Please leave a comment on one of the papers that we are discussing in Week 11 by the end of the day Sunday the 11th.
7 thoughts on “Comments on phonetically driven phonology”
Jon Ander Mendia
My comments are on Steriade (2008). I have found the article very interesting. Since the very beginning of the course I wondered how OT theories would accommodate context-dependent phenomena on a general basis. I like Steriade’s idea of solving the ‘too-many-problem’ by positing some means to measure the distance between two contrasts in a given context. The system, however, only works for those cases in which we observe some *invariant* preference for one repair strategy over others. I wonder whether there is phonological phenomena in natural languages showing some infrequent variation; that is, if, for the same context, repair strategy A were chosen 8 out of 10 times, whereas repair strategy B were chosen just 2 out of 10. What could this theory say about such a situation?
I also would like to learn more about how the system picks up different contexts for comparison. For example, if P-map ranks voicing relative to a distinctiveness scale in different contexts (see example (8)), how does the system select which contexts to evaluate for similarity/distinctiveness? I guess that not infinitely many, but only some relevant ones. Now, how does the system know which contexts are relevant?
My last question is rather a clarification question. In section 7.4.2, the author introduces a notation for correspondence constraint. If I understood the rationale correctly, it must be the case that if dif(x-y/_K.i) > dif(w-z/_K.i) then, CORRESP(x-y/_K.i) >> CORRESP(w-z/_K.i) (where “dif” stands for Steriade’s little triangle). What I do not understand is the discussion about “parallel constraints”. Why should there be a difference between Base-Derivative and non Base-Derivative constraints? Is it that parallel constraints only apply to output-output correspondences? I find the discussion about (20) confusing.
1) In Pater 2004, there is a mention of both I,O and O,I correspondence constraints. Why do we need O,I correspondence constraints? My best guess is that they are needed for epenthesis-related phenomena because in these instances there would be something in the output that does not correspond to something in the input, which you may want to reference.
2) I liked Lombardi 1991’s account of regressive and progressive voicelessness assimilation in Swedish. She accounts for this assimilation by ranking Agree and Ident[Lar] above *Lar. This ranking essentially forces voicelessness agreement (Agree and *Lar), but only let’s you change up to one segment to do so (Ident[Lar]; e.g. ägde, *äkte).
On that note, I did not understand why she introduced Harm’s generalisation (as a constraint?) to account for progressive voicing assimilation in the English plural. Earlier, she ranked Agree at the bottom of the constraint hierarchy for Egnlish because pigpen and *pigben, *pikpen. Clearly, these instances of voicing assimilation are different from voicing assimilation in consonant clusters. Something like AgreeCodaVoice or AgreeCCVoice and IdentObsVoice would have made more sense to me.
　I read Pater (2004), so I’ll make some comments on it. First of all, this paper elegantly offers a unified explanation for a wide variety of *NC effects with OT constraints ranking, including nasal substitution, nasal deletion, denasalization, and post-nasal voicing. Furthermore, this paper argues that so-called “conspiracy†effects provide us with crucial evidence to favor OT instead of rule-based phonology. Given this, my specific questions are as follows:
(i) Can we find empirically fatal flaws in rule-based phonology? (i.e. some data which OT can explain, but other phonological theories cannot)
(ii) Is “fusion†an operation in GEN? If so, does the fusion operation generate infinite candidates with not only nasal/voiceless obstruent fusion but also other arbitrary fusions? I’m just wondering whether the fusion operation in GEN component over-generates possible languages.
(iii) I came up with another possibility to repair *NC effects: metathesis. Are ‘metathesis’ solutions for *NC effects attested so far in natural languages?
(iv) I’m curious to know recent results in phonology which demonstrates that typological gaps discovered in this paper can be resolved in some way, if any. Especially, I’m interested in no epenthesis strategies to escape from *NC and a mysterious asymmetry in assimilation between progression and regression.
