How to enforce sentence_count in abstractive summary.

Question

How to enforce sentence_count in abstractive summary.

RoryCorneliusSmith-5098 5

Simple text abstractive summary question from my experiences using the Python SDK and the online language studio tryout.

The latter seems to provide much more concise abstractive text summaries as opposed to my efforts with the Python SDK and I am wondering why this is? I know the sentence_count parameter is an unenforceable parameter but I am wondering why there is such a discrepancy and how to overcome this?

AbstractiveSummaryAction(sentence_count=3, model_version="latest")

Using the above as an action input to begin_analyze_actions but I have also experimented with begin_abstract_summary equally with a sentence count set to 3. In both cases with the SDK I get an output longer than my input closer to 7/8 in sentence count.

Manas Mohanty 11,690 Reputation points Microsoft External Staff Moderator

2025-10-13T18:11:29.62+00:00

Hi Rory Cornelius Smith

Please share relevant SDK code and documentation for replicating the issue.

Thank you.

RoryCorneliusSmith-5098 5

Thanks for the response @Manas Mohanty , this is my code using the SDK for abstractive summary. I am using the latest model version from Sweden Central.

        input_text = ["This sprint includes a variety of tasks. The team is working on initiatives such as " + ", ".join(descriptions[:-1]) + \
         ", and " + descriptions[-1] + "."]

        documents = [{"id": "1", "text": input_text}]

        actions = [
            AbstractiveSummaryAction(sentence_count=3, model_version="latest")
        ]

        poller = text_analytics_client.begin_analyze_actions(
            documents=documents,
            actions=actions
        )

        results = poller.result()

Manas Mohanty 11,690 Microsoft External Staff Moderator

Hi RoryCorneliusSmith-5098

I am able to replicate the issue with Python SDK and testing with Rest API to isolate the issue.

import os
from azure.core.credentials import AzureKeyCredential

key = "<languageky>"
endpoint = "https://vmanlanguage.cognitiveservices.azure.com/"

import os
from azure.ai.textanalytics import TextAnalyticsClient, AbstractiveSummaryAction
from azure.core.credentials import AzureKeyCredential


# Authenticate the client
def authenticate_client():
    ta_credential = AzureKeyCredential(key)
    text_analytics_client = TextAnalyticsClient(
        endpoint=endpoint,
        credential=ta_credential
    )
    return text_analytics_client

client = authenticate_client()

# Example method for abstractive summarization
def sample_abstractive_summarization(client):
    document = [
        "The abstractive summarization feature uses advanced natural language generation techniques "
    ]

    # Begin analysis with AbstractiveSummaryAction
    poller = client.begin_analyze_actions(
        document,
        actions=[
            AbstractiveSummaryAction(sentence_count=5)  # You can adjust sentence_count as needed
        ],
    )

    # Retrieve and print results
    document_results = poller.result()
    for result in document_results:
        abstractive_summary_result = result[0]  # first document, first result
        if abstractive_summary_result.is_error:
            print(f"...Is an error with code '{abstractive_summary_result.code}' and message '{abstractive_summary_result.message}'")
        else:
            for summary in abstractive_summary_result.summaries:
                print(f"Abstractive Summary: {summary.text}")

sample_abstractive_summarization(client)

Output ( Output 5 lines)


Abstractive Summary: The source document highlights an abstractive summarization feature that employs sophisticated natural language generation methods. This technology is designed to encapsulate the core ideas from a source document into a concise summary. While specific details of the techniques used are not provided, it is implied that the system is capable of understanding and distilling complex information effectively. The focus on advanced language generation suggests a significant level of artificial intelligence or algorithmic sophistication in creating summaries that capture the essence of the original material. Overall, the document points to an innovative approach in summarization technology, though the depth of its capabilities remains to be inferred from the given text. The summary reflects the central theme of using advanced techniques for abstractive summarization without delving into specific questions or answers that might be present in the original context.

Throught Abstractive adds up clarity on given topic but sentence count param is not making any difference here.

Will sync with Product group internally and update here.

Thank you for your inputs.

