Primordial non-Gaussianity of local type is known to produce a scale-dependent contribution to the galaxy bias . Several classes of multi-field inflationary models predict non-Gaussian bias which is stochastic , in the sense that dark matter and halos don ’ t trace each other perfectly on large scales . In this work , we forecast the ability of next-generation Large Scale Structure surveys to constrain common types of primordial non-Gaussianity like f _ { NL } , g _ { NL } and \tau _ { NL } using halo bias , including stochastic contributions . We provide fitting functions for statistical errors on these parameters which can be used for rapid forecasting or survey optimization . A next-generation survey with volume V = 25 h ^ { -3 } Gpc ^ { 3 } , median redshift z = 0.7 and mean bias b _ { g } = 2.5 , can achieve \sigma ( f _ { NL } ) = 6 , \sigma ( g _ { NL } ) = 10 ^ { 5 } and \sigma ( \tau _ { NL } ) = 10 ^ { 3 } if no mass information is available . If halo masses are available , we show that optimally weighting the halo field in order to reduce sample variance can achieve \sigma ( f _ { NL } ) = 1.5 , \sigma ( g _ { NL } ) = 10 ^ { 4 } and \sigma ( \tau _ { NL } ) = 100 if halos with mass down to M _ { min } = 10 ^ { 11 } h ^ { -1 } M _ { \odot } are resolved , outperforming Planck by a factor of 4 on f _ { NL } and nearly an order of magnitude on g _ { NL } and \tau _ { NL } . Finally , we study the effect of photometric redshift errors and discuss degeneracies between different non-Gaussian parameters , as well as the impact of marginalizing Gaussian bias and shot noise .