

• In fact, these two constraints pull triple-duty:

1) They explain the very robust prohibition in Estonian against long vowels and geminates outside the primary-stressed syllable*.

2) They force LDCL by barring morae from landing in intermediate positions.

3) They explain why Estonian only displays lengthening from Q2 > Q3, and not from Q1 > Q2.

*Again, except in the genitive plural suffix, /-tte/, and the “long” illative, /-sse/.

130 A final comment: LDCL occurs elsewhere in Estonian

• Estonian phonology displays many instances of CL producing Q3, but also a high degree of opacity that often makes it difficult to glean the true underlying form for any given word.

131 A final comment: LDCL occurs elsewhere in Estonian

• Estonian phonology displays many instances of CL producing Q3, but also a high degree of opacity that often makes it difficult to glean the true underlying form for any given word.

• The partitive case was presented here because…

131 A final comment: LDCL occurs elsewhere in Estonian

• Estonian phonology displays many instances of CL producing Q3, but also a high degree of opacity that often makes it difficult to glean the true underlying form for any given word.

• The partitive case was presented here because…

1) Its underlying form is easy to justify:

131 A final comment: LDCL occurs elsewhere in Estonian

• Estonian phonology displays many instances of CL producing Q3, but also a high degree of opacity that often makes it difficult to glean the true underlying form for any given word.

• The partitive case was presented here because…

1) Its underlying form is easy to justify:

➢ the presence of /t/ coupled with deletion and lengthening explain the spectrum of [∅, t, tt].

131 A final comment: LDCL occurs elsewhere in Estonian

• Estonian phonology displays many instances of CL producing Q3, but also a high degree of opacity that often makes it difficult to glean the true underlying form for any given word.

• The partitive case was presented here because…

1) Its underlying form is easy to justify:

➢ the presence of /t/ coupled with deletion and lengthening explain the spectrum of [∅, t, tt]. ➢ the deletion of the vowel triggers CL, explaining /t/ > [tt] and Q2 > Q3.

131 A final comment: LDCL occurs elsewhere in Estonian

• Estonian phonology displays many instances of CL producing Q3, but also a high degree of opacity that often makes it difficult to glean the true underlying form for any given word.

• The partitive case was presented here because…

1) Its underlying form is easy to justify:

➢ the presence of /t/ coupled with deletion and lengthening explain the spectrum of [∅, t, tt]. ➢ the deletion of the vowel triggers CL, explaining /t/ > [tt] and Q2 > Q3.

2) It displays the full range of behaviors: LDCL, local CL, and deletion without CL.

131 A final comment: LDCL occurs elsewhere in Estonian

• Estonian phonology displays many instances of CL producing Q3, but also a high degree of opacity that often makes it difficult to glean the true underlying form for any given word.

• The partitive case was presented here because…

1) Its underlying form is easy to justify:

➢ the presence of /t/ coupled with deletion and lengthening explain the spectrum of [∅, t, tt]. ➢ the deletion of the vowel triggers CL, explaining /t/ > [tt] and Q2 > Q3.

2) It displays the full range of behaviors: LDCL, local CL, and deletion without CL.

3) An analysis of it requires only a set of processes that occur elsewhere in the language (apocope, syncope, T-deletion, and compensatory lengthening).

131 A final comment: LDCL occurs elsewhere in Estonian (cont.)

• Nevertheless, it should be emphasized that (LD)CL also produces Q3 in other word forms, e.g.…

132 A final comment: LDCL occurs elsewhere in Estonian (cont.)

• Nevertheless, it should be emphasized that (LD)CL also produces Q3 in other word forms, e.g.…

• The so-called “ma-infinitive,” underlyingly /-mɑ/ plus the illative suffix /-hɑ/.

• /ɤpːi-mɑhɑ/ “learn-INF” → [ɤpːːimɑ] • /kɑːlu-mɑhɑ/ “weigh-INF” → [kɑːːlumɑ]

132 A final comment: LDCL occurs elsewhere in Estonian (cont.)

• Nevertheless, it should be emphasized that (LD)CL also produces Q3 in other word forms, e.g.…

• The so-called “ma-infinitive,” underlyingly /-mɑ/ plus the illative suffix /-hɑ/.

• /ɤpːi-mɑhɑ/ “learn-INF” → [ɤpːːimɑ] • /kɑːlu-mɑhɑ/ “weigh-INF” → [kɑːːlumɑ]

• The genitive and locative cases of consonant-final nouns, which take an epenthetic vowel.

• /ʋɤːrɑsV-le/ “stranger-ALL” → [vɤːːrɑle] • /kɑlːisV-le/ “dear-ALL” → [kɑlːːile]

132 A final comment: LDCL occurs elsewhere in Estonian (cont.)

• Nevertheless, it should be emphasized that (LD)CL also produces Q3 in other word forms, e.g.…

• The so-called “ma-infinitive,” underlyingly /-mɑ/ plus the illative suffix /-hɑ/.

