Aphasia is a language impairment due to brain damage. In German-speaking countries, the \textit{Aachener Aphasie Test} (AAT) is the clinical gold standard for diagnosis and therapy planning. Although the AAT evaluates multiple language domains, including spontaneous speech, it is time- and resource-intensive. Advances in Machine Learning (ML) have shown promising results for aphasia assessment using speech and language analysis. This study investigates (1) the application of Automatic Speech Recognition (ASR) to spontaneous speech produced by people with aphasia, (2) whether automatically extracted speech and linguistic information can support the modelling of spontaneous language symptoms associated with different aphasia syndromes, in alignment with the ratings of the AAT, and (3) how model predictions relate to speech and language features.We collected~429~spontaneous speech recordings from~290~German-speaking people with aphasia while performing the AAT. We fine-tuned a transformer-based ASR model to transcribe the recordings. Multi-task regression models were trained to estimate the spontaneous language subscale scores: communicative behaviour, articulation and prosody, automatised language, semantic structure, phonological structure, and syntactic structure. While ASR results on spontaneous speech from people with aphasia remains challenging, ML models demonstrate encouraging results in modelling spontaneous language domains related to aphasia syndromes. Exploratory analyses further show associations between model predictions and speech-language features that are broadly consistent with established descriptions of aphasic language impairment. By examining relationships between speech-language characteristics and measures derived from the AAT, this work contributes toward a computational approach for studying spontaneous speech associated with aphasia in German language.