This dissertation develops a comprehensive response to the question of how agreement is computed in Hindi-Urdu – a language with a mixed agreement system where the verb can agree with a subject or an object depending on the structural context. This dissertation covers new empirical and theoretical ground in two domains. First, I identify three kinds of atypical agreement patterns which are not accounted for under traditional approaches Hindi-Urdu agreement -- verb agreement with the nominal component of Noun-Verb complex predicates, long distance agreement of embedding Adjective-Verb predicates with embedded infinitive clause objects, and copular agreement in identity copula structures. Second, I make a novel contribution with respect to the question of how agreement is computed in real-time by investigating the conditions under which agreement attraction errors arise in Hindi-Urdu. I argue for a uniform analysis involving downwards probing for Hindi-Urdu’s mixed agreement system. The downwards probing is proposed to be sensitive to case-marked nominals, as nominals with differential object marking are argued to be visible to the agreement system giving rise to defective intervention and preventing agreement with the nominal component of Noun-Verb predicates. I also propose number-gender and person should probe separately, with the person probe being an articulated probe to capture person hierarchy effects in identity copulas. Furthermore, I propose that upwards agreement is required in the context of adjectival agreement in order to capture the agreement asymmetries between two-subclasses of adjectives which are argued to either be unergative or unaccusative. I demonstrate that speakers of Hindi-Urdu are susceptible to agreement errors in the presence of a mismatching distractor nominal in both subject agreement and object agreement and thereby provide support for a uniform treatment of these two agreement outcomes. At the same time, the mere presence of a distractor is not sufficient to give rise to agreement errors in Hindi-Urdu. The pattern of errors as well as the response time evidence points to speakers utilizing a representational approach to agreement processing wherein the features of the distractor intrude into agreement computation, but only when the distractor is independently identified as an agreement controller.