• /ɤpːi-mɑhɑ/ “learn-INF” → [ɤpːːimɑ] • /kɑːlu-mɑhɑ/ “weigh-INF” → [kɑːːlumɑ]

• The genitive and locative cases of consonant-final nouns, which take an epenthetic vowel.

• /ʋɤːrɑsV-le/ “stranger-ALL” → [vɤːːrɑle] • /kɑlːisV-le/ “dear-ALL” → [kɑlːːile]

• The present-tense forms of consonant-final verbs, which take an epenthetic vowel.

• /ʋɑitel-en/ “fight-1SG” → [ʋɑiːtlen] • /hypːɑʔ-en/ “jump-1GS” → [hypːːɑn] (with ghost-consonant /ʔ/)

132 IV. Conclusion

133 Conclusion

• In this talk, I have shown that LDCL is indeed possible, and have presented a way to account for its presence in Estonian, as well as its rarity cross-linguistically.

134 Conclusion

• In this talk, I have shown that LDCL is indeed possible, and have presented a way to account for its presence in Estonian, as well as its rarity cross-linguistically.

➢ LDCL only occurs when MAX-μ(ω) >> *CROSS, and some other constraint dominates *CROSS, forcing additional violations of it—in other words, forcing LDCL over local CL.

134 Conclusion

• In this talk, I have shown that LDCL is indeed possible, and have presented a way to account for its presence in Estonian, as well as its rarity cross-linguistically.

➢ LDCL only occurs when MAX-μ(ω) >> *CROSS, and some other constraint dominates *CROSS, forcing additional violations of it—in other words, forcing LDCL over local CL.

➢ A welcome side effect of this account is the prediction that LDCL should be marked even within Estonian, a language that tolerates it in some instances. This is indeed borne out.

134 Conclusion

• In this talk, I have shown that LDCL is indeed possible, and have presented a way to account for its presence in Estonian, as well as its rarity cross-linguistically.

➢ LDCL only occurs when MAX-μ(ω) >> *CROSS, and some other constraint dominates *CROSS, forcing additional violations of it—in other words, forcing LDCL over local CL.

➢ A welcome side effect of this account is the prediction that LDCL should be marked even within Estonian, a language that tolerates it in some instances. This is indeed borne out.

• The existence of LDCL is a challenge to purely phonetic accounts of CL (e.g. Kavitskaya 2002) – there would seem to be no phonetic motivation for lengthening a segment several syllables away. By contrast, the phonological motivation presented here is quite clear.

134 Conclusion

• In this talk, I have shown that LDCL is indeed possible, and have presented a way to account for its presence in Estonian, as well as its rarity cross-linguistically.

➢ LDCL only occurs when MAX-μ(ω) >> *CROSS, and some other constraint dominates *CROSS, forcing additional violations of it—in other words, forcing LDCL over local CL.

➢ A welcome side effect of this account is the prediction that LDCL should be marked even within Estonian, a language that tolerates it in some instances. This is indeed borne out.

• The existence of LDCL is a challenge to purely phonetic accounts of CL (e.g. Kavitskaya 2002) – there would seem to be no phonetic motivation for lengthening a segment several syllables away. By contrast, the phonological motivation presented here is quite clear.

• Future work will refine *CROSS, as well as develop a more complete typology involving all MAX-μ constraints and all DEP-μ constraints, in order to determine ranking volumes and thus the predicted frequency of certain CL behaviors cross-linguistically.

134 THANK YOU

( AND STAY SAFE AND HEALTHY !) Bibliography

Anttila, Arto. (2012). Modeling phonological variation. In Abigail C. Cohn, Cécile Fougeron, and Marie Huffman (eds), The Oxford Handbook of Laboratory Phonology, Oxford University Press, Oxford. pp. 76-91.

Blevins, J. P. (2008). Declension classes in Estonian. Linguistica Uralica, 44(4), 241-267.

Bye, P. (1997). A generative perspective on ‘overlength’ in Estonian and Saami. In Estonian : Papers from a Symposium (pp. 36-70). Institute of Tallinn.

Carson, Neusa M. (1981). Phonology and morphosyntax of Macuxi (Carib). Ph.D. dissertation, University of Kansas.

Clements, G. N. (1986). Compensatory lengthening and consonant gemina-tion in LuGanda. In Wetzels & Sezer (1986), 37–77.

Collinder, B. (1929). Über den finnisch-lappischen Quantitätsweschsel. Uppsala.

Ehala, M. (2003). Estonian quantity: Implications for Moraic Theory. In: Nelson, D., Manninen, S. (Ed.). Generative Approaches to Finnic and Saami Linguistics (51-80). Stanford: CSLI. (CSLI Lecture Notes; 148).

Gess, R. (2011). Compensatory lengthening. The Blackwell Companion to Phonology, 1-24.