In Pater (1999/2004) we are presented with an analysis of different languages in order to show how the role of diverse faithfulness constraints allows us to account for the strategies that these languages follow to deal with NC˳ clusters. The general idea is that NC˳ needs to be ranked above one faithfulness constraint or above two in the case of conspiracies, as argued in the last part of the paper, but beneath the rest of faithfulness constraints. One of the strategies followed by languages that don’t allow this type of clusters is fusion of the nasal and the voiceless obstruent, as in the case of Indonesian. This is possible in a language in which Max and Dep rank above NC˳ but Linearity is ranked beneath NC˳. However, if we shift Linearity and Max, having Max below NC˳, there will be deletion of the nasal segment, as it occurs in languages such as Kelantan Malay or Swahili, in which the deletion of the nasal instead of the obstruent can be explained by using specific versions of Max: ObsMax ranks above NasMax. A language in which Dep is ranked below has not been attested, which may posit some theoretical problems and the need for future research. Denasalization is another possibility to deal with this type of clusters and results in gemination when NC˳ ranks above IdentI→O[nas], constraint that states that the features present in one segment in the input need to be present in the output; this is the case of Mandar. A different version of this constraint, IdentO→I[nas], permits to account for geminate nasals, which occurs in languages such as Konjo. Another strategy presented in the paper is post-nasal devoicing, possible in affixal post-nasal consonants in Puyo Pungo Quechua, where Ident[ObsVce] is ranked beneath NC˳. This brief summary of the paper aims to show the apparent adequacy of using faithfulness constraints to explain processes of final devoicing taking an Optimality-Theory perspective.
One of the things I liked the most about the analysis presented in Pater (1999/2004) is the restriction of some constraints at either one specific word level (root or affixes), as done with Root Linearity in order to account for the lack of root-internal nasal substitution in Indonesian, or to one specific feature, as in the case of IdentI→O[nas] for the description of Mandar. This restriction has been proved to be very helpful also in Lombardi’s (1999), where an identity constraint is restricted to the onset position to preserve underlying laryngeal segments in the output (IDentOnset[Laryngeal]) in order to better explain the direction of voicing assimilation of consonant clusters.
Pater 2004 introduced thus far unfamiliar subcategories of feature-dependent faithfulness constraints that regulate NC correspondence. And whle IdentI –> O[F], where F = Nasal accurately fits the described typology, I do not understand why Ident[Nas] does not do the job. While IdentI –> O[F] is the more general and thereby more universal, I do not understand how the outputs in 13a and 13b violate Ident[Nasal] to the same degree, as described in the paper. All in all, the paper on NC Effects is an enlightening study on phonetically motivated asymmetry within a phonological grammar.
While Lombadi’s account of voicing assimilation operated with a more narrow, simpler set of constraints, it did not sufficiently address phonetic asymmetry. Even though the typology did not predict exclusively progressive voicing assimilation, I imagine there is a phenomena that would best be described through an interaction of IdentOnset(Laryngeal) IdentCoda(Laryngeal).
Also, we have thus far been predicting typology as a factorial of the number of constraints we are using, minus equivalent rankings. However a more extensive ranking, which would encompass a much greater number of constraints than we’ve been dealing with thus far, what is the limiting factor on the number of constraints in GEN? What stops us from constructing an overly broad, or too specific of a constriant? For instance, Harms’ generalization seems to be too specific of a constraint and I wonder if it is used to account for languages other than English. If not, is the existence of the constraint still motivated?
I also wonder about the nature of Ident(Laryngeal) and whether Lombardi’s decision to use this constraint and not Ident(Voice) is crucially motivated.
