Morphophonology

Morphophonology (also morphophonemics, morphonology) is a branch of linguistics which studies, in general, the interaction between morphological and phonetic processes. When a morpheme is attached to a word, it can alter the phonetic environments of other morphemes in that word. Morphophonemics attempts to describe this process. A language's morphophonemic structure is generally described with a series of rules which, ideally, can describe every morphophonemic alternation that takes place in the language.

For an example of a morphophonological alternation in English, take the plural suffix. Written as "-s" or "-es" but generally understood to have the underlying representation /z/, the plural morpheme alternates between [s], [z], and [əz], as in cats, dogs, and horses, respectively. The plural suffix "-s" can also appear to alter phonemes directly surrounding it. As an example, the word "leaf" [lif] takes its plural by alternating the [f] with a [v] and adding the plural suffix, this time written as "-es" but pronounced as [z]. The result is "leaves" [livz]. Other words like "knife," "fife," and "dwarf" also display this alternation. This may be because the last phoneme in these words is actually an archiphoneme /F/ which may be realised as [f] or [v] depending on the context, even though those phonemes usually contrast. The archiphoneme is unspecified for voice, according to the rule: /F/ -> [αvoice] / __ [αvoice]. Because the underlying representation of the English plural suffix is /z/, a voiced consonant, the archiphoneme /F/ is realised as the voiced allophone [v].

Another example would be the different pronunciations for the past tense marker "-ed". After a voiceless sound, "-ed" is generally realised as [t], as in walked, hoped, wished, and so on.

Inflected and agglutinating languages may have extremely complicated systems of morphophonemics. Examples of complex morphophonological systems include:

• Consonant gradation, found in some Uralic languages such as Finnish, Estonian, Northern Sámi, and Nganasan.
• Vowel harmony, which occurs in varying degrees in languages all around the world, notably Turkic languages.
• Sandhi, the phenomena behind the English examples of plural and past tense above, is found in virtually all languages to some degree. Even Mandarin, which is sometimes said to display no morphology, nonetheless displays tone sandhi, a morphophonemic alternation.
• Ablaut found in English and other Germanic languages. Ablaut is the phenomena wherein stem vowels change form depending on context, as in English sing, sang, sung, song.

Morphophonemic Analysis

Morphophonemic Analysis designates the analytic procedures whereby paradigms with phonological alternations are reduced to underlying representations and phonological rules. The term "morphophonemic analysis" has a now obscure origin. In the 1940s and 1950s, many phonologists worked with a theory in which (roughly) all neutralising rules were assumed to apply before all allophonic rules. This in effect divided the phonology into two components: a neutralising component, whose units were called "morphophonemes," and a non-neutralizing component, which dealt with phonemes and allophones. This bifurcated-phonology theory is widely considered untenable today, but "morphophonemics" remains a useful term for characterising the study of neutralising phonological rules as they apply in paradigms.

• A Method for Morphophonemic Analysis

When we conduct morphophonemic analysis, we seek to establish a connection between data and theory. The theory in question is that morphemes are stored in the lexicon in an invariant phonemic form. They are then strung together by morphological and syntactic rules. Finally, they are converted to their surface forms by a sequence of (often neutralising) phonological rules, applied in a particular order. The purpose of morphophonemic analysis is to discover a set of underlying forms and ordered rules that is consistent with the data; and the payoff is that seemingly complex patterns are often reduced to simplicity. Morphophonemic analysis may be contrasted with phonemic analysis. Phonemic analysis is a more limited form of phonological analysis that seeks only to discover the non-neutralising (allophonic) rules of the phonology. In phonemic analysis, only the distribution and similarity of the phones is examined. Therefore, the data need not be grouped in paradigms, but need only comprise a sufficiently large and representative set of words. Like phonemic analysis, morphophonemic analysis can be pursued with a systematic method.

• Procedure for Morphophonemic Analysis

1. Examine the data, consulting the glosses, and make a provisional division of the forms into morphemes.
2. Find each morpheme that alternates, and locate all of its allomorphs.
3. Within each allomorph, locate the particular segment or segments that alternate.
4. Considering the logical possibilities, set up the underlying representations so that all the allomorphs of each morpheme can be derived from a single underlying representation by general phonological rules.

This overall scheme is elaborated step by step below.

• Pre-processing the data: phonemicization

It is almost always easier to do morphophonemic analysis with data that are already expressed as phonemes, so if this has not already been done, it is advisable first to reduce the data to phonemes.

• Morpheme division

The next step is to break up the forms into their component morphemes. A potential complication is that phonological alternations may obscure this division. In the hard cases, one must try more than one possibility for "placing the hyphens," ultimately selecting the choice that yields a working analysis. As the words are divided into morphemes, it is usually also possible to state and order the rules of morphology that are active.

• Setting up underlying representations

As with morpheme division, the problem of choosing the underlying representations often involves considering more than one hypothesis, with the final choice defended by its leading to a working analysis. The following strategy is often helpful. Suppose segment A alternates with segment B in the data. In such a case, the analyst should consider two possibilities:

1. Segments showing A ~ B alternation are underlyingly /A/, which is converted to [B] in certain contexts by one or more phonological rules.
2. Segments showing A ~ B alternation are underlyingly /B/, which is converted to [A] in certain contexts by one or more phonological rules.