Hayes, B. (1989). Compensatory lengthening in moraic phonology. Linguistic Inquiry, 20(2), 253-306. Bibliography (cont.)

Harms, R. (1968). Introduction to phonological theory. Englewood Gliffs,NJ: Prentic- Hall

Hualde, J. I. (1990). Compensatory lengthening in Friulian. Probus, 2(1),31-46.

Kavitskaya, D. (2002). Compensatory Lengthening: , Phonology, Diachrony. New York: Routledge.

Keyser, S. J., & Kiparsky, P. (1984). Syllable structure in . Language sound structure, 7-31.

Lipiński, Edward. 2001. Semitic languages: Outline of a comparative grammar. 2nd edn. Leuven: Peeters.

Prillop, K. (2013). Feet, syllables, moras and the Estonian quantity system. Linguistica Uralica, 49(1), 1-29.

Prince, A. S. (1980). A metrical theory for Estonian quantity. Linguistic inquiry, 511-562.

Topintzi, N. (2006). A (not so) paradoxical instance of compensatory lengthening: Samothraki Greek and theoretical implications. Journal of Greek Linguistics, 7(1), 71-119.

Veske, M. (1879). Eesti keele healte õpetus ja kirjastuse viis. Tartu. Some parting thoughts on

*CROSS

309 Future research: the Nature of *CROSS

• Nevertheless, more work needs to be done to refine *CROSS. Some questions that remain:

133 Future research: the Nature of *CROSS

• Nevertheless, more work needs to be done to refine *CROSS. Some questions that remain:

➢ Does it apply to the mora-to-segment line as well as the syllable-to-mora line?

133 Future research: the Nature of *CROSS

• Nevertheless, more work needs to be done to refine *CROSS. Some questions that remain:

➢ Does it apply to the mora-to-segment line as well as the syllable-to-mora line?

➢ How can we tell the two apart?

133 Future research: the Nature of *CROSS

• Nevertheless, more work needs to be done to refine *CROSS. Some questions that remain:

➢ Does it apply to the mora-to-segment line as well as the syllable-to-mora line?

➢ How can we tell the two apart?

➢ Does it treat all association lines the same way? For example…

133 Future research: the Nature of *CROSS

• Nevertheless, more work needs to be done to refine *CROSS. Some questions that remain:

➢ Does it apply to the mora-to-segment line as well as the syllable-to-mora line?

➢ How can we tell the two apart?

➢ Does it treat all association lines the same way? For example…

➢ CL that crosses vowels and moraic codas?

133 Future research: the Nature of *CROSS

• Nevertheless, more work needs to be done to refine *CROSS. Some questions that remain:

➢ Does it apply to the mora-to-segment line as well as the syllable-to-mora line?

➢ How can we tell the two apart?

➢ Does it treat all association lines the same way? For example…

➢ CL that crosses vowels and moraic codas? ➢ CL that crosses extrametrical or nonmoraic codas?

133 Future research: the Nature of *CROSS

• Nevertheless, more work needs to be done to refine *CROSS. Some questions that remain:

➢ Does it apply to the mora-to-segment line as well as the syllable-to-mora line?

➢ How can we tell the two apart?

➢ Does it treat all association lines the same way? For example…

➢ CL that crosses vowels and moraic codas? ➢ CL that crosses extrametrical or nonmoraic codas? ➢ CL that crosses onsets?

133 Future research: the Nature of *CROSS

• Nevertheless, more work needs to be done to refine *CROSS. Some questions that remain:

➢ Does it apply to the mora-to-segment line as well as the syllable-to-mora line?

➢ How can we tell the two apart?

➢ Does it treat all association lines the same way? For example…

➢ CL that crosses vowels and moraic codas? ➢ CL that crosses extrametrical or nonmoraic codas? ➢ CL that crosses onsets?

➢ Finally, do morae have to move one at a time (as assumed implicitly here)? Or can one displace the other, and so on down the line like a Newton’s Cradle?

133 Future research: the Nature of *CROSS

• Nevertheless, more work needs to be done to refine *CROSS. Some questions that remain:

➢ Does it apply to the mora-to-segment line as well as the syllable-to-mora line?

➢ How can we tell the two apart?

➢ Does it treat all association lines the same way? For example…

➢ CL that crosses vowels and moraic codas? ➢ CL that crosses extrametrical or nonmoraic codas? ➢ CL that crosses onsets?

➢ Finally, do morae have to move one at a time (as assumed implicitly here)? Or can one displace the other, and so on down the line like a Newton’s Cradle?

➢ This will not make much of a difference in most cases, but it will if there are no intervening onsets (where there will be no violations of *CROSS!).

133 Mora Movement: “Teleportation” or “Newton’s Cradle”?

Teleportation with onsets: Teleportation without onsets:

5 violations 2 violations

134 Mora Movement: “Teleportation” or “Newton’s Cradle”?

Newton’s Cradle with onsets: Newton’s Cradle without onsets:

3 violations 0 violations!

135