The paper on which my comments will be made is Pater (2004). This paper argues for the fact that Optimality Theoretic approach does a better job in analyzing NC effects cross-linguistically than purely rule-based approaches, because the effects of conspiracies can be formalized into the constraints while the functional connection between the rules dealing with NC violations in rule-based approaches is not directly retrievable from the rules themselves. Generalizationwise, the strategies used to avoid NC sequences in different languages can be accounted for by the constraints that look at input-output correspondence. Whether the strategy is nasal substitution, deletion, denasalization, or post-nasal voicing is determined by the ranking of the constraints that are posited to focus on the segments or phonological features of the NC sequence. However, my question arises here: would there be any influence, good or bad, to the OT framework itself if we incorporate phonetically conditioned constraints into the system? Do we have to differentiate the levels of representation of constraint evaluation in terms of segmental and featural or we can just put them all into one single evaluation process to account for NC avoidance strategies?
It is also said in the paper that by moving different constraint to the bottom of the ranking, we could get different strategies used in different languages to avoid NC sequences. In some cases, a language may even adopt two strategies for NC avoidance, say post-nasal voicing in the morphological context of prefix + root, and nasal substitution in the root-initial context. Does that mean there are two sets of constraints, hence, two grammars, in this given language? If this language uses another strategy in yet another morphological context, such as denasalization in total reduplication (ex. [ton] → [totton]: denasalization), does that make it a language with three different grammars in various morphological contexts? I think my real question is: do we just simply live with this result, saying that this is ideal in accounting for the phonological phenomenon under discussion although we are not sure if the conclusion of a language with many separate grammars is desired? Or do we want to find out if there is any possibility of formalizing different morphological contexts into the constraints so that all of the NC avoidance strategies a language adopts can be accounted for by the ranking of one set of constraints?
I read Pater (2004/1999). I find the concept of fusion shown in this article fairly interesting aspect of study, because of the investigative questions/challenges it provided. As could be expected the task of the investigator was to find out what constraints would provide cause a language to choose fusion instead of other likely options such as, epenthesis, deletion or denasalization. For this reason I feel as though this article is not solely useful because of its comment on the importance of the division between phonotactic and Faithfulness constraints, but also because of its use of constraints such as: IdentO-I[Nas], and Ident-[F]. I believe that this article shows how to appropriately define constraints in order to form appropriate rankings.
My comments are on Steriade (2008). I have found the article very interesting. Since the very beginning of the course I wondered how OT theories would accommodate context-dependent phenomena on a general basis. I like Steriade’s idea of solving the ‘too-many-problem’ by positing some means to measure the distance between two contrasts in a given context. The system, however, only works for those cases in which we observe some *invariant* preference for one repair strategy over others. I wonder whether there is phonological phenomena in natural languages showing some infrequent variation; that is, if, for the same context, repair strategy A were chosen 8 out of 10 times, whereas repair strategy B were chosen just 2 out of 10. What could this theory say about such a situation?
I also would like to learn more about how the system picks up different contexts for comparison. For example, if P-map ranks voicing relative to a distinctiveness scale in different contexts (see example (8)), how does the system select which contexts to evaluate for similarity/distinctiveness? I guess that not infinitely many, but only some relevant ones. Now, how does the system know which contexts are relevant?
My last question is rather a clarification question. In section 7.4.2, the author introduces a notation for correspondence constraint. If I understood the rationale correctly, it must be the case that if dif(x-y/_K.i) > dif(w-z/_K.i) then, CORRESP(x-y/_K.i) >> CORRESP(w-z/_K.i) (where “dif” stands for Steriade’s little triangle). What I do not understand is the discussion about “parallel constraints”. Why should there be a difference between Base-Derivative and non Base-Derivative constraints? Is it that parallel constraints only apply to output-output correspondences? I find the discussion about (20) confusing.
One question and one comment:
1) In Pater 2004, there is a mention of both I,O and O,I correspondence constraints. Why do we need O,I correspondence constraints? My best guess is that they are needed for epenthesis-related phenomena because in these instances there would be something in the output that does not correspond to something in the input, which you may want to reference.