In other words, always consider both directions.

To give a concrete example: if we were analysing Chimwiini, we would find many instances of long vowels alternating with short as in

[x-soːm-a] 'to read'

[x-som-oːw-a] 'to be read'

We would consider the possibility that such cases are underlyingly long vowels ('read' = /soːm/), and consider shortening rules (this turns out to be correct), as well as the possibility that these are underlyingly short vowels ('read' = /som/), and consider lengthening rules.

• Constructing underlying representations under a particular hypothesis

Assuming that you have picked a particular direction for the rules (/A/=>[B], or /B/=>[A]) and are trying it out, the next step is to construct underlying representations. Here is a recommended procedure:

1. Segments that do not alternate can (normally) be assumed to be phonemically identical in their underlying representation to their surface representation. (This presupposes, as already noted, that phonemic analysis is already accomplished, so any positional allophones will already appear in their underlying form.)
2. For segments that alternate, follow the hypothesis you made about underlying forms, implementing it consistently through the data. Thus if you are assuming that an alternation A ~ B found in a particular context, is underlain by A, you should set up /A/ in the underlying representation for all such alternations in that context.
3. Be sure that the underlying representation of each morpheme is uniform through its paradigm – this is a basic hypothesis of the theory you are assuming.

In our example of [x-soːm-a] ~ [x-som-oːw-a], under the hypothesis that the rule is a shortening rule, these principles force us to set up the underlying representations /soːm/ for the root, and /-oːw/ for the invariantly long passive suffix.

The final vowel /-a/ turns out to be a special case: its surface length is actually non-distinctive, being determined entirely by the phonological rules. Our grammar will work no matter what underlying length is assigned to this suffix.

• Working out the rules

When you have a suitable set of hypothesized underlying forms, it is helpful to arrange them in a row, aligning their corresponding surface forms underneath them, as follows:

'to read'	'to be read'	'to stop for one'
/x-soːm-a/	/x-soːm-o:w-a/	/ku-reːb-eɺ-an-a/	underlying forms
...	...	...	add rules here
[xsoːma]	[xsomoːwa]	[kurebeɺana]	surface forms

It is then a matter of coming up with a single rule system that will derive the bottom row from the top. If you get stuck doing this, you can try collecting the local environments for the sounds that change, as described above for phonemic analysis.

• A clue for choosing underlying representations

When you are deciding whether to set up underlying A and derive B from it, or vice versa, there is often a clue in the data to guide you, namely, a contextually limited contrast. In the present case, note that while vowel length is phonemic in Chimwiini, only short vowels are allowed when more than three syllables from the end of a phrase, or when a long vowel follows. Such limitations are a strong clue that there must be a rule that wipes out the contrast in these environments.

Another way of saying the same thing is: don't analyze in a direction opposite to that of a neutralisation. When we analyze Chimwiini with shortening, or analysis fits in well with the contextually neutralised distribution of long and short vowels in the language. If, however, we try to analyse Chimwiini with lengthening, the phonological distribution will stymie us. The following quadruplet of forms should make this point clear.

[x-kuːl-a] 'to extract' [x-kul-oːw-a] 'to be extracted'
[x-kul-a] 'to grow' [x-kul-oːw-a] 'to be grown'

The top row of forms shows an alternation between [uː] and [u], which we earlier analyzed assuming underlying /uː/ and the neutralising rule of Pre-Long Shortening. It is plain that Pre-Long Shortening is neutralizing, since the passive form of [x-ku:l-a], [x-kul-oːw-a], is identical to the passive of [x-kul-a], meaning 'to grow'. If we had wrongly chosen underlying /u/ for the root meaning 'extract', we would be defeated: no matter what lengthening rule we tried, it would be unable to derive [x-kuːl-a] for 'extract' and [x-kul-a] for 'grow', since these two forms would have the same underlying representation.

The Isolation Form Shortcut and Why It Sometimes Fails

When one is looking for underlying forms, it is tempting to appeal to a "shortcut" that finds them with great speed:

The Isolation Form Shortcut

"The underlying form of a stem is simply the way that the stem appears in isolation (taking away the effects of any allophonic rules)."

This strategy particularly suggests itself for languages like English, where stems frequently appear alone. Hearing an alternation like [ˈplænt] ~ [ˈplænɪŋ] (plant ~ planting; we are tempted to take the evidence of the isolation form [ˈplænt] as evidence sufficient in and of itself to justify the underlying form /ˈplænt/. This turns out to work fine for this particular case.

However, the Isolation Form Shortcut does not work in general. The reason for this lies in how the system is set up, and simple logic: it is certainly possible that neutralisation rules could apply just in case no affix is added to the stem. We would say that in such cases, the affix "protects" the stem from the neutralizing rule, serving as a kind of buffer.

