The problem of automatic translation of texts into sign languages (Sign Language Translation, SLT) is particularly relevant for agglutinative languages, including Kazakh. The complex morphological structure of the Kazakh language complicates the direct conversion of text into a sequence of signs and requires multi-stage linguistic processing. This study proposes a solution in the form of a text-to-gloss conversion system as the first stage in generating Kazakh Sign Language (KSL). The goal of the developed system is to extract the basic lexical units (glosses) from the input Kazakh text for subsequent synthesis of a sign-language message. To address this task, a hybrid approach combining deterministic and statistical methods was developed. Morphological analysis of the input text is performed using finite-state automata that formally describe the grammar of the Kazakh language, enabling the generation of all possible analyses of each word form, including identification of its lemma and grammatical features. A hidden Markov model subsequently evaluates these alternatives statistically and selects the correct sequence of lemmas, taking the sentence context into account. As a result, a software pipeline was created, implemented as both a console application and a web interface, which converts arbitrary Kazakh text into an equivalent sequence of KSL glosses. Experimental evaluation demonstrated that the system provides high accuracy in lemmatizing input words, confirming the effectiveness of the proposed approach for processing agglutinative morphology in SLT tasks. Thus, the proposed text-to-gloss conversion pipeline can serve as a foundation for developing a complete automatic translation system from Kazakh to sign language.