1 answer

Your answer

Manas Mohanty 11,690 Reputation points Microsoft External Staff Moderator

2025-10-13T18:11:29.62+00:00

Hi Rory Cornelius Smith

Please share relevant SDK code and documentation for replicating the issue.

Thank you.
RoryCorneliusSmith-5098 5 Reputation points

2025-10-14T13:07:52.4833333+00:00

Thanks for the response @Manas Mohanty , this is my code using the SDK for abstractive summary. I am using the latest model version from Sweden Central.

input_text = ["This sprint includes a variety of tasks. The team is working on initiatives such as " + ", ".join(descriptions[:-1]) + \ ", and " + descriptions[-1] + "."] documents = [{"id": "1", "text": input_text}] actions = [ AbstractiveSummaryAction(sentence_count=3, model_version="latest") ] poller = text_analytics_client.begin_analyze_actions( documents=documents, actions=actions ) results = poller.result()

Answer 1

Hi RoryCorneliusSmith-5098

Answer lies in the keyword itself. AbstractiveSummaryAction

Abstract summarization will add more context to input statements; Approximate count of Summary will be around the Sentence count (might exceed to give proper summary.

It will try to provide output summary and not intended to truncate summary to exact sentence count.

Please use Extractive Summary instead which truncates and give expected result with max_sentence_count.

   import os
   from azure.core.credentials import AzureKeyCredential
   from azure.ai.textanalytics import TextAnalyticsClient

   

   text_analytics_client = TextAnalyticsClient(
       endpoint=endpoint,
       credential=AzureKeyCredential(key),
   )

   document = [
       "At Microsoft, we have been on a quest to advance AI beyond existing techniques, by taking a more holistic, "
       "human-centric approach to learning and understanding. As Chief Technology Officer of Azure AI Cognitive "
       "Services, I have been working with a team of amazing scientists and engineers to turn this quest into a "
       "reality. In my role, I enjoy a unique perspective in viewing the relationship among three attributes of "
       "human cognition: monolingual text (X), audio or visual sensory signals, (Y) and multilingual (Z). At the "
       "intersection of all three, there's magic-what we call XYZ-code as illustrated in Figure 1-a joint "
       "representation to create more powerful AI that can speak, hear, see, and understand humans better. "
       "We believe XYZ-code will enable us to fulfill our long-term vision: cross-domain transfer learning, "
       "spanning modalities and languages. The goal is to have pretrained models that can jointly learn "
       "representations to support a broad range of downstream AI tasks, much in the way humans do today. "
       "Over the past five years, we have achieved human performance on benchmarks in conversational speech "
       "recognition, machine translation, conversational question answering, machine reading comprehension, "
       "and image captioning. These five breakthroughs provided us with strong signals toward our more ambitious "
       "aspiration to produce a leap in AI capabilities, achieving multisensory and multilingual learning that "
       "is closer in line with how humans learn and understand. I believe the joint XYZ-code is a foundational "
       "component of this aspiration, if grounded with external knowledge sources in the downstream AI tasks."
   ]

   poller = text_analytics_client.begin_extract_summary(document, max_sentence_count= 2)
   extract_summary_results = poller.result()
   for result in extract_summary_results:
       if result.kind == "ExtractiveSummarization":
           print("Summary extracted: \n{}".format(
               " ".join([sentence.text for sentence in result.sentences]))
           )
       elif result.is_error is True:
           print("...Is an error with code '{}' and message '{}'".format(
               result.error.code, result.error.message
           ))

Output

 At the intersection of all three, there's magic-what we call XYZ-code as illustrated in Figure 1-a joint representation to create more powerful AI that can speak, hear, see, and understand humans better. The goal is to have pretrained models that can jointly learn representations to support a broad range of downstream AI tasks, much in the way humans do today.

Hope It clarifies the intended behaviour of Abstractive and Extractive summary api.

Reference used - https://free.blessedness.top/en-us/python/api/azure-ai-textanalytics/azure.ai.textanalytics.textanalyticsclient?view=azure-python#azure-ai-textanalytics-textanalyticsclient-begin-extract-summary

Thank you.

Share via

How to enforce sentence_count in abstractive summary.

1 answer

Your answer