2) I liked Lombardi 1991’s account of regressive and progressive voicelessness assimilation in Swedish. She accounts for this assimilation by ranking Agree and Ident[Lar] above *Lar. This ranking essentially forces voicelessness agreement (Agree and *Lar), but only let’s you change up to one segment to do so (Ident[Lar]; e.g. ägde, *äkte).
On that note, I did not understand why she introduced Harm’s generalisation (as a constraint?) to account for progressive voicing assimilation in the English plural. Earlier, she ranked Agree at the bottom of the constraint hierarchy for Egnlish because pigpen and *pigben, *pikpen. Clearly, these instances of voicing assimilation are different from voicing assimilation in consonant clusters. Something like AgreeCodaVoice or AgreeCCVoice and IdentObsVoice would have made more sense to me.
　I read Pater (2004), so I’ll make some comments on it. First of all, this paper elegantly offers a unified explanation for a wide variety of *NC effects with OT constraints ranking, including nasal substitution, nasal deletion, denasalization, and post-nasal voicing. Furthermore, this paper argues that so-called “conspiracy†effects provide us with crucial evidence to favor OT instead of rule-based phonology. Given this, my specific questions are as follows:
(i) Can we find empirically fatal flaws in rule-based phonology? (i.e. some data which OT can explain, but other phonological theories cannot)
(ii) Is “fusion†an operation in GEN? If so, does the fusion operation generate infinite candidates with not only nasal/voiceless obstruent fusion but also other arbitrary fusions? I’m just wondering whether the fusion operation in GEN component over-generates possible languages.
(iii) I came up with another possibility to repair *NC effects: metathesis. Are ‘metathesis’ solutions for *NC effects attested so far in natural languages?
(iv) I’m curious to know recent results in phonology which demonstrates that typological gaps discovered in this paper can be resolved in some way, if any. Especially, I’m interested in no epenthesis strategies to escape from *NC and a mysterious asymmetry in assimilation between progression and regression.
In Pater (1999/2004) we are presented with an analysis of different languages in order to show how the role of diverse faithfulness constraints allows us to account for the strategies that these languages follow to deal with NC˳ clusters. The general idea is that NC˳ needs to be ranked above one faithfulness constraint or above two in the case of conspiracies, as argued in the last part of the paper, but beneath the rest of faithfulness constraints. One of the strategies followed by languages that don’t allow this type of clusters is fusion of the nasal and the voiceless obstruent, as in the case of Indonesian. This is possible in a language in which Max and Dep rank above NC˳ but Linearity is ranked beneath NC˳. However, if we shift Linearity and Max, having Max below NC˳, there will be deletion of the nasal segment, as it occurs in languages such as Kelantan Malay or Swahili, in which the deletion of the nasal instead of the obstruent can be explained by using specific versions of Max: ObsMax ranks above NasMax. A language in which Dep is ranked below has not been attested, which may posit some theoretical problems and the need for future research. Denasalization is another possibility to deal with this type of clusters and results in gemination when NC˳ ranks above IdentI→O[nas], constraint that states that the features present in one segment in the input need to be present in the output; this is the case of Mandar. A different version of this constraint, IdentO→I[nas], permits to account for geminate nasals, which occurs in languages such as Konjo. Another strategy presented in the paper is post-nasal devoicing, possible in affixal post-nasal consonants in Puyo Pungo Quechua, where Ident[ObsVce] is ranked beneath NC˳. This brief summary of the paper aims to show the apparent adequacy of using faithfulness constraints to explain processes of final devoicing taking an Optimality-Theory perspective.
One of the things I liked the most about the analysis presented in Pater (1999/2004) is the restriction of some constraints at either one specific word level (root or affixes), as done with Root Linearity in order to account for the lack of root-internal nasal substitution in Indonesian, or to one specific feature, as in the case of IdentI→O[nas] for the description of Mandar. This restriction has been proved to be very helpful also in Lombardi’s (1999), where an identity constraint is restricted to the onset position to preserve underlying laryngeal segments in the output (IDentOnset[Laryngeal]) in order to better explain the direction of voicing assimilation of consonant clusters.
