This report describes a computational system with which phonologists may describe a natural language in terms of autosegmental phonology, currently the most advanced theory pertaining to the sound systems of human languages. This system allows linguists to easily test autosegmental hypotheses against a large corpus of data. The system was designed primarily with tonal systems in mind, but also provides support for tree or feature matrix representation of phonemes (as in The Sound Pattern of English), as well as syllable structures and other aspects of phonological theory. Underspecification is allowed, and trees may be specified before, during, and after rule application. The association convention is automatically applied, and other principles such as the conjunctivity condition are supported. The method of representation was designed such that rules are designated in as close a fashion as possible to the existing conventions of autosegmental theory while adhering to a textual constraint for maximum portability.

Phonological processes can be characterized as functions from input strings to output strings, and treating them as mathematical objects like this reveals properties that hold regardless of how we implement them (i.e., with rules, constraints, or other tools). For example, Chandlee (2014) found that a vast majority of phonological processes can be modelled as Strictly Local (SL) functions, which are sensitive to a window of finite size. Long-distance processes like vowel and consonant harmony are exceptions to this generalization, although a key observation is that they look local once irrelevant information is ignored. This thesis shows how such selective attention can be modelled by augmenting SL functions with autosegmental tiers (e.g., Goldsmith, 1976). A single tier is sufficient to capture individual long-distance processes, and having multiple tiers available allows us to model multiple long-distance processes simultaneously as well as interactions between local and non-local patterns. Furthermore, probabilistic variants of these tier-based functions allow for a cognitively plausible model of what Zymet (2015) calls distance-based decay. Unrestricted use of multiple tiers is, however, quite powerful and so I additionally argue that tiersets should be defined from the perspective of individual input elements (i.e., potential process targets). Each input element designates a superset-subset hierarchy of tiers and pays attention to them alone; the tiers specified by another input element are either redundant or irrelevant. Restricting tiersets in this manner has beneficial consequences for learnability as it imparts a structure onto the learner's hypothesis space that can be exploited to great effect. Furthermore, tier-based functions meeting this restriction fail to generate a number of pathological behaviours that can be characterized as subsequential functions, a type of function that has previously been offered as a model of non-local phonological processes (Heinz and Lai, 2013; Luo, 2017; Payne, 2017). In light of their empirical coverage, their comparative lack of pathological predictions, and their efficient learnability, I conclude that tier-based functions act as a more accurate characterization of long-distance phonology.

Ever since Bengt Sigurd laid out the first comprehensive description of Swedish phonotactics in 1965, it has been the main point of reference within the field. This thesis attempts a new approach, by presenting a computational and statistical model of Swedish phonotactics, which can be built by any corpus of IPA phonetic script. The model is a weighted trie, represented as a finite state automaton, where states are phonemes linked by transitions in valid phoneme sequences, which adds the benefits of being probabilistic and expressible by regular languages. It was implemented using the Nordisk Språkteknologi (NST) pronunciation lexicon and was used to test against a couple of rulesets defined in Sigurd relating to initial two consonant clusters of phonemes and phoneme classes. The results largely agree with Sigurd's rules and illustrated the benefits of the model, in that it effectively can be used to pattern match against phonotactic information using regular expression-like syntax. / Ända sedan Bengt Sigurd lade fram den första övergripande beskrivningen av svensk fonotax 1965, så har den varit den främsta referenspunkten inom fältet. Detta examensarbete försöker sig på en ny infallsvinkel genom att presentera en beräkningsbar och statistisk modell av svensk fonotax som kan byggas med en korpus av fonetisk skrift i IPA. Modellen är en viktad trie, representerad som en ändlig automat, vilket har fördelarna av att vara probabilistisk och kunna beskrivas av reguljära språk. Den implementerades med hjälp av uttalslexikonet från Nordisk Språkteknologi (NST) och användes för att testa ett par regelgrupper av initiala två-konsonant kluster av fonem och fonemklasser definierad av Sigurd. Resultaten stämmer till större del överens med Sigurds regler och visar på fördelarna hos modellen, i att den effektivt kan användas för att matcha mönster av fonotaktisk information med hjälp av en liknande syntax för reguljära uttryck.