To make this more precise: neutralising phonological rules are often conditioned by word edge; that is, they have environments like /___]_word. When an affix is present, a stem will be buffered by the affix, and the crucial rule won't apply. Indeed, the rule will apply in only those members of the paradigm where there is no affix, so that the buffering effect is absent.

Phonologies that have this kind of phenomenon are quite common, occurring in Korean, Japanese, English, German, Russian, and many other languages.

Rule-Ordering Terminology

• Feeding

Observe that Apocope, when it exposes a consonant cluster at the end of a word, thereby makes it possible for Cluster reduction to apply. The following abbreviated derivation shows this:

/jukaɾpa/	underlying form
jukaɾp	Apocope
jukaɾ	Cluster Reduction
[jukaɾ]	surface form

This is said to be a case of feeding: Apocope "feeds" Cluster reduction. The term is defined in general as follows:

Rule A feeds rule B when:

– A is ordered before B, and

– A creates novel configurations to which B may apply.

• Bleeding

Consider next the interaction of /w/ Epenthesis and Vowel Deletion, shown in the following abbreviated derivation:

/papi-uɻ/	underlying form
papiwuɻ	/w/ Epenthesis
−	Vowel Deletion
[papiwuɻ]	surface form

It is clear that if /w/ Epenthesis had not applied, then Vowel Deletion would have had an additional chance to apply, creating *[papiɻ]. Thus, we might say that /w/ Epenthesis, in this particular derivation, "blocks" or "pre-empts" Vowel Deletion. The standard term used, however, is bleeding; /w/ Epenthesis bleeds Vowel Deletion. More generally:

Rule A bleeds rule B when:

– A is ordered before B, and

– A removes configurations to which B could otherwise have applied.

Orthographic context

See also: Morpho-phonemic orthography

The English plural morpheme s is written the same regardless of its pronunciation: cats, dogs. This is a morphophonemic spelling. If English used a purely phonemic orthography (the same system without any morphemic considerations), these could be spelled cats and dogz, because /s/ and /z/ are separate phonemes in English.

To some extent English orthography reflects the etymology of its words, and as such it is partially morphophonemic. This explains not only cats /s/ and dogs /z/, but also science /saɪ/ vs. unconscious /ʃ/, prejudice /prɛ/ vs. prequel /priː/, chased /t/ vs. loaded /ɪd/, sign /saɪn/ signature /sɪɡn/, nation /neɪ/ vs. nationalism /næ/, and special /spɛ/ vs. species /spiː/, etc.

Most morphophonemic orthographies, however, reflect only active morphology, like cats vs. dogs, or chased vs. loaded. Turkish and German both have broadly phonemic writing systems, but while German is morphophonemic, transcribing the "underlying" phonemes, Turkish is purely phonemic, transcribing surface phonemes only (at least traditionally; this appears to be changing). For example, Turkish has two words, /et/ 'meat' and /et/ 'to do', which in isolation appear to be homonyms. However, when a vowel follows, the roots diverge: /eti/ 'his meat', but /edir/ 'he does'. In Turkish when a root that ends in a /d/ appears without a following vowel, the /d/ becomes /t/ (final obstruent devoicing), and that is reflected in the spelling: et, et, eti, edir.

German has a similar relationship between /t/ and /d/. The words for 'bath' and 'advice' are /bat/ and /rat/, but the verbal forms are /badən/ 'to bathe' and /ratən/ 'to advise'. However, they are spelled Bad, baden and Rat, raten as if the consonants didn't change at all. Indeed, a speaker may perceive that the final consonant in Bad is different from the final consonant of Rat because the inflections differ, even though they are pronounced the same. A morphophonemic orthography such as this has the advantage of maintaining the orthographic shape of the root regardless of the inflection, which aids in recognition while reading.

In the International Phonetic Alphabet, pipes (| |) are often used to indicate a morphophonemic rather than phonemic representation. Another common convention is double slashes (// //), iconically implying that the transcription is 'more phonemic than simply phonemic'. Other conventions sometimes seen are double pipes (|| ||) and curly brackets ({ }).

Table. The underlying (morpho-phonemic), phonemic, and phonetic representations of four German and Turkish words. (In the Turkish examples, //Ü// represents an underlying high vowel that, as a result of Turkish vowel harmony, may surface as any one of the four phonemes /i y ɯ u/.)

	word	morpho- phonemic	phonemic	phonetic
German	Bad	//bad//	/bat/	[bat]
	baden	//badən//	/badən/	[badən]
	Rat	//rat//	/rat/	[ʀat]
	raten	//ratən//	/ratən/	[ʀatən]
Turkish	et	//ed//	/et/	[ɛt]
	edir	//edÜr//	/edir/	[edir]
	et	//et//	/et/	[ɛt]
	eti	//etÜ//	/eti/	[eti]

Another example of a morphophonemic orthography is modern hangul, and even more so the obsolete North Korean Chosŏn-ŏ sinch'ŏlchabŏp orthography.

References

• Hayes, Bruce (2009). "Morphophonemic Analysis" Introductory Phonology, pp. 161–185. Blackwell