What are UR and SF?
Speakers in all languages possess abstract concepts about the sounds they articulate. In other words, they believe that they have an accurate awareness of speech sounds that they utter. Typically, though, there is a discrepancy between speakers’ own ideas about the sounds in their language use, and what they are actually articulating in practice. The underlying representation (UR) refers to speakers’ abstract concepts of their phones (language sounds), and the surface form (SF) refers the phones that are actually produced.
This does not at all stand to mean that UR somehow ‘does not exist’, nor does it entail that UR is an ‘incorrect’ version of SF. To clarify, the UR does exist, and it is ‘correct’ at the abstract level. The SF is what ‘surfaces’ in speech after the UR has been modified by underging a phonological process. Surface forms are a net result of one or more of phonological processes that occur systematically during the last stage of language production. Phonological processes (e.g. substitution, assimilation, epenthesis) cause the alternation of phonemes based phonological environments in which they are found. Environments can exist within morphemes, syllables of word segments; they can also be phrasal and exist across word boundaries when phonological processes such as vowel harmony are at play.
Example #1: The Role of Phonological Processes
The table below contains examples of vowel reduction (a phonological process) occurring among two verbs in Palauan.
Note that UR is represented phonemically, in broad transcription, and SF is represented phonetically, in narrow transcription. This is a convention used when differentiating between UR and SF. Representing UR in broad transcription and SF in narrow transcription reflects the fact that UR is conceptual and general, and SF is a specific surfacing of UR, and there are potentially several surface forms to one underlying form.
The example of Palauan verbs illustrates the way in which surface forms vary systematically from their UF derivation. The UR of each word appears in the left column. The next three columns contain various surface forms of the word. (They are conjugations of the verb; to avoid confusion, bear in mind that the UR is an abstract concept and is not the same thing as the verbal infinitive.) In the table, certain phonemes have been emboldened in each SF to highlight the change that has occurred from the underlying level to the surface level.
As a phonological process, stress shifts and vowel reduction (V à /ə/) occurs when case affixes (/mə-, -l, -all/) are attached to the word. The change that occurs between levels is systematic and rule driven and can produce any number of surface forms; note that any one lexical entry can have only one underlying form, but will likely have several surface forms.
Example #2: Allophonic Variation as the Surface Form
To further illustrate the difference between these two levels, consider the contraction <can’t>. The underlying form of <can’t> is /kænt/. In some dialects of American English the /k/ is aspirated, the vowel /æ/ is nasalized, the nasal /n/ is deleted and the /t/ is glottalized. The result of these phonological processes is [kʰæ̃ʔ], which is how the underlying /kænt/ has surfaced.
Frequently the underlying form is simply the phonemic form. For example, in many varieties of American English, the phoneme /t/ in a word like <bet> may be glottalized or flapped causing it to surface as [tˀ] or [ɾ]. Note that [tˀ] or [ɾ], the surface forms, are allophones and do not contribute to alternations in the meaning of <bet>. They are a consequence of different phonological processes occurring as a result of /t/, the underlying form, being placed in different phonological environments.
Determining the UR with Phonological Analysis
Though the UF is rarely immediately clear when looking at surface forms, it can be determined by using phonological analysis. Remember, the goal of phonological analysis is to identify distinct phonemes and account for allophonic variation between them. Moreover, phonological analysis can determine whether not distinct phonemes have allophonic variation.
If they are found to have allophonic variation, then allophones (of a single phoneme) must be members of one underlying representation. In phonological analysis, this is indicated when they are found in complementary distribution. If they are not found to have allophonic variation, this means that there is a one-to-one correlation between their underlying and surface form. In terms of writing rules for phonological analysis, UR is the input, and SF is the output.
Further Reading: Designated and Non-designated Levels
To add to your understanding of levels of representation: Among all theories of phonology (not just generative), levels of representation can be classified as either designated or non-designated. Within the generative theory of phonology (known as ‘SPE’) there are two designated levels and unlimited non-designated levels. UF and SF are the designated levels, and the non-designated levels are all forms that exist intermittently in the stages of phonological processes.
B. Okum, 2015