It is well known that natural language usually can provide much more precise and convenient spatial information in the context of social geographical information service. This paper researches the method of the GIS path reconstruction based on natural language （Chinese） containing geo spatial information

which is called as natural language representation of path （NLRP）. Through analyzing Chinese sentences that contain path representation

a syntax model of NLRP for Chinese is built

which mainly contains a set of actions with spatial constraint and action objects. As a rule

the action objects usually are geospatial features. Considering the contribution to the forming of a concrete path

the actions in NLRP can be classified into three types: durative actions

transitional actions and parallel actions. Durative actions can last for a period of time

and each of them forms one segment of the whole path; transitional actions change the state of the activities

such as “turn left”

and usually they start flags or end flags of a durative action; when a parallel action happens

the state doesn’t change

and it occurs inside another durative action both spatially and temporally

such as “pass by”. Generally speaking

durative actions are corresponding to linear features and time segments

while a non durative action

can be mapped to a specific event point in spatial and temporal space. In the model

the actions with spatial semantics can be defined precisely using 9 I matrix.According to the model above

the NLRP syntax for constrained Chinese language is defined using BNF. Then the path reconstruction algorithm （PRA） and two interrelated issues

i.e. dictionary design and software architecture design

are discussed. In PRA

using intermediate result of the syntax processing

which is a sequence of actions

the action series are analyzed and the spatial relationships between objective path and related features are examined based on the geographical reference map. At last

the concrete path can be figured out.Furthermore

a NLRP expression contains some uncertainties inevitably

which may come from the spatial cognition process and imprecise expression. The uncertainties of NLRP bring ambiguities to PRA

and increase the algorithm complexity. They include uncertainties of direction

uncertainties of distance and uncertainties of topologic relationship. Meanwhile

solution to them is also discussed

the essential of which is to distinguish major flag and minor flag of a durative action

and the former determines the shape of path

while the latter may be used to check errors of an NLRP description.Finally

a case study of PRA is described to validate it.