Pater 2004 introduced thus far unfamiliar subcategories of feature-dependent faithfulness constraints that regulate NC correspondence. And whle IdentI –> O[F], where F = Nasal accurately fits the described typology, I do not understand why Ident[Nas] does not do the job. While IdentI –> O[F] is the more general and thereby more universal, I do not understand how the outputs in 13a and 13b violate Ident[Nasal] to the same degree, as described in the paper. All in all, the paper on NC Effects is an enlightening study on phonetically motivated asymmetry within a phonological grammar.
While Lombadi’s account of voicing assimilation operated with a more narrow, simpler set of constraints, it did not sufficiently address phonetic asymmetry. Even though the typology did not predict exclusively progressive voicing assimilation, I imagine there is a phenomena that would best be described through an interaction of IdentOnset(Laryngeal) IdentCoda(Laryngeal).
Also, we have thus far been predicting typology as a factorial of the number of constraints we are using, minus equivalent rankings. However a more extensive ranking, which would encompass a much greater number of constraints than we’ve been dealing with thus far, what is the limiting factor on the number of constraints in GEN? What stops us from constructing an overly broad, or too specific of a constriant? For instance, Harms’ generalization seems to be too specific of a constraint and I wonder if it is used to account for languages other than English. If not, is the existence of the constraint still motivated?
I also wonder about the nature of Ident(Laryngeal) and whether Lombardi’s decision to use this constraint and not Ident(Voice) is crucially motivated.
The paper on which my comments will be made is Pater (2004). This paper argues for the fact that Optimality Theoretic approach does a better job in analyzing NC effects cross-linguistically than purely rule-based approaches, because the effects of conspiracies can be formalized into the constraints while the functional connection between the rules dealing with NC violations in rule-based approaches is not directly retrievable from the rules themselves. Generalizationwise, the strategies used to avoid NC sequences in different languages can be accounted for by the constraints that look at input-output correspondence. Whether the strategy is nasal substitution, deletion, denasalization, or post-nasal voicing is determined by the ranking of the constraints that are posited to focus on the segments or phonological features of the NC sequence. However, my question arises here: would there be any influence, good or bad, to the OT framework itself if we incorporate phonetically conditioned constraints into the system? Do we have to differentiate the levels of representation of constraint evaluation in terms of segmental and featural or we can just put them all into one single evaluation process to account for NC avoidance strategies?
It is also said in the paper that by moving different constraint to the bottom of the ranking, we could get different strategies used in different languages to avoid NC sequences. In some cases, a language may even adopt two strategies for NC avoidance, say post-nasal voicing in the morphological context of prefix + root, and nasal substitution in the root-initial context. Does that mean there are two sets of constraints, hence, two grammars, in this given language? If this language uses another strategy in yet another morphological context, such as denasalization in total reduplication (ex. [ton] → [totton]: denasalization), does that make it a language with three different grammars in various morphological contexts? I think my real question is: do we just simply live with this result, saying that this is ideal in accounting for the phonological phenomenon under discussion although we are not sure if the conclusion of a language with many separate grammars is desired? Or do we want to find out if there is any possibility of formalizing different morphological contexts into the constraints so that all of the NC avoidance strategies a language adopts can be accounted for by the ranking of one set of constraints?
I read Pater (2004/1999). I find the concept of fusion shown in this article fairly interesting aspect of study, because of the investigative questions/challenges it provided. As could be expected the task of the investigator was to find out what constraints would provide cause a language to choose fusion instead of other likely options such as, epenthesis, deletion or denasalization. For this reason I feel as though this article is not solely useful because of its comment on the importance of the division between phonotactic and Faithfulness constraints, but also because of its use of constraints such as: IdentO-I[Nas], and Ident-[F]. I believe that this article shows how to appropriately define constraints in order to form appropriate rankings.