Raising Vowel Harmony-RVH is a harmonic process that affects the height of vowels. In presence of high vowels, mid-vowels of a word raise. RVH displays a peculiar property: raising may be scalar or complete. Scalar RVH raise vowels by a degree of height, while complete RVH turn any vowel into a high vowel regardless of its lexical height. RVH may also be asymmetric: not all expected triggers actually trigger the harmony, or not all expected targets are affected (a high vowel may not trigger RVH though being high, or a mid-vowel may not raise while all other mid-vowels do). In the case of parasitism, some targets undergo VH only if they share a property with the trigger. The cross-linguistic properties of RVH are explored through a typological survey. Original fieldwork shows that three similar Eastern Lombard dialects display different RVHs. In Bresciano all mid-vowels are affected by raising ; in Bergamasco RVH targets only rounded mid-vowels. In Camuno RVH is parasitic : rounded vowels undergo raising whenever they are followed by high vowels, while unrounded vowels do so only if the triggering vowel is also unrounded. It is shown that theories able to formalize asymmetries encounter problems with scalar raising patterns, while theories that address the scalar nature of raising are unable to handle asymmetric RVH. It is argued that asymmetries and parasitic RVH are a form of crazy rules (Bach & Harms 1972, Chabot 2021): the class of mid vowels splits into a subset that undergoes raising and another that does not. This requires the existence of phonologically active classes (Mielke 2008), rather than of phonetically natural classes. It is argued that RVHs can only be described when phonological patterning alone defines which segments share a given prime. Given this background, it appears that the origin of the problems of existing analyses of RVH is the shared assumption that melodic primes as well as their phonetic correlates are universal and innate. Most theories assume the existence of a fixed set of primes that is universal and innate, where the phonetic correlate of every prime is given at birth and is the same in all languages. The alternative (Radical Substance Free Phonology) is a theory based on emergent primes, which argues that primes and their phonetic correlates are learned. There are no primes at the initial state : children are born with the knowledge that there are primes and that they will need to construct them based on environmental information. Primes are language-specific and have an arbitrary correlation with phonetics, which depends on contrast and behavior of segments in phonological processing (Mielke 2008, Dresher 2014). Different models of prime emergence are compared (Dresher 2014, Sandstedt 2018, Odden 2022). These proposals consider contrast and phonological processing as factors in prime emergence. For Dresher and Sandstedt contrast and processing are equally important, while for Odden processing has a logical precedence over contrast. In an environment where primes such as α β are used, different theories of computation may produce different prime specifications. This thesis argues for an approach to computation where only addition and subtraction of primes are allowed. This corresponds to the state of computation in Autosegmentalism, where primes may be either linked or delinked. On the representational side, this thesis endorses unary primes. It is shown that Sandstedt's model fails to account for scalar raising patterns and Odden's approach can build the required sets of representations for both parasitic and scalar RVH. It can also account for morphologically conditioned processes based on the fact that emergent primes entail phonetic arbitrariness, i.e. an interface between phonetics and phonology where mappings are arbitrary. Odden's approach is formally simpler than the others: processing alone guides the learner to prime specifications without recurring to other additional assumptions.